GPT-5 in Cursor vs Claude Code: Testing If It's Worth The Hype LIVE

Yo yo yo, GPT5 has officially landed. Holy smokes. Today we're going to take a look at this model. It has claims for extremely high intelligence, a lot of coding capability, and all the different things that we're going to take a look at. And I have some sauce for you.

This model has also landed inside of cursor with about like eight different configurations. And we're just going to go ahead and take a look and explore that inside of cursor. And what this model has allowed you to do, ladies and gentlemen, is kind of unfathomable. And I want to let you in on a little bit secret. I've been kind of holding back a little bit. I know everyone's been kind of killing me right now, but I got secret access that I didn't know. I got access to a crazy model uh inside of this thing. And we're going to take a look and see how this thing actually performs because I actually have been using it to refactor my codebase and everything like that. And I'm actually going to cancel my Claude code subscription based off of using this model for the last like I don't know 24 hours. And my reaction is insane because there's some other sauce that I'm trying to confirm with the cursor team that they're going to be releasing soon that is going to be like bro I'm done with cloud code right now. I'm literally done with cloud code.

So there's a lot of hype right now, but it is actually justified and we're going to take a look because this is the first time I've ever felt that a model can actually like understand with you and actually come back at you. There was some refactoring that was done in my codebase that I want to share with you in a conversation where I'm like, yo, did you actually fix it and did you do this type of thing? I was just kind of pressing it really hard and it's like, yeah, bro, like this is an architecture decision that we did and this is why I did it. You know, it's like I was like, "Oh, wow." Cuz I was just trying to really kind of gaslight the model and it knew I was trying to gaslight it, which is kind of scary, but it's actually really amazing that they've been able to do that.

So, um, this is something that I was just kind of showing you, uh, because I actually had it generate a plan. So, one of the techniques that I want to share with you is I chat with GBT5 and I what I basically do is have it generate a plan for me and then once it does that plan in like a lot of detail, the next thing you want to go ahead and let it do is basically let it um like work on that plan and just open it up in a new chat. And so, one of the great things that Cursor has done at the very bottom is they actually tell you how many tokens are being consumed. And I want to keep that token window tight. So, as it's doing research on my code, I want to just make sure that it does the various things. So, in the thing in my previous plan, let me see if I can kind of go back to the history of chats and kind of show you uh what was going on cuz I had to basically refactor my app for O and I had actually confused and used I like vibecoded my O in a funny way where I used two different conventions and during that transition, it basically made an A, B, and C plan. And so whenever you ask the model to do something, you're going to actually see it grab through and do a whole bunch of checking of different things. And it's extremely fast at doing this type of stuff. And so once that plan gets generated, we'll see if I can kind of show you a little bit here. I'm going to get my other camera turned on. I just got too excited for this stream. I was having some technical difficulties earlier.

And so one of the things that we're going to go ahead and do is I just wanted to give you a quick overview on that really quick of it landing in the model. And then another thing I wanted to quickly show you too before we kind of get into some coding stuff. Uh, and I got to catch up with the chat a little bit because I need to give some shout outs to my members and everyone who's been waiting kind of getting on this because I was trying to get this model cracking super fast. So right now we have GPT5 Max. And so this currently shows us here that uh it's OpenAI's latest model. Obviously 272 token context window just by default here. We can have this thing turned on in max mode and it's going to be able to do that and it's it's currently free. And then uh we also have the fast model which is also here as well. So this is 2x the price and this is basically the the highest priority and I don't know if you saw the specs on OpenAI's website but I'm going to pull this up for you as well.

So you can also access the GBT5 API through these different uh configurations here on platform.openai.com. So in the platform here what they're basically showing us is there's batch, there's flex, there's standard and there's priority. I know, right? There's there's all these model configurations, but what you need to know is that the standard thing is probably what you're going to see passed around for most models, but this is the priority thing. And if you can actually see the priority pricing for GPT5 right now, it says that it's $125 for the input and then 10 bucks for the output. So, whenever you do the priority, obviously, it's going to be 250 for the input and then 20 bucks for the output. still extremely amazingly competitive when you're thinking about something for Cloud 4 Opus, which is about 15 per million input and $60 output, $75 per million output. Don't quote me on that last part, but I know it's just like ridiculously high. So, even even if you have priority set up here, this is kind of what you're looking at from a priority processing standpoint on its API itself. And then OpenAI has also released the mini, the nano, and they also have like a chat model as well, which is kind of what we're dealing with here.

So, um, this is the model configuration. And if you go into models, if you don't see it yet in your, uh, thing, if you just type in like GPT5, you'll actually see all of the different models. So, I can turn on high, I could turn on low, I can turn on high, fast. I don't know. Let's just going to turn some of these on. I feel like I got to play around with some of these. I haven't got to like these are kind of new to me. So, we're kind of looking at these newer ones here. Uh, and then GPT5 Nano, GPT5 Mini. Okay, cool.

So, I think um yeah, one of the apps here that we're kind of taking a look at is um let's see. I I want to see if I can I'm using a Bootstrap here. Let me see if I can just kind of bunddev this and I'll show you real quick what kind of what this bootstrap is doing for us. Let me clear this bunddev. Oh, yeah. There's some new sounds, too, by the way. So, in here, all right, so I have this little like template that's pulled up, and this is a an Nex.js SAS starter template that basically connects with Convex. So, we have like the full Convex database, and then uh it's hooked up to like a dashboard. So, when you when you log in, the users will actually come into here and then you can select different things like payment gate and so forth. And if you wanted to subscribe, you would do that subscription. All of this is actually handled through Clerk. So Clerk is doing the like web hooks and all that Stripe payments things. And if you want to actually download this, this is actually on my GitHub. So it's it's free. You can do that. Uh it's from this guy named our team. He's an amazing developer and he actually has this template available. Um, what's nice about it is that it's using Shad CN Tailwind CSS. You can use Tweak CN as well.

So, one of the interesting things here which would be really fun for us to test is how good is GPT5 at taking a template like this and then understanding what the codebase does because it has a database because has has all these things kind of baked into it. It's a very rich app already. Can cursor with GPT5 be able to take this template and really start cooking with it? You know what I mean? Can it like understand all the different hooks and understand my cursor rules for convex in there? Can it like understand that it has shad cen and all these different types of things in there as well? I know like with clawed code I have to be a little bit more explicit with things and so I'm really curious to see what it can do here and I'd love to know like what type of apps you want me to build, right? We're not going to build the next snake game. We're not going to build the next whatever the game is, you know. Um, but I think it'd be really fun to just kind of let it do its some research for us because that that would be really really incredible.

So, one of the apps that I'm thinking right now is I always wanted I like the Tinder UI. If you ever swipe left and swipe right, it's kind of a really amazing experience that these guys have been able to nail down. So, I thought it'd be really fun to maybe build like code reviews. So, you can just take your GitHub issues and throw them into a database and then you can just swipe left or swipe right if you want to work on them or approve or deny them. So right now, I don't know if you know, but when you're using background agents or any type of agentic coding, a lot of these agents will actually do the work and then spit them back down. So if you have them all done and organize in a GitHub issues type of thing, why can't I just review them and just swipe right or swipe left to just approve and merge, right? So that way we can ship fast with AI or then, you know, respond back with voice or something like that. So why don't we just work on that? I think that can be a really fun idea to for us to work while we let the model kind of do its thing and we can kind of preview what that looks like. Um, and then then I'll catch up a little bit to the chat as well cuz uh, everyone's has been super popping. So, I want a big shout out to uh, Alex Finn who just dropped a super chat. He says, "Good live, Bozo." I was I was a little bit late getting started, but definitely I wanted to give you a shout out. If you haven't had a chance to go check out Alex Finn, definitely go ahead and do that. Uh, big shout outs to the members as well. So, u, I'm a little bit behind on the chat, so I'm going to make sure uh, we're going to get to that.

All right, so um, yeah, that's Twec CN. This is that type of thing. This is our dashboard right here. We have this dashboard already that's built into our app. Um, it's a fairly rich app. So, we're just going to go and take a look. And let me see if I can pop cursor down a little bit so y'all can see a little bit what's going on. And then I'm going to minimize this other one for now. And I'm going to minimize this one, too, because we're going to be working a lot with code specifically. All right.

So, let's take a look here. So, in GPT5, we have GPT5. I'm just going to go straight up because if I do high, I don't know. Yeah, I don't I don't think I need So high specifically calls out the high reasoning effort and then we have the medium reasoning effort. I think medium for every day should probably be fine. So, we're just going to go ahead and do that. And then I think for my app as well is that um I'm going to just describe what I want to do in my app and I'll just let the agent kind of do its work. I'm going to remove whatever tab is here and I'm just going to let it cook for its thing. So, let's go ahead and see if we can figure this out here.

All right. I want to make a app that'll allow me to review my GitHub issues. And I can either swipe left or swipe right using a Tinder-l like UI so that I can the whole purpose of this app is actually to review my code. And as the code comes in through the GitHub issues, I'll be able to kind of pull it up just like this way you see a Tinder profile. And um I want you to get really creative with this and really fun with this because I really want to see you know how good you are at creating this type of app for me. It should be very fun, very playful. Doing development is also not really like playful type of thing. So by doing it this way, it's just going to encourage me to get on top of my GitHub issues and especially with these code reviews type of thing. So, I want you to recognize that there's a code review and it needs to be reviewed to be merged in. And I want to use the React swipeable framework so that way you can swipe left or swipe right uh to approve the changes and so forth. So, I want you to use the web and see if you can generate a plan and review my codebase and look at the different conventions I have here so that we don't rewrite new code for this. It was just kind of like a generic thing. So I basically this is just my like word vomit and I think it'd be really interesting to see how the model treats this whole type of thing. So I'm going to look up for web. Is web mentioned in here? And then I'm just going to say at web at web. Okay, cool.

So we're just going to go ahead GPT5 max. We're just going to let it do its thing here. Um I should have probably started off the demo like this, but I just came up with the idea right now live. This is a live stream, ladies and gentlemen. So this is really interesting. It's going to look up these different React components here. And this is actually my my website. It's called just three commas. So the goal of three commas, if you don't haven't watched the Silicon Valley show, is like whenever you've made it to the billiond dollar club, you have three commas in your bank account. And so so this is kind of our goal here is to make a SAS where we actually generate a billion dollars worth of u net worth here. So I'll scan the database of the codebase and understand the conventions.

So, what's really cool about cursor is that you'll actually see how fast it is at looking at different parts of your code. And um I'm going to check in with some folks to see if this other announcement has come through because then we'll have some other cool stuff loading up, but it may not be related to this, but uh we'll see. But yeah, so this is interesting. So now it's actually going through and reading the code. It's using the GPT5 Max model. I I have my different things. I actually don't think I have my convex rules in here. Do I? Let's see. convex rules. Oh no, I'm going to go ahead and grab the rules right now. Oh, actually it might be my cursor file. Yeah, I do have the convex rules here. So that's really cool because that's a super handy type of thing. I might do that later when I actually generate the code. So the first step right now is to just generate the idea. So the idea is here I'm trying to see how good is this model at reading my codebase, right? Taking my idea, trying to chunk it up, and then trying to see if it can come up with a plan. And you know, theoretically, we should be able to do this, right? I have an existing template. I have code. I have the starter kit for a reason because I like to use this stack. I don't want to have it go through all those steps and waste tokens to do that. So, it's always better to have a stack that you have that's working. See if you can make it a template into GitHub and then start with that and then start to make an app from it. Um, that's just like the like probably the easiest and fastest way to get going. You save a lot of tokens, but then on top of that to build on top of it. We'll see how that actually turns up for us using this new model. And so it'd be really cool to see how this type of thing starts going here. So you can kind of see it's a little bit uh thinking a lot more. So it's read a lot of the different files about cards and layout uh context provider. I didn't give it anything any rules files or anything specific to go search through. It's literally just using the intelligence the model cursor has made a lot of updates. I think let's just take a look and see what update this is on. So this is uh 1.4.2. So in this version they have made a lot of improvements to their app. I was primarily using cloud code extensively inside of cursor and now I'm actually preferring the output using this model inside of cursor because of like it's really strong agentic cap capabilities and this is like for me it's performed a lot better than I've been using cloud code uh for for my apps and for solving problems and stuff.

So, while that's going, let me see if I can kind of pull up um some of the chats and stuff and catch up with you guys because it's been a little while and I know a lot of people have been kind of up here. So, Eric, how's it going, man? Let's cook for sure. I'm having a withdrawal. Oh, yeah. Yeah. Uh earlier I had a bladder issue. Yeah, I had to go to the bathroom and stuff. Yo, big shout outs to Alente. Big shout outs to Codeex as well. Uh Alex Finn is over here trying to cook as well. Uh yeah, yeah, yeah. Let's see. Cooking, cooking. All right. All right. No, big shout outs to no code Dan as well. Um, get ready for let's go, fam. Yeah, y'all are real ones. Oh, man. Colin, thank you so much for becoming a member. I really appreciate that. If you guys haven't had a chance to, um, if you want to get in on the Discord, this is probably the best time to do it right now before I double my prices. U, if once you sign up as a member, there's going to be a members only post. Just follow the instructions there to, you know, get me, um, some information and then I'll get you right into the Discord. So, y'all y'all log in with me. Odell, how's it going? Welcome back. I thank you so much for becoming a member again. I really appreciated that. Uh, it's just really awesome to see everyone here come and support. Okay. Yeah, it's fast, y'all. I just sent my first GPT5 Max prompt. Ooh, get next. Get next. What What are you building, boy? I want to see what you're cooking with. This thing is cooking right now. It's It's cooking hard.

So, you notice how long it's thinking? By the way, it just didn't say, "Here's your app. I'm done." This thing is still like churning like we are putting it through the It's interesting. It's like now going to add the React swipeable too. Okay, let's go. So, it it found it. I don't even know if it actually read through the documentation. It It must It must have. It is. So, it says deciding on a library and then it says PMPM lock file. Oh, I think it sees a lock file. Cards are out. Okay. Okay, so it made a little noise at me saying it's already installed and then I think and now it's thinking here. Planning the next moves. So this is really cool. So we're seeing the next moves being planned out. So planning the GitHub issues integration. So I need to implement a plan and possibly add a new page under the dashboard. GitHub issues. I'll reference the web search results and include citations. Uh and then damn setting up the GitHub issues fetching. Let's go. Creating a a swipe deck component. Bro, this is crazy. Damn. Okay, let's go. App sidebar. Okay, so it's adding a new sidebar for me. Dashboard isue review. So, this is in my actual dashboard of the app. Let's see. Can I see here? Uh, let me refresh the site. Oh, I don't even actually have the site up right now. Blendev probably killed it. Let's go. This is so crazy. Let's see. Yeah, issue review now is is here. Oh, yeah, that's that's right. I I did this whole type of thing a while back. This is crazy. What? Okay. Okay. Site header. Let's go. It's actually writing the page for us right now. And so, I think what I wanted it to do is there's a new thing where you can interrupt it. And I'm going to tell it some interruptions. is like I don't want to necessarily have it build the whole thing right now. Um I want you to get to finish this file that you're working on, but the next thing I want you to do is to actually write out an extensive detailed plan in a markdown file and put that in a documentation folder so that we can review it before we get into fully writing out the code. I just want to make sure that we have everything we need in that plan file. And so I'm basically just telling it that type of information. And I'm going to go ahead and hit enter and sending it. So cursor allows you to cue the messages now. So now that we have the message cued, we're just going to let it do its thing. And so it's going to figure out itself how to place this message in our type of thing. You know, if we really want to forcefully send it, we can hit stop and then force it. But you know, now it's actually trying to figure out and plan its next moves. So it's basically I told it to just get to a stopping point. It's going to read about that. It's going to be like, hm, let me look through the chat again. Let me figure out what's going on. And by the way, this is just like one conversation. I'm really curious to see like how long the like context window is going to be for. You know what I mean? Like like h how much of the context window are we going to eat up? That's primarily the reason why I'm stopping it right now. I forgot to tell it that I just wanted to generate this. I think I told it I wanted to generate this plan, but I didn't really tell it that where to generate the plan. I just started seeing it going off to write the files. And this is kind of why I want to manage that context window.

So while that's doing that thing, now it's actually going to figure out to write the uh schema changes and planning the animation integration and creating the documentation plan. So this is super helpful because it's going to say GitHub swipe review plan MD. And in this plan, basically what we're going to go ahead and do now, it's going to Oh, even considering a mermaid diagram. This is this is like, oh my god, this is so fun. This is the type of coding that I've always wanted. It's like a pair programmer that's like really good to be with you side by side, but also like generate plans like this. And you know, I think I think right now the model is like super anxious if you you know, it's like it wants to go ahead and perform for you. But I'm really curious to see what this plan looks like and then see if we can take it and iterate from there because I can open up a new chat in that new context window and then I can start adding more like detailed features to this and say um I really to be honest when I want to make an app I want to make the basically like the minimum lovable feature. The only thing that is going to get people to talk about this is if they see a swipe left or swipe right for a code review, right? So if you see that, like you instantly are going to share it with your friends. You be like, "Yeah, bro. This guy solved it." And that's the only feature I need. I don't need anything else for it to work. I just need to work reliably. I need to go grab the GitHub issues. I need to swipe left, swipe right, and we good, right? Then we got like a billion dollar SAS. No, no, no.

All right, let's see what's happening. Uh cuz this can actually be my entry into the Cerebrro Valley um hackathon cuz they're doing a hackathon right now. And they're just like, "What are you working on?" And I was like, I didn't put anything yet cuz I had no clue what I was going to work on. But I guess I'm going to be working on this. Let's go. Oh my god. Okay. Okay. Okay. Let's go ahead and review this plan. Let's cook, my friends. All right. So, this is up. I'm going to kill this session, the terminal. Okay. All right. All right. All right. So, let's see what's going on. There's this little preview up here and read through this plan and see what's going on. So, there's GitHub swipable review or product technical plan. Purpose is to make a code issue review playful and fast. Yes. using a Tinder UI UI. Uh, swipe right is to approve the PR for issues marked as acknowledgeable in our local log and then swipe left to request changes. Oh, I like that. And so MVP success user can fetch their open issues, can swipe through them for PRs, automatically create a review in GitHub. So by seeing like this, I could see what the model was thinking before it's writing the code. That way I can update the features. Uh if it's I mean first of all we've seen a couple of things here. One is that we gave it this really long spew of an idea. We put that into cursor. We put that into GPT5. Set the max setting. And by the way that max setting just by default was just set for medium reasoning. I didn't have to do high or longer reasoning. We already got quite a bit out of that. Two after that request went in, it started to literally dig through the entire codebase and figure out what is this codebase all about? I'm using a template. I have clerk, I have convex, I have all these different providers like shaden, nex next.js, all of it's installed, ready to go and configured and it's just going through and read all those files and then it in between those types of things, it does thinking and it tells me like what it's thinking about after it's reading all these chunks of files. So look at all that thinking that it's doing. So you don't need a lot of thinking. In essence, you may even save yourself if you don't have the high reasoning because now you have just enough to get by. And I think that's kind of what is going to be really interesting here to play around with these configurations. Um, after that, by the way, then we got into it trying to write the code for me. I'm like, yo, yo, yo, just chill, chill, chill, chill right now. We can't be going in that fast. Like, you know, pull out. Make sure you pull out, bro. And then the next step in the middle of it, while it was going, you can actually feed it a conversation. So, we told it, let's go ahead and get some documentation in before you go a little too deep, man. I don't need you to, you know, just just chill. chill. And then after that, that's what we're reviewing right now. So, that's just a quick recap for those who are just popping into the stream right now where we're at. We're basically building the Tinder UI for GitHub issues. I don't want to say what I was going to say in my mind. It was a little bit of a 30 thought, but yeah. Okay, whatever. We're doing the Tinder UI for GitHub issues. And this is kind of our plan now that we've generated. And just doing a quick review of what this is looking like. And this is all generated using cursor and GPT5. So we're just trying to test the intelligence of the model. Is it really good at looking through my codebase? I know a lot of other u model providers are trying to like do this type of stuff now with their IDE and like is this enough for you to actually cancel your Claude Code Max plan? We're going to find out. I mean, like, I'm gonna literally copy and paste this prompt and then like see if I can clone this repo into a different place and then try to see if I could have it do the same thing and see what type of results we'll get because I'm really curious to see what is going to happen, you know, between the two and and who would actually win, right? Like in this type of uh data field here. So, um I just wanted to have this plan to see if we can get to the stopping point to compare and then maybe I'll just pop this into cloud code and see what it does too. So uh the scope of the MVP uh fetch the combined feed of the open issues and PRs involving the current user. Okay. And then so okay so this is basically it saying about the user experience and then the scope and the architecture overview and then doing this type of current implementation. So real quick what I want to go ahead and do right now while before we continue reviewing the rest of this code. I mean look at it gave us the schema and everything. Sheesh. Oh my goodness. Wow. It It's It's cooking for sure. And so what I'm going to do is I'm going to show you where this repo lives so that way you can copy this as well. So github.com rayfernando137 in here. There's one called um is it? Yeah. Yeah. Elite convex starter. So this is from my boy Artim. I'm going to try to have him on the live stream to like Friday or Saturday, I think. But he made this cool starter kit. And this starter kit is what we're using right now that has Clerk, has Convex and in like 10 minutes you can get an entire SAS setup up and running charging people and everything with payment gated like super fast. So uh I have a video on this as well like on my live stream and I have like an actual recorded video that's going to be dropping in the coming weeks. So make sure you stay tuned for that. But this is a template that I made and I put it on my GitHub. I'm going to put this in the chat so y'all can get cooking with this one. Could you share the GitHub link? I got you, fam. Jack Martin, I got you. Uh, I just dropped it right now in the chat. So, you you able to see this and you'll see this link here. I actually have a shortened URL. I think it's like dub.sh. Um, and it is I think it's called speedrun. No. Um, starter r f.me/starter. So, that's the one that you want to check out there. So, check that out. And this will take you to the starter kit. And um I'm going to clone this kit and then try to do uh cloud code Tinder or something like that. So let's see terminal. Let me see if I can pop open any terminal without having my windows change all over the place. New window, please. Okay, cool. So this is what we're going to do. cd workspace. Make this a little bit bigger for y'all. All right, so I'm going to do get clone this and I'm going to call it uh CC Tinder or no CC. Yeah, Tinder. Why not? cdcc tinder. Okay, cool. So, now that we have this, I'm going to do uh bun install and just install all the packages and then then we're going to open up cloud code and literally just copy and paste the prompt back back and forth. Right. So, okay, now that we're done cla yes, proceed. All right, cool. Now, now we're cooking. So, we're in cloud code. this this actual repo already has a claude uh MD file. So that it's probably why it's not going to be complaining. And I've already initialized that for the starter kit. And so let's go ahead and see if we can grab that in our chat. Just literally the same prompt and just kind of see what it does. I'm I'm really curious to see what it does here. So let's go back to terminal and then here we go. I'm just going to paste this in here and I'm going to put it into plan mode. Shift tab. Shift tab. Plan mode. So that way like I know cursor doesn't have an explicit plan mode but that way it won't just start writing files here and let's just see what it does with the plan. I didn't give it any ultra think. I didn't give it anything. I just said let's just go straight up and then see what happens here. And then if it doesn't do anything interesting I'm just going to create a new window and I'm going to try to see what the difference is maybe between Ultraink and then like think harder and so forth. So um this would be really interesting. So now it's actually searching the web. It's going to do its thing. So, we have a cloud code version running in a whole different repo that I cloned from the template, the exact same template, right? Uh, and now it's going to just try to see if it can actually leverage and do the frameworks. One of the things that we notice right now is like how it's it doesn't really reveal its things, but in cursor it was showing exactly what files it was reading, all the different research it was doing. And I'm not sure if I should have just said we should do this with parallel agents or not, and if it would have gotten us a different result because that could even have an effect on the context window. I feel like I could just pop open a new window and see what it does. I mean, why don't we just do that, right? Let's just do that because I can. I have the max plan, by the way. I still have the max plan and we'll see if I need to cancel it. We're gonna do this. Let's just do it, my friends. All right, let's go all out. So, paste it. And then I'm going to just say, please ultra think and use parallel agents for your research. Okay. And then shift tab. Shift tab. Okay. Now, we're in plan mode again. All right. So, I'm doing two different things, right? I'm literally running the same prompts in cloud code twice. One is just the regular non-thinking and then the other one is with ultrathink and then on top of that we're also saying you know let's just go see if you can use parallel agents. I'm really curious to see if you can kick off more than one agent for this type of research task. And um ideally this is kind of what I do. I don't know about you, but whenever I see a single input or single text box, that's just my opportunity to just dump as much information in right now or maybe like a year ago, that probably wasn't the case because you used to just have to break down your ideas into smaller chunks and then spit that in. But I have ADHD or something. I don't know. I'm not undiagnosed, but I have a lot in my mind. I always want to do lots of things. I just want to dump everything out as much as possible and then just go ahead and capture that later. And that's I think where language models are going. The bigger the models get, the more they're able to use the general knowledge and we don't have to write as much software. We don't have to have as much tooling to figure this out. It would just try to understand everything from you. So that's just kind of where things are going, right? So 2 years from now, we could have the next GPT6. So that will be a 100x leap from GPT5, which is crazy. So GPT5 theoretically from a parameters and like as far as how big the model is and all the data and everything that goes into it is a 100x leap from GPT4 and GPT4 is 100x leap from GPT3. So that's the scale that like things keep getting bigger. And what that means and that this is why I'm kind of stopping right now for us to like think about it is that you have to learn how to prompt these models. Everyone's going to come out today with all their existing prompts and their existing tooling and that could actually cause some pollution. So you may not see the results that I see if you're using a lot of different MCP tools or you're doing a lot of different stuff. So it's important to open different windows and try different prompts kind of like the way that I'm doing right now. Especially since it sounds like some of the models could be free or available inside of cursor. Try different prompts. Try removing all the MCPS. Try to just straight up raw dog it and see where you get right. Raw dog prompting. Bro, chill, chill, chill, chill. Okay, okay, okay. I got to get back to my chat. I I've been kind of ignoring uh what's been going on because now I've been I've been kind of low-key waiting, right? Like waiting to see what some of these results will yield us. And uh let's just see what's up.

All right, so in our plan, this is a nice plan from cloud code. Let's see what's going on here. So it basically in the non-thinking like I just did cursor claopus 4 I think it's 4.1 yeah cloud opus 4.1 as far as a researcher and trying to understand my crazy prompt it's it has an implementation plan to add the react tinder card the graphql for the API access the react confetti and then the github integration for ooth as well so there's an app integration using clerk for authentication and and then store the access keys securely in convex. This is pretty cool. It so it knows about my convex database. And then it's talking about adding the schema as well here. Daniel and so now it can create new p new pages and components. So the main swiper interface is here. Okay. Okay. Components to create. Oh, okay. That's actually kind of hot. I like how it broke it down like this. This is really good plan. Uh card feature designs. Each card will deploy display profile. I mean, I feel like I should just keep going with this and see which one wins, right? Like side by side, right? And so, I like this. So, in React Swipeable, if you've ever used that framework, you can have like swipe up, swipe down, swipe left, swipe right, you know, you have like four actions. And I like how it's actually understood that here from cloud code opus 4.1 and actually is detailing that. So, it says like swipe right is going to do this, swipe left, this star up, and so forth. Uh, daily streak counter, XP review system, speed reviewer. This is kind of fun. Damn. Opus is Opus is cooking. I know this is not an Opus stream, but still. Damn. Is it ready to be cancelled though? I mean, would you cancel your Claude Max plan? Like, hit me up. Test Gro 4, please. Oh, yes. I still have the heavy plan. So, that'll be really interesting. I don't think I have the Oh, Grock 4 is in the cursor API. Dang. I'm going to have to fire a new window and then do like Grock 4 and then do the same thing. That'll be that'll be really interesting because that'll be that'll be really cool to see how how much it fails or not. Just testing it with ADR to keep things simple. Can't wait to do some more agentic tests, but it's looking interesting so far. May not. Um maybe not the leap I was hoping for, but we'll have to wait and see. Yeah. Yeah. Yeah. Um so GPT5 is in VS Code in public review. You have to enable it in your copilot settings and then just show up for sure. Okay. Uh why cursor doesn't have GPT5? I just updated it. you may have to enable like early feature access in one of your like beta settings. And so if you go to the settings at the top for the user, you'll see like a thing saying early access or whatever the the beta things are cuz I'm pretty sure they're rolling it out to the early testers and or people who signed up for the that that type of uh configuration. So uh how can we use GPT5 in cloud code? That is to be determined um because there are two different things. There's a standard for the way that OpenI has their API spec and there's a standard for the way Anthropic has their spec for tool calling. So if you just straight up change the base URL when you give it that type of tool calling thing, it's going to freak out and be like, "Hey, I don't know. I don't I'm not familiar with this object." There are things like open code and there's some other interpreters like some proxies that will try to do that type of finangling for you, but it may not get the results that you're thinking of. And as of now, these folks it have have had more time to actually fine-tune, not fine-tune, but actually it sounds like I don't know how long these these companies have had these models for, but they've been tuning the tool calling for it, right? Cuz there's a lot of parameters they can tweak to make this a really good experience. Um, and it thinks a lot. Yeah. Okay. Yeah, that's what's up. Let's get back into here this plan. Okay. So, this approach makes the review. It's talking about the Tinder card manually, accept improved edits. So, that's an interesting one. Let's just see what the Ultra Think did for us. It's still thinking. Wow. Okay, so it did a task here and it says, "Okay, so now it wants to do this. It's like, yes, don't ask for the docs anymore." Dang. Ultra Think is just like chilling. It's taking more time obviously, but that's taking it it's sweet time. Let's go back to um cursor. I think for here, let's just say yes and auto accept. I'm just yoloing. I'm yoloing the the first clot code thing. And then we're going back to like GBT5 over here. All right.

So, um, now it says, and now it added the detail like what do we need to do next? Okay. So, added the complete detail plan, wired a new issue review page, added issue API routes, created the plan, and then added GitHub token, added dashboard issue review. All right, cool. So, it's basically has gone this us this far and let's just see how much it's consumed. So, so far it's consumed 9% or 200 of the 7 7 272,000 tokens for GPT5. And so, now that we have that there, we're looking at I'm just going to skim through this plan here and see how it it kind of has fared out because it already did some work by the way before it generated the plan and that that last chat was kind of telling us. So the overlays just to say approve changes and this. So it only got the left and right directions which is kind of what I said. The claopus one did a little bit more than that for different directions. But that was really interesting you know swiping up and swiping down. Uh the scope of the MVP is basically going to be query GitHub API involves is me at open and then for the right event approve request. So, this is actually dealing with the GraphQL database or GraphQL queries. And this is the stuff it's going to send if I swipe right or swipe left. Yo, who wants to yolo test this out for me? Are you going to be down? This is going to be crazy. Y'all going to be pushing some crazy stuff in when you go live. All right. Frame motion. We got server rules. We got off. Okay. Uh, yeah. Okay. It has the future per. So, the use a server side GitHub app token. Okay.

Um, current implementation. All right. App sidebar. So, there's a data model, convex processing additions. Okay. And so, what I wanted to do is have it review itself and see if you can improve some of this and add some more details as a code architect because, yeah, this looks pretty cool. And then I wanted to also review my cursor rules for this type of thing.

So, what I'm going to go ahead and do, since we still have a lot of token window left and it has a lot of this history, I don't think, let's see. I might just open up a new chat because I don't want to cause any more distractors for this type of thing. I just want to have a clean slate and just I want to grab this plan and then do step two. So, in the plan thing, I'm going to grab it and say, "New thing here." So, I'm going to just grab the the little file up here and just put it in here. And then what I'm going to go ahead and do is also grab the convex rules and put that in there as well. And then I'm just going to say, "I want you to review the plan and review the convex rules as well. And I want you to act like a senior architect here and give me a lot more detail in this plan because this is going to be our master plan to go off of so that we can make this a real app in our template. And that will probably mean that you'll have to think harder and review a lot of code files."

So, one of the key terms that they talked about in the keynote today that I saw was that this "think harder" keyword, which is really interesting that they also mentioned that too in the OpenAI keynote, literally from Anthropic, right? And I'm just going to go ahead and stage these changes so that I can see what the diffs are from here on forward. And that keyword is interesting because one of the things that it got me thinking, it's like it does listen for that keyword and can increase the reasoning effort automatically. So, if in Cursor here, the reasoning effort by default is set to medium, and I don't know if it's still dynamic in terms of like medium-low or medium-high, but we're we're going to try it out. I'm going to throw it a bone. I could have even selected like the specific high one to do this type of reasoning. And maybe I might do that as a second pass. So, after it does this, what I'm going to go ahead and do is see what the output is. And then I'll just go back and just say, you know, "Select from medium to high," and see if we get any difference. Similar to what we were doing over here with Cloud Code, how we're seeing this is just basically the like no reasoning effort, and it's doing this type of thing to over here with the high reasoning effort. Damn. Um, and kind of seeing if there's any difference here. So, yeah, that's interesting.

Let's see. Okay, so this right now, it's going through it's it's thought for 24 seconds. It's going to scan the key files to implement the MVP API routes and so forth. And it's also going to follow the Convex guidelines. So, for those who don't know, if you use Convex as a database provider, it's basically a reactive database provider. It's super duper awesome. And, um, oh gosh, let me close this thing here. And what they have done is that they've created their own guidelines to make the AI language models like really adhere to their standards, and they run eval against them. And every time there's something new, they just keep putting and updating the rules as they go. Super duper handy, especially if you're writing stuff with LLMs. Uh, and the tooling and stuff in Convex is super rich, which is really great. And so right now, it's just going through and it's going to go ahead and implement the master plan. Is it going to implement it? Oh, so I can produce a detailed, yeah, implementation plan. So, I scanned the MVP the routes and it says, "You have a working MVP to make this production-grade template, centralize GitHub operations. Persist the state, swipe, and preferences in Convex, and harden the security, error handling, add pagination, and to build a testable abstraction."

So, this is a really good back-and-forth conversation, and you want to treat GPT-5 like it is, like give it that respect as a senior engineer. Take a plan that you've already generated, even if it's rough, throw it back in, and then see if you can actually have this type of conversation to help it improve its own code. So, what the changes are that we see here are: one, we started with a rough plan. Two, I pulled in the Convex rules. So, I'm basically saying, "Hey, this is my database. These are the conventions for Convex." And for Convex, the source of truth will be in the database. And then the front end is basically just going to be reactive to those changes. And the nice part about it is, as the data is streaming into the back end, it's going to create a WebSocket event, and we're just going to just query it from the front end, and that's it. We don't have to write state code for the front end. We don't have to worry about the caching. We don't have to worry about anything. The website loads. You can have three phones pulled up. If you swipe right, that issue will go away on those other phones, right? So, if you have three people reviewing the code and they're looking at the issue, and the person as soon as they they actually take care of it, it's going to disappear from the other phones, which is crazy, right? So, you'll see that happening in real time, which is kind of like the perfect use case for something like this, you know, Convex, right? I don't have to write that extra code. Giving it those rules files, talking with GPT, reviewing the plan as an architect. Now, you can kind of see that intelligence that it's doing right now, right? It's thinking like a senior engineer. It's like, you know what? Since you have this modeling already, why don't we just actually do that, right? It's talking about our state changes, keeping them up in Convex, so that way when somebody does something, it just immediately shows up in the UI on the other side, you know? That's super duper smart, and I think that's like probably the best way to do it.

So, uh, Alan, thank you so much for the super chat, man. Let's cook with GPT-5. Alan Chua, thank you so much also for being a member as well. Y'all are cooking in here. So, that's what's up. Oh, man. This is so great. Uh, I've been, uh, Hello Ray, new subscriber here. Hi, folks. How's it going? Yo, tell GPT-5 to think hard. And it's really hard. It thinks it's time to analyze and think and plan. Really cool. So, that's this thing we're going to try with this model right now. So, I'm going to give it the same prompt and see if I can actually get it to think hard and see if it makes any difference here.

So, there's a new feature in Cursor that lets you basically split the chat or copy the chat and have the same type of conversation. And so, in order for us to do that, we can go to the very bottom of this chat and say, um, there's this little three dots here. So, now you can duplicate the chat. So, if we wanted to, we can duplicate the chat from right here and then continue down this new code path of like, "Okay, now that I have the plan set up, I want to do everything. I just want to continue that." So, you'll grab all the context from this and go on forward, and you can try something, and if it doesn't work, just throw it away, right? You can snap back to it and just say, "Throw it away," which is super handy. If we go to our Git, we now see that there's no changes, right? It literally was just this conversation of research that it did. Did a lot of research. Holy smokes. And it thought pretty hard about it. That's pretty epic, actually.

So, what I'm going to go ahead and do is I'm going to duplicate this chat, and we're going to see the difference between the medium reasoning and the high reasoning, right? So, let's go ahead and do this. I'm going to duplicate the chat. So, now we have the same duplicates shows the number one up here. And then in here, I'm just going to say, "Instead of GPT Max, let's get high, my friends. Let's get high." Are y'all ready to get high with me right now? Oh my god, I'm so hyped about these models. It's It's just It's absolutely crazy that we are entering like this new era, I feel. And it's a new era for a lot of different reasons, and you'll kind of see why the more you play with this model. And like, let's just go ahead and go. I'm just let it do its thing. So, literally, we're we cloned the chat. We went back to the previous thing, and we're just selecting high now. And so this is the equivalent of like "think harder" or "ultra think" in Cloud Code. So, if you're familiar with using Cloud Code, I want to see how this like model thinks intelligently about things. And so, um, I'm really curious here because one of the things that's kind of my favorite is interacting with the UI. And that's a big component here to understanding whether I should keep my Cloud Code plan, you know, is is Cursor with GPT-5 good enough for me to just use as an everyday type of model compared to just Cloud Code and those like weekly limits, right? So, this is a really hard problem because it's literally $200 for the same thing. But over here, I get a UI, and I get GPT-5, and I still get Claude Sonnet, and I still get Opus, right? Like, you get everything still, right? So, maybe it might be a balance instead of paying $200 for Cloud Code. Maybe I'll just pay $20 a month and use some Opus, you know, limits whenever I can. Um, or actually, Opus is only available if you have the $100 a month plan. If you have the $20 a month plan in the CLI, it's only Claude Sonnet, but in the user interface, you can use Opus, but it's low-key nerfed, but they don't give you a lot of usage in Cloud Code, uh, in in Cloud, you know, Anthropic system. So, if you put that money over into like OpenAI, you still get access to the all of those models and GPT-5, and you know, you can kind of experiment. So, maybe this month you cancel Cloud Code and then just go all in on Cursor with the ultra plan and try it out, like, you know, and but I you don't even have to do that. You could just try it out apparently for free. I mean, sheesh. I mean, cuz like the medium model was cooking for us already. Like, apparently I was coding with the medium model. I wasn't even using high. Like, right now we're getting high, and it's even searching all of this stuff on the web. I don't know. I don't know. You you tell me. Oh man. Uh, Claude is offering 20. No way, bro. Wait, that's the hack right now. So, if I go cancel my plan, 20% off for the next 3 months, bro. Where has this been? They're all like shaking everyone down for the dollars, and then OpenAI comes out be like, "Yo, yo, yo, yo, don't leave us, bro. We're still here. We got you. 20% off, though." Damn, bro. I can't right now, bro. I can't. I can't these model providers. I got one word: drug dealers. I've seen this game before. I've seen this game before. Oh man. Y'all are not playing right now. I cannot wait to click cancel on Cloud right now. Bro, the same here. I recommend having VS Code plan, $39 bucks on Cursor. $20. $20. Never hit a limit. Mayhem. It'd be interesting. Do you get the speed that you get from using Cursor? Right now, I'm high, and it's I mean, it's it's doing a lot of work right now all throughout the codebase. And just the medium plan alone for me was it's it's super snappy. You know, medium is snappy where I feel like this is, you know, like everyday model. I I can code with this every day, and it's just fun. It's fun for a lot of reasons because it takes more time. It looks through more of my code. It's everything that I wanted from Claude Code, right? It's now happening in Cursor, which is kind of crazy. And like, if you want to step it up, like we're doing right now, we want the high. We want to solve that challenging problem. Snap that thing into high for one request, right? We don't have to go max all out like I am right now, right? The only reason I'm doing max right now is because I'm balling out. You know, you dropped the $200 bucks. Like, we're going to put this thing put this thing to work. But look at this. Damn. This was the medium one. This is not even the high one. Current implementation for quick references. Feed the route shape in the GitHub token usage. Review the route. Swipe up, uh, swipe UI and interaction. Okay, it's got that. Damn, it's talking about the middleware as well. Like I didn't see that middleware plan everywhere else. Let's go. Convex schema. Okay. Okay. It's the data model. Follow the Convex schema guidelines defined in here. Okay. No dot filtering queries. Jesus. This is just the medium model, like medium reasoning. You know, like you have GPT-5 medium reasoning. Oh, this is nice. Oh my goodness. Public mutations. Swipe records. It's talking about my Dang. Let's just cook with this. Let's RUI architecture. Okay. Off, bro. Use GraphQL search. Come on. Oh my god. Sentry. That's This is Let's just Let's just go. Let's go. All right. All right. Okay. So, I think Did it Okay, man. This plan is epic. Can you write this out into a markdown file? Oh, no. And to, let's see, let's see. Oh, I'm using WhisperFlow. And with WhisperFlow, by the way, if you say "@" any file, it'll actually grab that file and put it in here for you. So, that's a new feature in WhisperFlow. Uh, if you just say "@config", it's going to actually do that for you. So, you don't have to say the whole file name. It searched it up for me. It's so fast, right? It's so crazy. Yeah, that that's so fire right now. Um, so, yeah. Uh, if somebody wants to drop their referral link, go ahead and drop it. I have so many referrals now that it's just like beyond like how I can use it for like 5 years now. It's like, "Oh no." If somebody else dropped the referral link so that people can use that type of thing. If you have that, uh, make sure you do that for sure. Um, all right. Let's go. Let's go. Okay. So, uh, man, this is epic. Oh, can you write this out into a markdown file so that others can um, oh, markdown file so that another language model can start to take chunks of this plan and start to implement it for us. The goal is for it to be organized in a plan so it can follow, follow it in a sequential manner and very detailed. So this engineer is really good at writing code and can take these instructions from an architect and start to go into the details and search through the different files it needs to search through in order for it to write really great functional code and do a small test iteration to make sure that it's implemented correctly in accordance to the larger plan. There's no way I could type that out. All right. All right. Let's go. Let's go. Let's go. Let's go. Let's just see what happens. This is the medium model, by the I'm still medium right now. Just the default GPT-5 Max right here. We're maxing on medium right now. And I think this is offered free with credits for paying users with during the launch week. Try it out. Damn. Right. Like, by the way, I'm not like sponsored or anything. I'm not getting the bag from Cursor. They give me access to the model because I've just been like like I've been talking to them for a long long time, and they're they're here in Silicon Valley. So that type of access has allowed me to, you know, reach out to the team, establish this type of relationship. So I've, you know, they don't tell me exactly what model or something I'm testing sometimes. And I'm like, "Yo, this is crazy that like we can't talk about it." I was like, "Okay, I don't know what to talk about. There's nothing really written." But now that GPT-5 is actually released, I was like, "This is what I've been testing, bro. This is this is what I've been testing." Oh my god. And if you haven't seen my memes, I dropped a pretty hard meme on X right now, and it's probably going bonkers because I can't right now. I can't. I can't. This is so It's so tough right now. It's so tough.

All right, so check this out. I started out with this. I just said, "Yo, I'm canceling my Claude Max plan. Holy moly, more tomorrow." And then like I had to kind of follow it up, you know what I mean? So, as the as the keynote was going, I was I was revealing that it was like the GPT-5 thing, and then like eventually Cursor with GPT-5, obviously, like if you kind of figure that out. So, that's kind of what we're doing right now. We're testing it, and we're basically making a Tinder UI for our GitHub issues, and we're trying to have it generate a plan and generate like a very detailed plan. And now that it's generated the detailed plan, this is going to be where we see the sauce with the actual models, right? It's been really interesting to see how much context window it's been taking up. Has not taken up a lot of context window like I was thinking it was going to, right? I thought it was just going to take up like, you know, 60, 70% of the context window just by reading all these files, putting stuff into the context. But one of the richness that you get in Cursor that you don't get with Cloud Code is the fact that it has its own indexing, has its own file tree association with what files are changing in your file system. So that's that's actually an advantage because they're basically extracting that context away from you and they're not counting it against you when you use your your requests, right? Especially for research tasks. The fact that I can do research with this GPT-5 model on Medium is super amazing because you see how long that research has been taking to go on the web to grab information to look through your codebase to understand the structure. I didn't have to explain a lot of different things. I just used my template. Especially if you have a large codebase. I think this is going to break the bank and going to put some of these other companies out of business, if you know what I mean. If y'all all you other IDE providers out there, y'all should be scared right now because this is I'm I'm hyped about this. I'm so hyped about this. Like like it it's just changing the way that you think about coding models in a way that I don't think other people are really talking about, right? Everyone's going to talk about benchmarks. Everyone's going to talk about specs. Everyone's gonna build a stupid snake game. If I see another snake game on YouTube, get out of here, bro. Like, we can't be doing that right now. You can't get like a PhD in your pocket and just tell them to go build another snake game, bro. Like, stop. So, stop with that. No, man. We got to make billion dollar SAS. We got to contribute to the economy. We got to have like developers be excited about changing code just by swiping left and swiping right, right? So, like I can't right now. I can't. I can't. I'm having a moment right now with with AI right now. This is This is my AGI moment right now. Oh yeah, it's terrible. Yeah, I'm ready for Cloud Code to respond for sure. Yeah, snake games and one-shotting things. Yeah, come on man. Yeah, pharmaceut phenomenal cosmic power l one rule. No more snake games, bro. I swear if I see another snake game, y'all want to make a snake game right now? What's Ray building? Gary, we are building the Tinder UI for GitHub. So right now developers like if you want to merge code in or if you're having agentic work done like I want to have a UI that looks like Tinder where I swipe left and swipe right to just merge the changes because the agents like the if you have GPT-5 level intelligence like a PhD person like this all up in your codebase doing stuff in the background with background agents. They're going to queue up all these different things. I want to see the change. I want to see a beautiful UI. I just want to swipe right, swipe left. And if I swipe right, I want to dig right into the details. Okay. Oh, this is a nice, sweet little thing. Yeah. So, that's what we do. That's what we do. That's what we're doing right now. Uh, I'm testing it with finance and logistics. This will be a total game changer. Oh, that's going to be really interesting. Next Gemini opening. Hey, has anyone checked on Meta? Like, Meta, are you still alive? Are you good, fam? Like, somebody check to see if they still got a pulse right now because they like they're the Los Angeles Lakers of AI. If you for those who don't know, like they spend a lot of money getting all these like really expensive players, and we're going to see what they drop. Like it may take 5 years for them to like drop something phenomenal, and I'm really curious to see what happens. I think there's going to be more drama than we do realize, but hey, if they come in in with the heat, that's what's up. But yeah, we need to check in on Meta right now as well. Uh, give you all our impression of the new model for coding. Is it beating Cloud 4? So, this is kind of what we're still figuring out right now. We're in the planning phases. Um, as a daily driver model right now, if you're using it, because I I actually got to use it like within the last 24 hours. I like this conversation back and forth is is really really really nice. Like this is kind of my favorite thing actually, is to have these conversations, go for a walk, kick off a background agent, kick this off in the background, just let it do its thing. Right now, right now it's writing a very detailed review. It's It's been thinking for like a minute and a half, bro, on a medium. We're not even ultra thinking this, or we're not even like uh thinking high on this, you know. We're just literally just doing this type of thing. Oh my goodness.

All right. So, what I'm basically having it do is is write out all of the instructions so that I can kind of review it a little bit before we have it just implement the code in my codebase. And then here, this is this is the final, this is the one with the actual thinking with the high model. So, this is GPT-5 high. Um, I I'll probably nickname this model Snoop Dogg. I'll just call this like uh GPT Snoop. Let's get high, friends. Let's get high with Snoop Dogg. All right, so this is actually offered for free, too, bro. They're just gonna give you high reasoning for free. I'm telling you, that's how these drug dealers work, bro. They just give you that first hit for free. Hey guys, come on in. GPT-5 is here. It's free for the first usage. Like, you can have as much as you want, then try to leave us. See what happens. Make sure you pay your bills, friends. That's all I'm saying. Oh my god. This is the casino. Oh my goodness. Let's go. Let's go. Look at this. Oh my goodness. This is GPT-5 high. Thought for a minute 59 seconds. It went through a lot of code. It went through a lot of searching on the web. Try to understand some things. Then I'll draft the expanded master plan with architecture, APIs, Convex schema, functions, security, testing, rollout steps, proposed security measures. I'm thinking through proposals that include token management, route protection, and environment variables, rate limits, backoff strategies. Testing is necessary, so I'll suggest the unit, integration, end-to-end test. Metrics and logging are important too. This is the era. I feel comfortable now being a vibe coder. Like, if the model is actually thinking this out loud. Oh, that's gorgeous. That's gorgeous. Clarifying route prop. Damn. I might have to stay high all the time, my friends. Damn. Let's go. That's Wow. That's gorgeous. Oh. Oh, so this is the executive summary. Keep the MVP and XGS routes. Introduce Convex for stateful persistence. Yep. Expand the data model for an API support, item potent review post. Yep. Issue only decisions and user level filters. Formalize Convex queries, mutation actions using the new function syntax, and validators. Oh my god. Let's just cook with this plan. This is I'm just skimming through it right now. This is This is gorgeous. This is so nice. Oh my god.

Okay, so I'm going to have it write this to an MD file and call it the "goated plan" because this is goated. Look at this Convex function skeletons. Concise the swipes. It's even got the swipes. Damn. List risks and trade-offs. So, okay. What do you need and want from me? Implement the Convex schema. Add this and wire to the UI. Extend this and return the labels. Okay. Wow. Okay. I want you to write this into the documentation folder and also want you to kind of name it as the "goated master plan." Yeah. There we go. There we go. There we go. Oh my goodness. All right. All right. So, I know it's taking a little while to get the planning stuff, but the planning stuff is actually the most probably the most important thing for then like we're just going to let the model rip. It's probably going to go for like Dude, look at this plan. This is This is the This is the one from the medium model. Can GPT-5 rip through almost like a half like 500 lines of a plan? This is amazing as far as like level of detail, by the way, from a medium model or like medium thinking, right? So like GPT-5 is a is a pretty strong model, but like the level of detail for coding is impressive. Damn. Okay, cook. It is cooking right now. I like this name though. I'm not going to lie. Oh, do you think GPT-5 will outpace Grok 4? Heavy Tyler? I think I don't know. I think they're two different things to be honest. One is this tool calling slash reasoning. So, what we're seeing in Cursor, which they're kind of showing us, and you won't see this from the user interface, by the way. If you go to Chat GPT, it's not going to do this level of like multi-step thing. C the Cursor team has had time to integrate the entire model because there's like eight different configurations this model have, right? They've had time to integrate that plus build what like I call hooks into their their IDE. So when you're using Cloud Code, it has to generate these shell commands to like literally GP through your codebase, look through different things, try to find semantic reasoning, and then try to understand and build its own context to feed that off to the language model. So that's that's like one layer of things that's Cloud Code up here. You have Cursor. So Cursor is above Cloud Code right now because they have a rich IDE. The Cursor team has ML models that they train that can look through the code and understand these changes. They have their own like ways of interpreting how a files are actually hier like are stored in your file system with some type of hierarchy. So when there's a change, they can detect it, and they like they'll literally just do this like old school method of like, you know, a tree, right? Like for data modeling, right? And so they can very quickly understand what's changing in your code, understand relationships. So they have to build all that tooling themselves, and then they feed that into a language model like GPT-5, and this is kind of why I'm saying like this is unlocking a whole new era, and this is going to allow them to really expand out. So like, if I was an investor in Cursor, I would throw like double my money back into that thing because that tooling is extremely rich. So you have the rich tooling capability, right? Cursor has like a really long runway, and then on top of that, you have a smart model like GPT-5, which can follow instructions very well, which is what they were documenting in the GPT-5 announcement, so that you can tune in, like you can tell it saying, "I want to extract data points. I want you to think again. I want you to think again," and I want you know, it can do this multi-step type of thing and just keep going, can keep going. So whenever you have really high intelligence, you have really good tool following like that. Now you're opening up people like Claude Code, you're going to get other tools like if you have like Open Interpreter or some type of open code type of platform that's a CLI as well. You tack that on the GPT-5, it's just going to feel smarter right out the get-go, right? Without any tuning of anything else. So, you know, Cursor is already kind of doing that for us, and then kind of going a little bit higher and like, I forgot what I was even launching into, but basically kind of saying that, um, I think I don't know because Anthropic is not making their own IDE, they can't get that level of integration and understanding about your codebase, and they can't get that like, there's only so much you can do with the model, and then the rest is really up to everyone else who builds on top of that, right? So that's the advantage that you have with an IDE like Cursor or Cloud Code, right? Like the IDE, the terminal can do so many things, but Cursor itself can take it like that much further and do it on steroids, and they could do it multi-threaded, and they don't have to eat up all the context window. They don't have to send that back and forth to a language model and eating up all your tokens. They can just do that in the background for you. So damn, damn. Damn, damn. I think this is the moment that I had when I used Claude Sonnet 3.5 inside of Cursor in August of last year. Oh, it's literally a year from now. This is how my whole account started with like less than 100 followers and less than 100 subscribers or something last year on YouTube. The same exact moment here. Oh my goodness. We're cooking.

So, I created this. Let's take a look at the "goated master plan." Let's see how many files. Damn. Oh, and then okay, Grok 4. So, Grok 4 tool calling like Grok 4 has a long ways to go. It can get there. It has some intelligence as a base model. So, it just needs those other pieces that GPT-5 has like tool calling, the code model. That's why those models haven't dropped yet, right? They dropped Grok 4. They dropped Grok 4 to be able to do a lot of processing, but they don't have those other tool calling stuff that they have to like actually do it in post-training. So, in this post-training phases, they're actually basically trying to train the model to accept tool calls, to train it for the outputs, to make sure it works well with human instructions and back and forth. That's what GPT-5 has already done, and they just said, "Here we are." They just said, "Yo, what's up? Our balls are heavy, bro. We're swinging." So that's why I'm thinking like it's Grok 4 probably has a long time to go. I will probably won't see that until Grok 5 probably. Maybe we'll see. Like, unless I get inside access. Yo, Grok, holler at me.

All right, so this is the plan. This is the master plan. Dang. So, this is the master plan that we were reviewing up here earlier using the GPT-5 high. So, the Snoop Dogg model. You got five on it, right? Yo, what's up? Okay. All right. Let's go. Let's go. Let's go. So, I'm just going to tell it to I'mma open up a new window. So, I'm just going to say, "I want you to implement this into my code," and we're just going to let it rip. So, before we do that, I'm just going to go make sure I check in these changes up here. Oops. Yeah, check in all these changes. I'm going just say, um, I should put it in a different branch, huh? But that's how we roll. We just do YOLO. Okay. What branch? I'm on main right now. Yeah, I probably should stage it in a new branch. Okay. Let me um so one of my my favorite ones to work with right now is Git Tower. Let me make up open up a new window. Open up a new window. Okay. So in Git Tower, I think I'm going to call it three commas. Yeah. So G tower dot. All right. Cool. So in here, I'm just going to make this adjusted for this type of thing. All right. Right. So, I'm going to create a new branch, feature yolo plan. So, I'm going to put that in there. I'm going to stash the changes so that they come back on and get snapped right back in. Okay. So, all of our changes are snapped back in. And now we're in the feature branch so that way I don't check them in the main like a good developer should be doing. Not me. Nah, I'm just vibes, bro. All right. So, then let's go ahead and check these changes in. Uh, we can go ahead and publish it if we want. Sure, why not? Um, and then what I'm going to go ahead and do is now create the new thing so that way we can kind of work here from here on forward. Okay, so let's see "@" goated master plan, right? Cool. And I'm going say "@" convex rules. "I want you to start implementing the goated master plan and you'll need to think hard a little bit and search through as many files in my codebase and don't try to tackle everything at once. So I want you to kind of chunk this out and implement the plan sequentially so that we can implement this feature. This should be what I call a minimum lovable feature. So it needs to work and work reliably, and that's kind of what we're focusing on here. Not anything that's too fancy or too crazy. Uh, but it should also leverage the existing patterns and stuff in my codebase. So please go forth and cook. Let's go."

Okay, so we're literally just lo yoloing in this plan. Oh man, things not to do. Like, do not take my any of my coding advice. This is just me doing this for entertainment purposes. Uh, anything that is said here, uh, I am not liable for. This is literally just me, uh, exploring this whole type of model in the way that I do it. For those who are wondering, uh, I used to be an engineer at Apple for like over 12 years or so, uh, over a decade. And now I'm just kind of using AI and playing around with it and trying to understand what these models are doing. And I like to try to share back as much knowledge as I go. And I know a lot of people who have become members have really enjoyed kind of hang out in the Discord with other people who are building. So if you're wanting to do that, all you have to do is just, you know, hit the little join button on YouTube and you become a member. And then there's some instructions to become a member in the Discord. And so like with with that, that we kind of share best practices and kind of do different things as well. And um, I I just kind of like hanging out with other folks who are kind of like-minded in this community who are trying to really cook with the software because I feel like uh, we're doing some really amazing stuff in this uh whole type of thing here. Oh my god, there's 496 people on YouTube watching right now. I had no clue. I had no clue. What am I doing? Okay, at le 1,129 on X. Y'all are crazy out here. Oh my goodness, bro. I'm so behind on the chat right now. Oh my god. Okay, I should probably make someone a moderator. Okay, so this is going this is going to be really really fun here. Y Oh my god, I got 111 likes. I got to start dancing, too. Oh my god. Okay, I'm so behind, man. But I'm trying to explain things, you know, as I go. So I feel like you can get a better understanding of the way I think about these types of things. And then like I've been playing around with these AI models for the last couple of years. So I've seen a huge evolution of the way that these models have really progressed in this era. I mean from GPT 3.5 to four and then to now five, right? So we've we've gone a long way in this type of agent encoding. And then what took me like over the top was last year when I started using Cursor was Sonnet 3.5 because I was literally copying and pasting code into Anthropic's actual UI and then getting the output from Opus and then pasting that back into, you know, my IDE, and then when that got finally integrated into Cursor and I had to like generate some apps, that was like technically like I was the first on the scene to like vibe code, right? And then um like that's was like really blew my mind how you can just go into V0, make a prototype, pull that into Cursor and just have it keep going and fixing its own errors. And I think that kind of like really started this whole movement of like the capabilities of understanding these something is really intelligent underneath the hood, and it's only going to get crazier from here. And it scared a lot of people too. A lot of people who are experienced developers are like, "Bro, it's not good at coding yet." But it wasn't that it wasn't good at coding. It's more that the more data you throw at these models, the bigger the models to get, the more general they are. And then if you can make it do these types of things like tool calling, now you have an intelligent system that doesn't need a lot of instructions or specific code to do different types of stuff. You can just say, "Just go participate in the world and then see if you can do something interesting." And that's kind of been my journey as a curious engineer. I like to do stuff like this to play around with it and see if I can get anything back from it. And um, what we're what I'm actually witnessing right now is the next leap up in AI. And so it's kind of hard to see it right now because it still looks like everything else, right? Your still looks like Cursor. It still looks like, you know, coding with Anthropic, but like some of the sneak peeks that I'm seeing here are like these extra. We're going to see some things, right? Like as as the code comes back, I'm going to show you kind of what I've been seeing. Uh, going to proceeding open open relevant files in parallel. So like this is actually doing a parallel tool calling here. So they they've done a lot of work. It looks like the Cursor team's done crazy amount of work to try to get this integrated, um, for everything to get working. Let me turn on my other light. I forgot to turn it on. Okay, cool. Now I looked pretty. So, okay. And so that's kind of what we're seeing here. We're seeing this model take advantage of this. And we're not even like flooding the context window, which is amazing. And right now I'm using the high model obviously, but I could have probably done this with the medium model, and I think it would just cook super fast and super good. And um, yeah, that's I think what would really happen right now. Um, uh, you're blowing up. Nice. I could be mod. Yo, Carl's so chill. Yeah. Yeah, I might hit you up with the mod mod roll. Um, yeah, let me know if there's anyone. I think I'll make Mr. Frank mod because he's he's like my oldest member right now. I think six months or so. Yeah, Mr. Frank, you're going to be the uh, the mod for my uh, add as moderator. Let's see. Can manage block words and all that stuff. Okay, cool, cool, cool. Let's go. Cool. Yes, Mr. Frank, I gave you a job. Uh, Meta, you need to wake up, bro. Okay, so yeah, Meta needs to wake up. We need to see what's going to happen with Anthropic. Are they going to respond back? Because they I think OpenAI did a really clever shakedown where they were trying to get everyone to release their models on Tuesday, and then like a bunch of people dropped their stuff, and they're everyone's like, "Yo, is GPT-5 you good, bro? Like, are you still in there? Are you going to come out?" So, yeah, that's what's up. Yo, Mr. Frank. Yeah, with he's got the little the the moderator role, so that's what's up. Um, yeah, I agree. Everyone is sincere and sharing rather than posting toxic remarks. Yeah. Yeah. No, that's what's up. Um, that's what's kind of going on in the Discords as well. Uh, which chat tool is he using? Um, I'm using, you mean as far as like the voice dictation stuff? It's called WhisperFlow. If somebody has a a referral link, please drop yours in. I have way more referrals than I could ever do with, and I feel like other people need to get some of the love, too. Uh, it it's such a good one. I really like it. Do you need Cursor Pro and have a subscription to each AI you want to use? I do right now, Justin, because I'm still like I like to evaluate, and part of being an engineer, too, is just trying to try some of these tools. Um, hopefully you get paid well enough that you can actually try a lot of these tools as well because I don't know if you know this, but like the amount of time you can save just by researching through code or having like these really nice conversations with the model like that, that's that intelligent, this does save you a lot of time whether it's Cloud Code, whether it's Cursor, whether it's, you know, whatever your favorite model is. I just think the bang for the buck is really going towards OpenAI now, especially with this new model because I haven't even launched the GPT-5 app, right? The ChatGBT app, right? There's a bunch of cool stuff in there that we're going to take a look at eventually, right? So, um, yeah, that's just nuts. Absolutely nuts. Yeah. Yo. Okay. Like, uh, that's why I like to hang out. This channel is vibrant, full of like-minded folks. Yeah. Yeah. No, this is a vibe here for sure. Uh, so, the normal mode has the thinking mode, and then there's a thinking mode. Yes, there are multiple modes. Let's just kind of go through those while we're waiting here. So, let's take a look here while we cook for a sec. I'm going to let it do its thing here. And I go to models. So, this is really funny. So, we have GPT-5, right? So, GPT-5 says it's offered with free free credits, and that's basically the version that they call medium reasoning. So, there they say it's medium reasoning. And then we have GPT-5 high with the high reasoning. I call that the Snoop Dogg model because all we do is get high. And then we have GPT-5 fast. So, this is the same GPT-5 model but using the fast priority. So, in the OpenAI uh API specs, it costs more, and you get like, you know, responses super fast. You're basically not in the queue. And so, they're offered with the free credits as well, and that one's the medium reasoning. So, if you want the high reasoning with the fast, also the high priority, you would just basically just tag that one in for it. Um, let me see. GPT-5. These are all the GPT-5 configurations, by the way. So, we did GPT-5, which is medium. We have GPT-5 high. Then you have the fast versions of them. So you have high and fast, and then now you have the low ones, which I did not enable. So, the

Here is the translation of the provided text from Scots to English:

low the low one here they let me do I have to enable it for it. Okay. So the low one says it's basically low reasoning effort. It's the same thing. Yeah. So low fast and then we have nano and then we have mini. Okay. So what's nano? Does it say anything about it? No. It doesn't have any tool tip here. Okay. And we have mini. Okay. The baby. We have the baby model. Okay. These are all the configurations they have so far. And on OpenAI's website, let's just take a look at that real quick. And here's the pricing. So, if you're going to just do an API call, this is what you would do here. You basically just would see the standard priority and you have priority pricing. So, the priority pricing is only for the five in the mini model and they have basically 2.5 input and then you have 20 bucks for the output there. And then you have flex, you have batch processing as well. So, all of them are there. So yeah, that's that's an interesting kind of take here. So yeah, that's kind of that there.

All right, so let's just see what's happening right now. A lot of files are being changed. So it says I'll add convex tables and functions for the swipes. Nice. Here's the schema. Here we go. Approve act needs discussion. This is great. Super chill. Swipes.ts. Okay, so this is all in the the database side. Swipe session. Cool. So, this is all in the com side of things. Let's see. Noise. Okay. Swipe sessions and then GitHub settings. Cool. I mean, I didn't I didn't do the fast model because I just wanted to basically not hit rate limits, but it would have been interesting to see the high priority ones and see what how fast that is. So, now it's now it's actually doing the swipe route. Let's see what's going on with the swipe route. So, the decision is here. It's just saying you only have these decisions. That way you can't pass anything else. So there say repo swipeable sessions as string. This is nice. Okay. Okay. Wow. Yep. It doesn't exist yet. So this is where it's going to do the next move.

This is kind of my favorite thing about cursor is the fact that like it does the code changes and then GPT5 as a model can like rethink itself after all the all the stuff has been done. Similar to what was happening with Claude Sonnet and Claude Opus inside of cursor but now I feel like it's kind of taken to the next level right holy smokes this is nice so this we're seeing its thinking reasoning here I'll add a public convex mutation to the log of the swipes using convex off and update the new server API route to authenticate with convex and call the public mutation bro we've been letting it cook oh my goodness okay so now we were earlier talking about how does this compare if you're just using is it like worth canceling my cloud code subscription yet you know we have the same thing is it as smart as intelligent can we do the same types of things that we've been doing right now in cursor with GPT5 but also do with cloud code like if you're already paying the max plan is it going to be even worth kind of going that way right cuz if you're just in the IDE and you have your workflow I feel like I've invested like a month and a half plus in cloud code so much like all of these workflows sub aents even the way that I think about how to talk to the AIS are all right now claw pilled. Like I'm so clawed pilled right now. But in the last 24 hours of using GPT5, I'm have to blow up all of that old school thinking, right? Like literally, I don't want to be an unk anymore. I need to go back to my my young self and just blow up the way of thinking because when you're seeing model act so intelligently. I don't have to give it all these explicit instructions and rules and claude files and all this really complicated stuff to put in all these different folders and like configurations and MCPs for planning like you saw no MCPs right we didn't no MCP tool calls all that planning was done all inside of chat GPT I mean in a GPT5 inside of cursor right it searched through all the code I blurted this big old like idea for a Tinder UI for GitHub issues it took my existing template and read through all those files I didn't have to do any of that stuff, right?

In cloud code, it's it's doing the same thing a little bit, right? So cloud code has an understanding too. Had to search through the files. It's generating a plan. Does does it compare? Right? Like how do these two compare? I'm seeing something different because the way cursor has just finished their latest updates for uh 1.4.2 and I think this is the new paradigm and people won't be able to see it yet until they've been coding for like about a week probably with cursor, right? using the new models and then you're going to really see like bro was talking about this. This is scary. This is really scary. So, but I just want to give it a fair shot though. I'm not trying to like I think it's right like you should try new things out. Man, this this is this is going to be really hard. This is going to be really hard. Uh but I'm not married to a company, you know what I mean? Like I'm not paid to bag by any of these companies. So, this is just me as an independent engineer. Um this is my own whatever. I'm just having fun. Let's just chill.

All right. So, let's take a look. So, this was the claude code. We did ultra think which basically gave it the max tokens for thinking window. We did parallel agents as well saying go do all this stuff in parallel and research. And then it came back with this plan. So, swipe right, swipe left, and then super like. So, add to favorites and priority review. I don't know where that super like is going to happen, but I like its description. And it even talks about like a plugin for throttling. That's interesting. I don't think I'll need that because there's already something that's built into convex for this. So, I might have to tell it that it should do something like that that's different. Uh maybe I'll I'll capture some of that. I'll say no. Let's go ahead and man uh keep planning. Okay, let's see. For the throttling, there is already a convex component that can handle this. So you may need to do some research on the web to look that up instead of bringing in that plugin that you mentioned from Octoit. And then I'm just going to review the rest of the plan here. So let me just kind of give it some more updates here. So add GitHub convex schema. Okay. To store the GitHub tokens and user preferences. Okay. Review history repositories and then create API routes. And like, can you review the GitHub token storing scheme? I thought that if I use something like Clerk, I'd be able to do some type of OAUTH refresh token thing, but you'll need to do some more research here. I'm not really sure how this works, but it just kind of feels wrong. I'm just kind of curious. Uh, let's see. Um, what else do I need to improve here? Created API routes. So the OAUTH flow issues review web hook. So this is an interesting scheme. Like I always kind of question like why is it creating all these routes? Can we keep it something simpler? Um why are we creating a lot of API routes? It just seems inefficient, doesn't it? I want you to think like an architect and kind of keep things cleaner and simpler to maintain. Oh, I forgot to also like give it the convex rules. So I think it's at um is it at cursor at cursor slash rules convex? Yeah, I will also need you to review these rules as well to make sure that we're following the conventions when using convex. And then what else are we looking at here in the plan? So we have stack of cards, swipe true, match animation, review stats, fun, playful elements, a PR bio. Okay, this is cool. Yeah, the gamification is kind of cute. I like this plan, but like it needs some work, but we're just going to let Cloud Code continue to plan on this and just kind of see what it comes back with. Um, all right. So, CL code in Ultra is going to continue to do the second review of the plan and take some additional considerations and see what it's going to do here. Um, yeah, it's really interesting. Let's just kind of take a look and see what it had done before and see if it tells us anything interesting compared to like GBT5, right? So in our our big plan, we just told it like this is our idea. I want you to do some research. Here's a here's my repo and like go ahead and go cook. So it did some thinking and then what it did here is it you know it learned about rack swipable UI libraries. It used one task with like 24k tokens to do a GitHub API integration. Uh it take a look the package.json to see what I had installed. It like listed these paths and components. It read through some more files. And so this is kind of like what's happening and sort of the limitations of cloud code in a funny way that it has to do all this type of stuff in the background, you know, send all these different tokens. And you know, I mean, I'm not saying that one's bad or the other. It was just kind of you're kind of seeing the differences now, but I think it's going to be more apparent when you start using it inside of cursor like how much smoother and fast the experience is when it's really integrated with a really intelligent model like GPT5 that has that nice function calling, tool calling, and additional stuff that you can kind of pull in together here. So that is interesting. This is a really cool plan. U and then we're just letting it do its thing. So right now it's going to search the web for the integration, the GitHub API integration. Okay. So it's going to do it its thing. So I'm going to let it do that while we're check back up on GPD5 aka Snoop Dogg with the high version. We got the high version. So I'll add the convex tables. Okay. Okay. Okay. So it's still thinking. Okay. And files being edited. Let's go ahead and accept. Oh, okay. Cool. It was just waiting on me to accept. issue review page. Okay. What is it doing here? Log to convex and options post a GitHub review server does both. All right. And then after this is done, what I might do is try to like create a prototype UI. So I might try to see if I can see the database with some fake data and some fake issues to see if we can actually kind of interact with the UI. Uh because that'd be really fun to see how this actually turns out um as far as a whole plan here. So, yeah, this thing is really just kind of doing its thing, yo.

All right. Um, okay. Let me catch up with the chat. I'm so behind. I'm so so behind. I'm kind of low-key going to just go backwards a little bit. Uh, and kind of prioritize my members. Hey, Colin. How's it going this year? Oh, yeah. I'll still be around. This will turn into a video on demand. And a big shout out to my friend Dan. Uh, so Dan is I have hired as an editor. My goal at the very beginning of July 1st when I started streaming was to get my first hundred members. And so I went from like, you know, 10 members or 20 members or something. We got to our first hundred and I was able to hire and bring on Dan. So he's going to be cutting up some of the live stream. So you'll be some of the seeing some of these snippets. Uh, you'll see some of me me kind of preaching and so forth. I'm also going to plan to do some recorded content. So I'm going to have some recorded stuff where I'm going to walk you through like an even simpler plan. Maybe something not as complex as this. But I'm really trying to push the model here and kind of show you a little bit of what's happening in the tool calling like why it's a little bit different compared to like cloud code. And so you'll probably see a shortened video. Uh, my members will get to see that first and then later on it'll be available to the public. And uh, yeah, a lot of this stuff takes a little bit time to film. So I'm going to be definitely doing that. Uh, tonight I'm going to a meet up in San Francisco. And so for those who don't know, Riley Brown is a content creator. I think he has over like millions of views on TikTok and has like tons of followers. Uh, he's also meeting up with this other builder. His name is Yoni and they're throwing a little bit of a party in San Francisco. So I'm going to be there. Uh, so I won't be filming tonight, but I'll definitely be hanging out with them. So that'd be really cool. If you want like an impromptu live stream, let me know and I'll probably just do a live stream on Tik Tok or something. It would be kind of fun. Um, so yeah, that's what's up. Uh, we'll test. Yes. Uh, oh, Ben, I didn't think of all the changes as negative. Ride the AI wave because these these shills will need to be regardless of what you probably get. Yeah. Yeah. Any idea when we get GPT6? Uh, likely I think like 36. No, is it 36? Is it 20 12? Like 20 to 30 months from now. I think it's like the timeline's like 26 months or something like that. It's generally about like a 2-year time frame or something like that is so forth. Like less than two years 18, but I'll I'll actually find someone if I can get them on. Yo, tech friend, man. What's going on? Yo, plus one to me. Are you going to go to go to the Riley party? Yo, that's gonna be fire, bro. It's gonna be crazy. We need GPT5 integrated into that Unit R1 that we had. For those who don't know, I had a robot on live. This guy name is King Butoshi, and he has one of these robots. And so, if we get GPT5 in that phase, that'll be so crazy. That'll be so fire. That'll be ultra fire. I want to see that for sure.

Oh, man. Okay, so I'm switching daily between hyped and being scared of u not getting the chance of having a career. I'm finishing my degree in 12 months. Hey Yan, I have some advice for this actually and being in the business a little bit and now actually kind of being out of it. I think the advice here is um that like if you focus on understanding how the code works like if you work at a company like a a company like Apple was so big right so like they have like 30 plus years of frameworks and of the operating systems and things have been reused over time right so at all these companies like Oracle as well they don't just throw away the code they keep reusing it because it becomes this compounding interest on their returns so either the deeper you you have a couple of options. You can either go really deep into the stack and get out the metal layer. So the the changes you have there are going to be what the AI calls to do the things. So if you want to do something that's really unique and interesting uh as an AI, it's not going to know about it until it writes the code to get to that part of the metal. And so as of now as humans, we're kind of guiding that expertise because the AI doesn't really know where to go with it. So you as a developer can get really deep on the frameworks, use AI to help you write even better, you know, mappings to that and kind of keep building that out. The other layer that you can go is way up in the stack and what I call like the product layer. It's kind of like my favorite place to be. So in the product layer, you're leveraging AI, but you have to also know about how some of these frameworks come together as products. So what you'll want to do is try to get expertise not only in one single domain but cross different domains and try to go really deep on them like really truly understand. So if you are really into medicine and you also understand coding, go really deep with them. Maybe combine something with finance, combine something with like something else, right? Those perspectives, maybe something very creative can help you cross those barriers. And so the era that we're currently in in like our current lifetime, I feel like by the time I die, you know, like a hundred years from now, is like all of us humans discovering that we can actually bridge all of these domains and tie them together with these like unique buildable systems that's going to last like 60, 50, 100 years from now. And the AIs will get smarter and better at doing these things and they're only going to help us cross those domains a lot faster. Right? So, like if you previously used to just be like an airline pilot and you know about airlines, now you can pick something up like this and learn how to code, but then start solving problems for the airlines. You can start solving your own problems. You can make your own inventions to help with airline flying. You know, there's a lot of more capabilities that will be unlocked and that is just a very long ways to go because it's going to require lots of energy, lots of resources. And I wouldn't look at it as negative. I would just look at it as more of like an enabler and then think about it broadly. Think about it across different domains and how can you think like this and start to like tinker with the models like your goal is to throw everything you like every every 12 months like literally a year ago right when I was using GPT uh cloud sonnet 3.5 right GPT 3.5 or 4 was just out I threw away all my way of thinking right like I discovered a way of prompting a year ago with using markdown right to to make these models really kind of follow everything you know you're doing less and less of that. I'm not doing markdown prompting anymore. I'm just plain English, right? And then it's going to get even easier from that, right? You just do like you have to almost throw away everything you know previously and just be born again and approach the models in that type of way because they're kind of foreign and they're being developed as we go. So, um, having that skill set and with the curiosity is just going to make you unstoppable, Yan. And I feel like that's kind of what we need right now in this era of people. Like we need people who are very creative, very passionate, who want to like use this technology because the fact that you're even here and trying to learn about this already shows that you're ready for that next level. Uh and you're around other people like this who are trying to build. So um yeah, definitely start cooking. That's what I'd be doing. Uh and yeah, if you guys want to join the Discord, join the membership now before I double the price. Easy easy way to get in. Other members can vouch for it as well. We're just kind of hanging out. Uh, I'd say there's a lot of alpha because it's not a free Discord and it's super cheap. It's like $6.99, so y'all can get in and chill um for everyone else. So, yeah.

Um, let's see what's going on. I heard the cursor just go off. So, let's see. Cool. So, what do we have here, Daniel? All right. All right. All right. Let's see. What do we get? So, status. I reviewed your convex schema. Okay. What changed the additions? New convex modules. So we made a added the persistence added reliable updated react swipe regenerated types and the lint pass. So we should have an app bunddev. So the bunx convex dev. Let me just make sure all the stuff can run. Wow. Okay, cool. And then that's there. And then here, let me just clear out this Okay. Bon dev. Let's see what happened. Vamp dashboard. Let's see. Issue review retry. Cool. So, we have issue review here. And so what I'm going to go ahead and do is going to say I want to have it generate some mock data for us so that we can reread the issues and see what's going on. If I go into like a mobile view like this, this is kind of what we'll we'll kind of be seeing as far as like our dashboard. And um, yeah, I mean to me it looks like it's actually hitting stuff because we're seeing it like hit the API. And then over here, what are we seeing? Any functions ready? Okay, cool. So uh I'm just going to keep all we accepted all the issues. All right, cool. And let me just go ahead and add these changes. Uh, let's call yolo dev. Uh, oh, whoops. Oh, shoot. If you hit this little star, what it's going to do is going to generate a commit message for us automatically, which is really nice. And then what I'm just going to do is add some information. YOLO dev. And I I just like to check in the changes very frequently, especially since I created a new branch for it just to keep track of what's been going on. And then so what was really interesting here is that um just to for those who are kind of new to the cursor thing at the very bottom, they've added this ability to see how much context window you've taken. So this is really good. We've only used 12% of our context window. We did all of that review with all of those files and all those types of things. did all that thinking, writing all these files, and we still only used up like 12% of our our context window. That's amazing for those who don't know. Like, usually this takes up a lot, and then you have context raw, and you have to keep opening new chats. So, that's super refreshing. And uh what I want to do now is I might open a new chat or I might just duplicate this chat because I want to maybe keep building features here. So, I think the best way to approach this is I'm going to just duplicate this chat and then see if it can actually create some dummy data, you know, based off this conversation. So, that way, um, I don't have to compact things. I don't have to start fresh and it just can continue the conversation. If it doesn't do what we want, we could just reset everything, which is kind of the joy of cursor, right? That's kind of why I like to use cursor, too. And I'm really happy these changes are here and it's so close to canceling and going back to cloud code, you know? you know, like you don't have to worry about like for me it's like do I need a CloudMax plan anymore after all this? Probably not. I mean, that's it's going to get really addicting to start using this model daily. So, let's go ahead and do this type of thing. Duplicate the chat. So, now we've duplicated the chat. We got this here. I'm going to say Oh, when I go to the what do they call it? The slash dashboard issue review. Okay. when I go to and then this I don't see any issues which is good but I want to make some mock dummy data in convex so that we can test the UI and play around with these things. Can you go ahead and see if you can leverage the convex MCP and do this for me? Let me see if I can make sure I turn that on and like watch it do the tool call. So where is the tools and integrations? Okay, so we have that here. Convex is enabled. Cool. So awesome. Let's go ahead and do that. So this will be our first MCP integration and it should be able to talk to the convex database and hopefully throw some dummy data in there and then we can kind of see, you know, what that would look like and then we can just start interacting with our UI and seeing that kind of updating the database which would be super fun. Um, yeah, because it's it's important to kind of test these UI changes and see how that kind of works out because that'd be really cool. And then see if I can publish this live. Yeah. So, this is it's doing its thing. So, it's now doing an MCP tool call and trying to read uh Yeah. and figure it's it's figuring itself out. So, that's really cool. Uh, let's see what's going on here, y'all. Yo, big shout out to Super Kraken. Just became a member. Thank you so much. Yeah, there's a post in there, by the way, if there's going to be a members only post on my channel. You go ahead and follow that and you can send me the instructions in the email and that's how I'll get you right into the Discord uh to get you to join as a Discord member as well. So, yeah. Uh the difference between the high and high maximum ox enable. So, with the max, we're getting the longer context window. And so, if you don't turn on max, the context window I think will be probably half that, but I'm not entirely sure. We'll have to try that out. So, it's going to want to go ahead and accept the code here. So, this has a schema noise and then it's going to do like a mock issues thing. So, it's going to create some issues that we can mock. So, this is the script that's going to run in convex and then it's probably going to let us do a a route call for this type of thing. Let's see. Yeah. So, it's going to call this function which is my script. And yeah, cool. Let's go ahead and run the tool. All right, cool. Nice. Okay, let's see. Yeah, welcome to the club. Yes, see you in the Discord. That'd be so fun. But the first time over, OpenAI has a good model for coding, dude. Um, like for me it was GPT4 back in the day and then 01 and like 01 Pro was really good for architecting. 03 Pro kind of came really good for my research, but this is the first time where I feel like this is my daily driver. Like GPT5 and all the different configurations is now my new daily driver. Like it was Claude Sonnet, then it was Claude Opus and there's it's hard. It's going to be really hard. I there's a lot of things right here. I think one is cursor, right? the cursor sauce. That's right now Jeff's kiss. And if Yeah, I think that's that's kind of where where we're at right now. I feel like GPT5 in its its various like flavors, right? We haven't even done fast requests or priority or whatever. You don't even really need it to be honest. You're just rolling right now. Um that's kind of where we're at right now. Yo, Yan, thank you so much for becoming a member, man. I really appreciate it. Uh you're going to be around some really awesome people and I'm really excited to see you in the Discord. Uh, is Ray using a template for his current project? If so, which one? Yes, the the project template is I think it's rfer.mme/starter. And this is the starter kit from uh this guy whose name is Arkim. And so he's actually just in the Discord. We're going to do a stream this weekend. He made this really great Nex.js starter kit which basically has Convex, has Cler, has Stripe all integrated into it. And you just basically have a SAS in the box in 10 minutes, bro. Like in 10 minutes, you can start charging, putting payments behind it, doing all this cool stuff. And so the configuration is here. And I'm making a dedicated video for this as well that it'll like literally take you through like the 10-minute setup of like add this file, do this, that type of thing. And so that'll be coming out, but I have a live stream as well like walking through it in a little more detail uh for like members only. So you'll be able to kind of check that as well. And let me go ahead and just drop that in the comments so y'all can get this. Um, so that's the Yeah, I put that in the link there. Uh, Kais Daisy, so you can get that. Um, yeah, a little model just has more context window for my use. I'm not limited by cursor. Yeah. Yo, awesome. Uh, would you say looking into Python ML is good or is it much to do? You know, if you're still new to coding, just pick something, right? Like Python and get really good at it because that translates to so many other things in the future like other languages. If you're so inclined, one of the interesting techniques for learning is to actually learn two models at or two languages at the same time because you'll start to sort of cross-pollinate on some of the concepts. It's maybe that's the way my ADHD type of brain works, but like if I'm learning an instrument, I want to learn it like three different instruments at the same time because I I tend to learn a lot faster that way. Um I get to see this the same like concept of music but in all these different ways. And like you know if you learn drumming and then you learn like something that's like not like actual like a like a flute instrument and then you play like a piano like those are three majorly different perspectives but they help you like one keeps you in timing and then like one keeps you in like you know understanding chords and stuff like that and like the visual concepts right cuz you're taking in different inputs and same thing with programming languages they're actually different ways of thinking and structuring data and how you know where they touch at the layer of the OS is going to teach you a lot too. So like those perspectives are very rich and there's a lot of information out there. So um yeah, I would just say definitely experiment and play around because like I feel like if you've learned React enough then you can kind of pop into Swift and you're like okay I'm going to manage UI now instead of dealing with client server you're kind of dealing with more your client right client and UI state and you know passing data off between a UI layer and maybe something that's happening at the at the data layer, right? So um it doesn't hurt but take it at your own pace or whatever you want. I feel like with these AI tools too, they can help you learn pretty fast. So yeah. Uh what's what means tool calling on how cursor can other tools? Windsurf. Windserf. I forgot about windsurf bro. I forgot about dre. Oh I haven't. Uh so windsurf if they have the same tool calling it's just going to be a difference in how they implement it into their IDE. And you know you you can have the same model in the same in like two different IDEs. They're VS code forks. They don't behave the same. Like the cursor team has been doing a lot of work to hook into those different parts of the IDE to grab either context or to do like specific coding tasks and then you know the tab model like there's just like too much sauce, right? So I mean to in to my advantage I think like cursor has already cooked everyone. So and they're going to cook more people with this type of thing. So let's see what's going on here. So it has the issues for the mocking. It falls back to Okay. So I've seated Oh, it already did the seating for it. Let's just see. Do we have issues here? Let's see. Refresh. Retry. Issue review. dashboard. Go back. No. Issue review. Refresh. Retry. I may I may have to how to test. Go to issue review and see if if it returns empty, it will autoload 25 mock items. Okay, cool. Let's just go to dashboard. Oh, let me rebuild it too. O issues convex dev and then you can recede the MCP via MCP later or add more. Okay. You want me to add a toggle to the UI to switch between live and mock data? I think I might do that. I'm just going to keep all these. So, let me go back here. Refresh. Okay. Uh, okay. Okay, so right now I just see error fail to load issues and then that's it. Okay, I think we may have to add a toggle to load the mock data because it tries and then it just goes to show me this. Okay. And then we just do that. So, I think that's what we'll have to do is add some UI because basically it's technically working, but I haven't put in all of the tokens for GitHub to go scrape my issues. I haven't told it what specific repo to pull it from. Um, we're basically just sort of making this really dirty prototype, and I'm just trying to get some mock data to to come in so that we can actually start to interact with the swipeable component and then we can kind of iterate more on the UI. I like iterating on the UI and experience because that's kind of the feel part of things. So I guess Yan, another advice I would give to you is that something I said in a previous live stream, which is we are currently in our craftsman era. And so previously, like you may have seen a tweet from Sam Walman who says we're in our fast fashion era. Your boy is on a different, you know, level or something like that. That that dude's wearing his crazy fashion and driving fancy cars. But I don't know about you, but I'm not a billionaire. So I'm just a normal dude just like you, right? I've worked at this fruit company for a while and I got kicked out for basically getting sick for a couple of years and now like on my own like do I have to go back to work at a company or can I just go work for myself and then use AI to generate freedom. So that's kind of where I'm at right now, right? It's like I can build my apps any ideas I have I can make them come to life. So this is kind of what I'm realizing is that we are in our craftsman era. So what that means is that you can start to learn how to do something and become really good at it. And the things that humans can do that AIs can't do is have feelings and emotions. And so you can do that with the work and level of detail that you do. So an example is just a person who's a woodworker and a craftsman. If you build really good like cabinets, your your work is going to stand out way beyond everyone else. And so whenever you do something good, have you ever been somewhere and then at a dinner table these people will tell you, "Hey man, have you heard of this new like chat GPT?" Like if five people are telling you about it without being paid or unprompted, it means something so good, right? So that's the type of level of good that you can get by solving like one simple problem and doing it extremely well at a level of detail that only you understand because you really have been in that problem and you solved it and then you can keep building and building on top of that. So you basically just get this asset that you make and you keep making it better and keep making it better and refining it over and over again. Sure, it's not going to generate you a billion dollars tomorrow, but in this era that we're in, these companies grow really fast, right? They were teased for making these, you know, like somebody was like, "Oh, I have something that generates code." And you're like, "Oh, you mean you're just a GBT rapper?" It's like, "Yeah, a GBT rapper that's basically worth a billion dollars now, bro. Like, what's up with that?" Right? Like, don't listen to other people. You have to go build. Be curious. Be getting your craftsman bag. Start learning those tools. Start solving those problems and cook, bro. Like, you're with us. We have to do it as a community. We have to show up together. Like, y'all on the stream, use GPT5 as much as you possibly can. Share some tips. Share back what you're learning. Right? I'm showing through some of the planning stuff I'm going through. I'm showing you how like you can mock data to play with UI. Like, this is the stuff you want to do to like really understand how your apps load. Look at that. Look at Look at that. I'm swiping left. I'm swiping right. Let's go. Let's go. Let's go. Let's go. Oh my god. Family. Let's go. Let's go. Swipe left. Swipe right. Swipe left. Request changes. Request changes. Approve. Open on GitHub. Yo, this is crazy, bro. No way. Use. Hey, dude. This is crazy. What? What are we getting into, bro? Sheesh. Let's go. Load the mock data. Oh my god. I'm like swiping left, swiping right. She Oh my god, bro. This is fun. This is what? Let's go dark mode. Let's go back to the phone. Let me refresh. Let me show you the mock data. Come on. Come on. Swipe left. Swipe right. Swipe left. Approve. Request change. Approve. Approve. Like I want to have like the card actually slide and animate. I could probably tell it to do that. We're going to Let's look at the D. Like that's crazy. Let's see. Dashboard.com concomx dashboard. I'm just curious to see what the database looks like right now. It's probably a mess. So, this database is called three commas because we're trying to hit that. We're in our billiond dollar SAS era. Oh, man. All right. All right. Swipe. We have swipe sessions. We have swipes. We have payment attempts. Mock issue. GitHub settings. Mock issues. So, these are all the mock issues that show up in the database. It shows you who the author is and who this type of thing is. Damn, that's what's up. That is what's up. So, PR114 example title, right? So, like I don't even know how these things load. And this has like an ID. Cool. HTML URL is pull request true false number 124. Repository Open AI titles PR example YOLO 124. Let's go. Yeah. Let me refresh. Load mock data. YOLO. So, yeah. And then I think watch I'll show you something pretty cool. So this is the reactive nature of convex. And I think this is kind of like a really cool demo I could show you. Um, so I'm literally changing the database right now. Oh my god. And theoretically, if we're using the right API calls, this should actually update. And that's how I know it's not actually being reactive because this should be using the the query from uh React from Convex. And that should be pulling this back in from here instead of just like reading it and kind of keeping it on the device. So yeah, we have we have some bugs to fix. It's possible that the issue can be duplicated. No, it's not. Okay. Yeah. Yep. So we have some more bugs to fix, but at least we got to kind of test it out and see what's happening. So yeah, more on that in a little bit. But damn. So wow, this is really funny. This is really crazy uh for us to be here. So uh let me catch up on the comments. Y'all have been like bearing with me throughout this whole entire stream. Also, shout outs to everyone who's on YouTube right now. We have 414 people who are watching the stream. Absolute beast mode right now. 1,572 people on X watching. Y'all are killing it right now. Also, if you aren't aware, uh I'm also trying to get my Twitch streaming up. So, if you're on Twitch, you can hit that up. Uh that's Ray Fernando 1337. Let me see if I can get a link, copy video URL, and put this into the chat because um I also need some Twitch love, too. Cuz apparently like they can get you can get monetized or something like that if you have enough people watching. And I think there's only like one one person there just kind of chilling on Twitch. But I mean, y'all y'all are coming from different places and I definitely appreciate the love. This this is so fun though. Like look at this type of UI that we got here. This is what I live for though, you know, like like dark mode, light mode, dark mode. Let's go. Let's go. Let me let me get this other change in. I kind of want this. I want this like nice reactive UI. Okay, so this is what we did. We basically made a new chat. We duplicated it. We said I want to um add a swipeable type of thing and then the next thing I want to do is just log all the changes into my documentation so that way whenever the next agent picks up it can kind of take it on from there and understand all the work that we did cuz we did quite a bit of work here for this type of plan and I want to make sure that they're documented so that's easy to hand off for a couple of reasons. If coding is not your everyday thing and you're kind of doing this as a side project, a whole week can go by and by the time you get back to it, you're trying to spin up your mind. You're like, I just lost what I was even doing, right? Was I working on the UI yet? Is it even hooked up? What what is happening here? So, doing this is going to do a couple of things. One is it helps your other AI background agents to get up to speed on where you're at. They can take those pieces of things and start working on these little mini tasks. That's the first thing. The second thing is for your own sanity. You want to know what has been done and what hasn't been done yet and you want somebody else to review it with a lot of detail. So then this is where I maybe select GBT5 high and to to because you want to spend more time in your code reviewing things, see what still needs to be done because there could be a bunch of other things that aren't done yet. And just like anything, you want to trust but verify. These models can like actually convince you that it did work when it actually didn't do work. And with the latest clawed Sonnet 4.1 model, that model is actually very deceiving. It's actually mentioned in the model card. It can actually convince you that it's it basically for me it feels like it's gaslighting you. But everyone will have different experiences. That's why I'd say trust but verify. So this is kind of the I would you know it's easy to go fast. To me it's literally like I got the blindfold on. I'm driving the Lambo down the Golden Gate Bridge. I could fall off at any moment if the AI steers me the wrong way and it's like this is the right way right in the ocean. You know what I mean? Yeah. Yeah. Cloud code is a legendary gas light here. For sure, for sure. For sure. For sure.

All right. So, let's get to the documentation part because we still got a lot of context window left. 19%. This is beautiful. This is so beautiful. Okay. So, let's check in these changes since they work, right? And I'm going to make sure I get this all checked up here. Um, and so what I did to check in the changes, I went down here. I just said, you know, like the changes will be here. I hit plus. It stages all the changes. Uh, and then once the stages are changed, I hit this little star thing. It automatically made our commit thing. And so now we got two of them in there. Let's go ahead sync it up. And so now that it's synced up, we're all up in here. And yeah, theoretically, this is live. I mean, let's go ahead and see. Uh, if I refresh, I think it's in Verscell. Let me just double check because that's what I'm deploying. Three commas and let me go to my deployments right now. It's building the the latest thing, but our our previous demo should technically be here. It's in a preview branch. I haven't merged it to main, but the cool thing about a preview branch is I can actually see what's up. So, I can log in and then um once I log in, I can kind of like, you know, theoretically show you what what that looks like. So, I'm just going to do that right now while we log in. So, while while it's logging in here, we're going to go back. And so yeah, this

is actually deployed live on my Verscell thing. I go to issue review, and then we should have the issues here. And, um, the latest changes that we were talking about with that has like the mock data will actually be deployed in my latest build. So if I go back here, it's still building. So probably like 40 minutes. And if if I want to, I technically can merge this to master, and you could just go to like three commas.burcellapp or something, and then you could play around with my little mock thing. But, um, yeah, that's kind of fun, actually.

Uh, so yeah, that that the the joy of deploying and doing this type of feature branching is the fact that like you have the developer experience built in. So if you do slash feature, it's not going to go to my main branch. So if you're ever working on a website, you always want to do this type of thing where you create the feature branch and you push it up. And then with Versell, it automatically gives you a preview of whatever you're working on, which is why we're checking in at every single step of the way because everything gets built. You get to see it. If there's a failure, make sure you fix that. Then you can kind of play with it live, right? And then two, when you merge it back to main, that's your main website. That's what I consider production. So, you don't want to mess with that. You want to make sure that everything's kind of tested and working well.

Cursor has a feature called Bugbot. So, whenever you're doing that pull request into main, it's going to review all of your changes to make sure that like you're not really going to blow anything up. And Bugbot currently costs, I think it's like 40 bucks a month. But the level of review and details is actually really good. It's not just going to tell you about like, oh, you had this linting error and like going to fix that. It's actually going to fix real bugs. Like I've had it fix some really bad state issue bugs, and I've actually merged the changes without doing Bugbot review and then later on I see the review and it's like, "Hey, Bugbot told me about this bug," and I had to like go back and roll back my changes in prod. But I was just moving too fast, and that's just my personal experience. I've had really good experiences with it. It's worth 40 bucks just to have that sidekick.

Um, I know if you're like in college trying to pay all this money, it's like really hard to do. But as a as a software engineer who's worked in the industry, you're like, "Bro, 40 bucks a month? That's easy. That's easy to have a second pair of eyes that can do that level of detail." I'm really curious if they're going to use like they have their own model that's trained to do this. But I think with these newer models, I mean, these guys are just they're cooking so hard, bro. They're so Let's go. Let's go. Oh my god. Cursor is back. Like BACC back.

All right. So I should have the latest branch here, and let me go to my preview. And I go to this URL now, I should have my So this is live in production. I have to log in again. So, um, I log in, and it's going to pop in the little login thing. And so that should basically let me log in. I have like 50 Gmail accounts, cuz every account that you create for, um, your business, you know, your whatever website you have is has its own email account. So that's kind of why I'm having to log in and out here. So we go to issue review. Now I should see the mock data now in production. Um, this is like production. It's like the preview preview, uh, actual thing. So, yeah, this is us swiping. And so I can actually test this out if I put this on my phone theoretically. And I should be able to go to that specific URL on my phone and kind of like cook there. So let me just make sure I could do that cuz I want to test to see like, is this actually capable to do that, um, from, you know, let's see, from a phone. Let's see.

Okay, let's just do that. It's going to paste paste and go. Okay. Authenticating as my Verscell thing. And so now it's actually asking me to sign in, which is pretty cool. So it's actually like doing that type of thing and say, "Hey, sign in through three commas." So I'm going to go ahead and sign in, which is be kind of fun. And, uh, we're we're going to be doing that there. So this is cool. I don't know. I I love live streaming, by the way. This is really crazy.

And so, yeah, what do we have here? Okay, bunch of people who are kind of catching up with us. Okay, cool. All right. So, as of now on my phone, I actually have that UI now on my phone, like actually on the website. So, it's not like localhost or anything. It's actually deployed to Verscell. And then what I can do is actually load the mock data for it. So, what I could do there is just say load mock data. This is kind of hard to watch it on the mirror. Oops. And then I can swipe left or swipe right to do the issues, which are pretty cool, right? Like literally swipe. I'm swiping left and right. It's Yeah. So, it doesn't show any animation, which is kind of the next feature I wanted to build in. But I want to make sure I document these types of things because as of now, technically React Swipeable as a framework is working underneath the hood. And the next step that we need to do is just add in the, um, animations to show like as I put my hand down, it should be able to track that and then animate it the card moving, just like you would normally do on the phone.

Whenever you're working with web, it's a little bit different, right? On on iOS, it seems to be like there's a lot of other frameworks that are easier to animate and draw, and it just looks way smoother. But on the web, it's a lot of a different stack, and that's kind of what I'm still learning. And, you know, I'm not a web developer, but I'm learning to be one. Uh, but leveraging AI has been really fun because I get to go the fast route, but I have the ideas, right? So like they're coming out a lot faster, which is really cool, and this is like the era we're in. This is so beautiful, man. That's so cool.

All right. All right. So let's go ahead and do that. So now that we have stuff checked in, it's already in code, it's already working, it's already in, uh, on my preview server. I have a documentation folder, and that's where I'm going to add in what's happened, basically. So, um, uh, so far, I have implemented the Goated Master Plan and have created the mock data, and everything is now working in my preview environments. So I want you to go through this entire chat and take a look at all the changes that we've checked in so far in this feature branch and describe them in a markdown file in our documentation so we can update ourselves on the progress on the master plan.

So what it did is it basically did this here, and, uh, my whisper flows lowkey failed me. So because maybe it was like looking for it, it thought go to, but I was looking for goated, go a see. So let's see. I'm going to just do this at goated master plan, okay, Goated Master Plan feature. Okay. And so then what I want to do is then tell it to write it into my documentation folder, which I already did there, um, marked in a in a new markdown file, yeah, um, in our documentation so we update seller still. Okay, cool.

So, basically what I'm telling it to do is go ahead and generate this new markdown file. Um, and then please put today's date in it at the beginning with something like 2025 08 August and then 07. Okay. And then we're still using the high mode, and then we're going to send that off. So this type of step for me is really important because I can track like for me, it's just my own sanity, right? Like a lot of stuff has happened sometimes, and I just can't really keep track of it. And this is my preferred way of doing things. It's kind of by hand. It's a little old school, but you have to learn this type of practice, right? Like you work on stuff, and you know, you would typically go to a meeting and give a status update, but if you're working by yourself, you don't have to do that anymore.

And I can't tell you how many times I've like slipped on not doing this type of thing, or I've slipped on certain processes like not tagging my code in terms of like labeling what's in production and whatnot. And those little hiccups do bite me. Like it literally bit me like last week when I shipped my app and I had to roll back a month's worth of changes to track a regression on like users could like once they logged in, there was all this state that was stuck everywhere. So like they could not process a new transcription because the UI was stuck. And I'm like, "Oh my god, I did not hit this during my testing. How how did this show up?" And so I had to like roll back. I was like, "Okay, this is the exact date where I started actually modifying this code, and I needed to work immediately because like three people hit me up in 24 hours. Like I need to act fast." So, uh, I did a live stream about it, and it's like you can take a look at that live stream, and it basically teaches you how to do a Git rollback so that you can roll back to that specific thing. And that's where I learned the best practice of like, I I've already knew the best practice. I just was being kind of low-key lazy. I was just vibing so hard that I didn't I didn't really think that it would be able to bite me that much.

So, um, there's a reason why these things exist as far as processes. And they do help you actually when you're in production and you're actually collecting money from users. You know, I don't have like a lot of users, you know. I have under $200 that of monthly recurring income for my app. So, it's like it's not like thousands and thousands of dollars, but hey, it's something.

So now we have our documentation. Let's just go ahead and review what we've been doing here because I think this is going to just be important for all of us. Um, because we were able to make this really cool like fun concept and take a look at the model itself. So, um, what it did is implemented the M0 slice of the plan. So like that's a really cool milestone. So it's it broke up my milestones of the plan, uh, with the swipe logging, idempotent GitHub review posting, and the reliable mock data flow. And here are the actual major changes that it did, which are pretty awesome: the Convex function for the swipes, the Next.js API routes, UI updates, and then like the decision mapping and stuff, and then the behavior and reliability here as well. That's pretty cool. And then environments and seeding is also set up here. And then how to verify, use the mock toggle notes. And then the next steps are the M1 targets. So that's actually wiring the GitHub settings filters to apply them to the issues for the live data, uh, and add a history panel for like swipe/recent. So that's a pretty good like nice review of our milestones. And we we've barely touched like phase one was actually phase zero of the master plan, right?

So it's not going to let me check in the changes until I actually write something up here. So that's why I need to do this up here. And then I'm going to make sure I check in these changes. And let me just go and close this here. And I don't know what happened. Uh, did I change something? There's a bug. Review the next file. Oh, I have to close this here and commit and sync these changes. Okay, so now that I have these changes, I I'm kind of like set and ready to do like the next steps of things that I want to do here. And I still have a bunch of other stuff that I want to take a look at. Like we were doing the the plan as well in Claude Code, and we were trying to figure out like, is it is Claude Code good enough to actually, um, like do the same thing as we were doing with ChatGBT5? Is the experience going to be better? Is it worth canceling if you've already invested 200 bucks a month? You know, I I I

So, as far as like plan for plan, you know, specific to specifics, I hate to say this, but it's really just going to be on where do you sit in this stack and how like how much do you do agentic coding and so forth, right? I just signed up for the ultra plan like right before the stream because I know I'm going to be using this model a lot as my daily driver because I just like to talk to something that's intelligent. It knows about a lot of things. It takes the time to think through things. Um, even just as a medium reasoning, I don't have to think about it too much. It's so good. Like two, like the user experience of using Cursor, right? Like Cursor as an IDE is something that I I I still used in Cloud Code, right? I literally installed a Cloud Code plugin so that Cloud Code can be inside of Cursor, and then as it was doing code reviews, it would pop into Cursor to do the review. And so like it's not like I didn't use Cursor, I use Cursor all the time with Cloud Code, and it's so like for me it's like, okay, cancel Cloud Code Max, go all in on Cursor with the latest model, see what happens. If it blows up, then I'll just resign up for Cloud Max and then maybe balance the two, right? Maybe I'll do some stuff over here with Cloud Code Max that I like. Right now, I'm I'm preferring the outputs. I'd rather pay per use with the GBT5 because of its level of like searching through the codebase and doing this type of understanding where, you know, Cloud Code is still it's I mean, I don't know like, but if you've already spent like a month and a half investing on everything Cloud Code, it may not be a good switch for you, right? You might be already pretty chill in Cloud Code, you're already used to sub-agent prompting, but I would not like ignore GPT5, and I would not ignore Claw, uh, what Cursor is doing. And when I say don't ignore them, I say that this is going to be a new paradigm for prompting and for agentic coding that is a little bit different than the trajectory that Cloud Code is currently going down. Cloud Code could possibly adjust to this. But as of now, because of all those hooks that Cursor has into the IDE, into like the your code itself, it is it is super impressive for what it's able to do and understand, um, the task that it's trying to, you know, set up for you. Like we took a very vague prompt of like my spew of ideas, and it's generated like the full-on PRD, and then from there, like actually acting on it, and then giving us updates and these multi-step things. Um, super duper cool.

So here we are. So here's Claude's plan. So Claude's talking now about the the rate limiter limiter from Convex. It's talking about the request GitHub data, and then it's talking about like the token budget because I didn't realize there's a token budget. So today I learned proper handling query hooks for fetching GitHub data in real time. So yeah, I mean, you see that there's like differences between the two models. What would be really fun is if we actually had one plan from GPT5, hand it over to Opus, right? Have Opus take a look at it and do some research, and then come back to GPT5. I think the two in the conversation will be super fire, and that's something that you can do in Cursor too, right? I can just open up a new chat window, switch it down to Opus, and say, "Here's a plan. Can you go ahead and review it and just give some feedback?" And then you could do you could do this cool loop, you know? I think that's to me, I think that's going to be actually where a lot of value is.

So, um, yo yo Pavo, thank you so much for becoming a member as well. Y'all are cooking here today. That's what's up, man. You guys are definitely, uh, sliding through with with the vibes and stuff. Yeah. So, if you become a member, basically Andy, as well, thank you so much for becoming a member. Dang, so many member requests today. So, this is great. Um, you'll get access to the Discord, you know, basically before I double the price. Get to hang out with other people who are thinking and cooking with us as well. Uh, how do I join your screen? Non, uh, are you on X right now? Are you still on my, uh, thing? I want you to Yeah, I'll get the headphones on because I'll get you in on my, uh, stream. I would I would love to see what you're doing cuz you So Niston is awesome, man. This guy's like the OG of like vibe coding, but like not only vibe coding, but like really knows a lot of stuff about model training, and they are part of like him, uh, and you have Alex Vulov who run, um, and a couple of other folks, I think Wolfrem and, um, some of the other regulars who run this thing called Thursday. So every Thursday they do their AI news, and they just, that's how I stay up to date on all those latest like open source models, what's happening in everything like from image models to like, you know, all that cool stuff.

So, uh, let me let me just pull up my X privately real quick and see if I can kind of see if I can get you over and see if I can get a stream link over to you. Um, yeah, and that way it would be really fun because I've been waiting wanting to see if I can cook with some folks. Yo, let me cook. Okay. So, yo, you there? And then I'll get the headphones on as well. Uh, let's see who else is up in the chat right now. Uh, let's see. Does anyone have any experience in Cursor and Windsurf where it reaches a point where it gets stuck and loading something, completing a tool call, or fix it? Do you have to reload the IDE? It can. You may have to just re like quit and then restart, or Yeah, cuz it like there could be a couple things. It could be a timeout from a tool call. Uh, it could just be the IDE stuck right in a in a process, and it's just not able to get out of it. Uh, or opening up a new chat and then trying to the same thing again. Um, hopefully it doesn't happen too often though. But that's what's up. Yeah.

Uh, let's see. These live streams are fire. Man, this is so cool. I I'm surprised how many people like show up. Like we we got like 400 something people in in on YouTube, and it's like 396 now. We have like five people on Twitch, you know. Oh, it's so so awesome. Like 1,700 people on X, that's just such a fire stream right now. So yeah, uh, the line using AI to build freedom is fire, dude. That's where we're at right now. Like I don't know if you realize this, but corporations don't care about you as much as they pretend to do that. And I'm just tired of that stuff. You know what I mean? Like I feel with 2025, like this is the last half of the decade of 2020, right? Like 2020 burned down. That was a big reset. And it reset my life too, because like I was literally in a hospital. I was going to die. And like I was part of this corporation. I've saved in like I don't know half a billion dollars easily, you know, through all the different stuff I worked on. And they're just like, "Yeah, like you're sick. I hope you get better. Let us know when you get better, and we're happy to have you back." I'm like, "Bro, I thought we were family, bro. Like I gave so much, you know?" And that was that moment I realized that like I see the game that's being played here. I see like everyone's low-key out for themselves, but that's not the type of energy that I roll with. I come from a like culture and family of values of love and like taking care of your community. And I'm not saying that they don't take care of the community, but I'm just saying that that's just not the way I like to be treated. And so, I like to give back and give treat people with a lot of respect, especially if they worked for me for so long, right? That's just kind of how I roll. And like I realized at that moment that I can actually take stuff on on my own. And I learned a lot being there. As much as like, you know, it was a two-way street too, right? Like I learned while being there. I dropped out of school. I don't have a degree, right? I was able to have large impact because literally I showed up to work. I was curious. I talked to people. I learned how stuff works. And I just kept contributing at a level that like I just didn't care. I was just just what we're doing right now with AI, right? Playing with these tools, learning how they work. And then you start to have impact more than you really realize. And that's kind of what you do.

And so, uh, the AI generating freedom is kind of like like the bigger theme of what I'm trying to do because that's like I could see the intelligence in these models, and you can see where it's going just off the last two years, right? This is this is blowing up, and this is our time to be in it right now, right? Start prompting, start getting in the way, start asking questions, blow up your way of thinking every time there's a new model release. Don't assume anything, right? Treat it very differently coming in like as you're a freshman. You don't really know what's happening, and you're going to learn a lot, and you're going to learn new techniques, and I want you to share them back with the community. Right? That's kind of why I think we have to build as a community right now because if you do it by yourself, you're only going to have such a narrow perspective on this giant iceberg that is literally sitting there. We're just seeing that tip of it. That iceberg goes so massive, and like all of us alive can't even surround the iceberg. That's what's crazy, right? So, we're just sharing and shouting our perspectives and hearing other people who are using this AI systems that are much larger than us, and they're we're all kind of seeing similar things.

So, yeah, building with AI super fire. This is like the movement that we're in. That's why I I named this, you know, like the membership, let me cook, because I feel like so many people are misunderstood or not heard because now this is your moment to be understood and to be heard, and you have the power to leverage the these like icebergs of intelligence to bring them in to make cool products, to help your community, to do good things. And that's the that's the energy I roll with. I roll with like, we're going to do good things. People are going to get out of our way. Like, you know, I just learned a lot about Mr. Beast's like philanthropic stuff. Like I knew him as a content creator, but I didn't realize how deep this guy went like in caring for other people in the world at a level of detail that like so few people do. And I feel like that's our energy right now. This is the energy that I roll with. That's the energy that I'm on right now, too. And that's what we're trying to do with this stuff. You know what I mean?

So, yeah. I'm just blessed to be able to have a camera, to turn on the live stream, to have this equipment to be able to communicate with you and share my passion for what we're doing right now because this is just absolute fire, you know, all this stuff that we're doing. It's like it's kind of fun and games and all that stuff, but at the end of the day, I feel like, man, you we're going to be rubbing elbows with people who are doing some really cool stuff, and there probably awesome people who are doing some really good stuff and stuff. Reuben, we high out here. GPT5 high. Yo, what what model you do? Do you guys got five on it? Are you on GPT5 right now? Is it loading up in your chat GPT window yet? Drop the five if you if you got that. Drop the five if you got five on it. That's what's up. Yeah, that this is this has been like a really crazy fire stream. Oh, man. This is wild.

Um Oh, I got the monkey emoji. Oh, that's so cool. I think that's probably on, uh, that's like Oh, yeah. I think that's on, um, those are my folks on Twitch. That's what's up, man. Oh, Vzero has the UI thing, too. Yo, greetings from YouTube. Oh, man, I'm so far behind in the chat. Wow, I just got here. Does it have to take which which is better, GPT5 or Cloud Code? Um, man, it's kind of like a rough question cuz like I feel like GPT5 as a daily driver for sure. Cloud Code, I'm still trying to figure out right now. I mean, I I have Cloud Code figured out, but like there's there's a lot of stuff that's going on with Cloud Code.

Uh, okay. So, here's the plan. Here's everything. Would you like me to proceed with this implementation? No. Um, let's see. I want you to write the detailed plan in a markdown file in a documentation folder so that I can go ahead and review it. And I'm going to take it out of plan mode, and it's going to do auto accept edits on. And that's shift tab to cycle between the two. So this is the Cursor with UltraThink enabled, and sorry, this is Claude Code with UltraThink enabled doing that thing. So it's going to create this markdown file. Uh, this was it took some of the changes that we did to review the initial plan, and I'm just going to make sure that, you know, it writes out this big plan and so forth. Uh, cuz we're basically trying to compare how does GPT5 compare to, you know, what what was happening over in Cursor land with GPT5. You know, can we get this type of really cool mock UI swipeable thing that we did, uh, or not, you know, deployed to production, uh, like we did earlier? So, right now it says yes, it's going to go ahead and proceed. And you can see the differences, like, you know, obviously one is like command line, and one is not, you know, there's some that there too, but, um, yeah, we're kind of seeing that thing there too. Uh, for me, rule code is back. Oo, okay. Transparency about usage. Yes. Oh, Dan, not super impressed with GPT5. Yeah, I got to take a look at your tweet because like for those who don't know, Dan's really awesome.

Uh, visit link here. Yeah, Dan's a great content creator, and he has a a couple cool apps that he's been building, and he has a great live show too that he does with another builder. So, let's just go ahead and take a look at that, um, thread right now. Oh, Thread Boy fashion. Let's go. Dan is also like the ultimate thread boy. Let's see. So he says, "Across coding task, what's the best model? This will surprise you." So he says, "Making a basic landing page." So a simple product landing page inspired by Notion. Wow. Nailed it. Okay. And then, "Making a game, GPT5 left versus Opus right." So like a tank game. Both one-shotted. Both. Okay. GPT5 basically unplayable. Physics wrong. 04 Opus playable but difficult. Okay. Yo, Dan, did you actually build like did you refactor any of your apps or something with it? I'm just really curious. I've been having a different experience in inside of Cursor, by the way. So, I'm really curious. Is it just like a web thing that you're doing, or kind of how that works?

Uh, "Making a modern landing page using existing codebase." So, like this is similar to one but on top of page AI app with existing rules, component chats, and kind of like what we basically did with our tests. So, the results Opus 41 made some wild things, but okay. Uh, a bit light on content. Both use existing components and rules. I kind of dig both. See the videos. So, this is like exactly his prompts. This is like his, you know, fully fleshed out type of thing there. So, this is basically like a really quick version of what we did, and cuz he has he has a template for his type of thing. And so, it's kind of hard to see, but you can kind of see maybe a little bit there. So GPT5 versus they kind of look the same to be honest.

"Making a web app just from scratch." GPT5 left, the lightweight web app, mini time buddy. Yeah, I think this is cool for like mini web apps, but I don't know. I'm trying to build some really bad like really good stuff. So bad news. Not even close. I mean, there were signs. H yeah. Spend $20 on these demos. Okay.

So for me like I've had a little bit different experience, and I think the the whole thing is maybe because I'm using GPT5 literally inside of Cursor, and that has enabled me to do maybe a lot more than it would have if I'm just generating a simple HTML website. And when you're doing those types of generations, to me you're just leveraging what's inside of the training set for the model. But the true intelligence and kind of why I'm super hyped about it, which is probably why my take is very different from a lot of other people on Twitter. Um, right now people on Twitter are just literally just trying to cool do like really cool demos. But like in this live stream, we've been kind of getting into the little details, like these little micro details of what's happening. And it's very rich. And what we've been able to see here is like how the model is doing its thinking, how steerable it is, how good of a job the Cursor team has been able to take the model and implement it and hook it into all these little pits and pieces. It's like they remove a lot of friction, and Cloud Code can't do that same level of like hooks that GPT5 is doing inside of Cursor because, you know, it has to run its own bash commands. I'm not I'm not saying that like one is better than the other, but I'm just saying from that perspective, like it's only going to get crazier from here. So like even a month from now, people are going to be like, "Yo, I think I prefer Cursor over Cloud Code." You can try it out. I I feel like that's the whole point of this type of thing. You can kind of you're basically getting to see me see me do this thing, uh, in the way that I take a template and kind of build on top of that. And it has a pretty big design system. We we use the web inside of Cursor to go do that web research, right? We tried the medium model. We also tried the high model, right? The Snoop Dogg model, right? So that's a like the difference almost felt like I could just be medium all day. Medium chilling that back and forth. And there's like I think there's also another thing that I have here. It's a relationship with the model. So the I didn't feel like gaslit from GPT5. GPT5 felt like a really smart friend. It's like, "Oh yeah, I got you, bro. Like you just want to chill over there and like make something? Yeah, I got you. Don't worry about it. I'm like, yo, can you make me like the next billion dollar S? It doesn't like gaslight me and be like, you can't make that. You didn't go to school. It's like, yeah, I got you. Yeah, don't worry about it, man. Just just chill. Like, do you want to I could place an order for you for some food while like I just cook. Is that cool? It's like, bro, yo, let me dab you up real quick. Yo, you know what I mean? That's that's what I feel like. Uh, GPT5. GPT5 is finally that cool guy who's entered the room, and he's got like this the shades on. He's just so cool. And he takes off the shades and, hey, what's up, man? What's going on? And everyone's like, whoa. Like the aura. The aura is coming out. Like GPT5 has an aura that you just can't explain right now. Cursor has that aura too. Like the fact that they're together is like the power couple right now. GPT5 plus Cursor is that power couple with the aura, or they're just like, I don't know. That's that's that's what I'm feeling right now.

Claude Claude is like the like nerd glasses on, like, "Hi, I graduated from Harvard. I got a PhD at Stanford. I also went to Berkeley. Oh, and in my free time, I also like graduated from all these different places and I started a billion dollar SaaS." It's like, "Bro, have you like ever washed your hands like in a bathroom or something before?" It's like, "Nah." Uh, "Can you clean this toilet?" "No, I I don't do that. I don't know how to do that. Have you ever served French fries before?" "No, I've never lived real life. I've only lived this like privileged life before." Like that's the that's what I'm feeling from these two models. I feel like one's a privileged life, you know? And then over here, we're getting like the, "Hey, it's actually real. Like I can actually be real with it." That's just my That's my vibe. That's my vibe right now. That's That's what I'm rolling with.

So yeah, let me catch up with the chat. I'm so behind right now. Damage Inc. Cursor CLI's out. Yo. Okay, this is what I want to show you right now. This is such a baddie right now. Cursor. Uh, if you got the the link, let me let me just get the CLI right now. This is This is the This is what I'm trying to show you. This is why you're going to cancel that Cloud Max plan. Yo, damage coming through though. Damage coming through with that super chat. Cursor's out. I'm I've been waiting for this. I've been waiting for this. This is my moment. So, let me just go on X real quick. I I have so many notifications right now. I I can't. It's like 20. So, it's so hard to keep up. Okay, let's see. Um, let me just go to my notifications. It's probably going to show up in here. Let's see who's got the Cursor CLI thing. Dang. Vice versa. GBT5. Okay, cool. They banned. Okay, it's got to be up in here. Um, midwip meme. Is this it? Yeah. Cursor agent. Let's go. Got to get the star. Oh, yo, this is so hot. Oh, finally. Okay, so this this right here, ladies and gentlemen, is the showstopper. I think I think this the these boys are cooking, and like this is the the the one more thing out of the whole like out of today. GPT5, you know, Claude. Oh my god. I don't think we're ready for this, bro. I If you are you ready for this in the chat? Yo, drop the C5 if you're ready for this in the chat, because this is about to blow up. I don't I don't think y'all ready for this. I don't think you're ready for this right now.

"Cursor is now in your terminal in early beta. Access to all models. Move easily between your CLI and editor." So all you have to do is run this command, and then you're cooking. So, I don't know if you know this, but your boy has already had access to this. And this is the one I was really hyped about, and I didn't want to say anything cuz I didn't want to blow it up. But now that it's officially here, oh my goodness. All you have to do is open up a new window, type in this command, install this bad boy. It's going to download the whole installer thing. It's going to be like, "Okay, cool." So, for the next steps, all you have to do, so if you're using the bash, all you have to do is then, uh, copy this in. So, I'm using ZSH. And the way you can see that is at the very top of the terminal says ZSH. So, if you're running a Mac, if you're running the latest version of Mac OS 10, unless you are specifically using some type of tool, you're likely running ZSH. So what you want to go ahead and do is just go ahead and do this type of command. Just copy and paste this, and this will allow you to run, uh, Cursor currently in in the, uh, environment. So in order for you to start using this, you type in just cursor-agent, and then cursor-agent will do its own thing here. So if I I run this now, we're currently in the the CC the Cloud Code Tinder here. So like, you know, they may be kind of interrupting their own thing. So it says, "Welcome to Cursor Agent. Agent's CLI is in beta. Security safeguards are still evolving. It can read, modify, delete files, and execute shell commands as you approve. Use at your own risk." cursor.com/security. Thank you, ladies and gentlemen. So we can plan, we can search, and we can build anything. These are slash commands. We can check the different model out. So we can see the different models that they have. And we got GPD5 in the building. We got GPD5 in the building. Hello. And I don't know if you know this, but running the model here, I'm going to let you into some sauce real quick. One of the things why I'm super excited about this is the speed at which their CLI runs compared to Cloud Code. And we're going to get right into that right now because this is about to be fire. Man, this stream is just I might stream all day. I might not stream stop streaming, bro. I need to go to the bathroom. But like, let's go. Cursor CLI. Let's go. Let's get cooking, fam.

Okay, cool. So now we're in Cloud Code Tinder. We're still running this planned document. So I'm just going to see if it can actually like remember how I was telling you maybe GPT5 can actually review our plan. Um, we have a couple of things. I don't want to mix the models up, and I want to see if I can actually go to the other repo and see if it can actually maybe add that swipeable feature that we're talking about because in our current thing right now when we swipe it, it doesn't do the animation, but the swipes do work. So, I want to see if it can do that, and we'll see the power of Cloud Code CLI coming in. This is such perfect timing, bro. Yo yo yo yo yo. Ray's been cooking. It's a command. Yes, that's all you have to do is just run that. Uh, Mr. Frank coming through. Mr. Frank is MVP right now. All RP. Okay. RIP Cloud Code. RIP Windsor. RIP like your favorite IDE, bro. Meta, where you at, bro? Can someone check on Meta right now? The temperature is not is is so spicy right now. This is a This is probably like the hottest summer right now. This is really really hot right now. Oh my goodness. Oh my goodness. All right. All right. I got I got to lock in right now. Let's lock in. Let's Let's lock in. Let's lock in. Okay.

So, this is Cloud Code. It finished its plan. It finished its review. It has this documentation. And I'm just going to see, um, I'm going to put this aside from now. Like Cloud Code, you're going to be put to the side. You know, this is kind of where we're at right now with with Cloud Code. We're going to have to like, you know, set you aside right now. This is really tough because I was a big Claude Stan, and I I still am. I I still want to be a big Claude Stan. But literally, it's this meme right now. Like this meme is actually why I posted this yesterday because when I was playing around with it, I was like, "Y'all are not ready for this. Y'all are not ready for this." Claw code. Like he's still holding his girl's hand, but like Cursor, I will embrace thee. I am so sorry. Oh man. Let's go. Let's go. Oh, I I can't I I can't right now. I can't. This stream is so ridiculous. I mean, how many more features are we going to drop today? I don't think we're we're not ready for this. They're even popping it in my IDE. Like, they're watching my live stream right now. I swear they're watching this live. They're listening to my chats, bro. Bottle GPT5. Hey. Okay, so you can bring an agent to an IDE or remote machine, bro. Stop it right now. It's now available in the most intelligent model we've tested in Cursor. Users can try for free. Drug dealing 101, bro. Just give your people free product at the beginning, get them hooked, and we're in, bro. Oh my god. Let's try install the CLI. We already did that. Cool. Let's cook, fam. Let's cook. Let's cook. Let's cook. Let's cook.

Okay. So, we have the IDE in here. I'm just going to go back in down here. All right. So, I'm going to do this. Uh, I think cursor-agent, right? So, we're we're in cursor agent. So, now I'm in my three comma app. So, we're trying to make a billion dollar. This is our goal to use AI to generate freedom. We're going to make our billion-dollar SaaS. And this billion dollar SaaS is basically going to be Tinder but for code reviews, right? So, T, you can do a Tinder code review type of style. Code reviews are no longer boring when they have this really addictive UI to make you swiping left, right? The feature we're going to add with the CLI now is just going to be swiping left or swiping right. We have that underlying framework working as far as the UI bits, but now we got to make the UI a little bit more saucy, right? We got to make it more craftsmanlike. We're in our craftsman era. I don't get down with that fast fashion stuff, but I get down with craftsman level stuff, right? That's what I get down with right now.

All right, so let's just see what we can do right here. I'm just going to let it rip. So, we have GPT5 already set as a default. I'm going to go back to my other terminal and kill that one that we have here because I don't need it running. And, um, I have all the changes checked in. We're in our feature branch right now down below. So, we're in our feature branch. We have the stuff checked in. We had this little thing that we're working with. I'm going to close that up. And all we do is all we have we have the code up here. And so, I'm going to let it appear. So, let's just say, um, so far, I have been able to get the swipes to work and in the UI when I'm reviewing issues, but what I want to do is take it a little bit further. I want the user experience to be really rich. So, as the user starts to swipe, I want it to actually show the card moving similar to like it does in the real Tinder app. And I want the UI to be polished for this type of interaction. So I want you to think hard and look at the existing code in my repo and leverage some of those existing hooks and patterns so that we can create this really nice rich UI experience.

Okay, so this is like my request to GPT5 using the new Cursor agent. And let's just let it rip. I'm I'm curious. I'm really curious. Let's go. So we're going to see how fast this thing moves through our code. And look at the tokens that it's just eating up. So, one of the things that I wanted to talk to you about on why this is

Really special compared to cloud code is a couple of things. Cursor has their own hooks and their own API calls to their own app. So, they can do stuff at a faster speed than what Claude code can do.

You're like, "Ray, what does that really mean?" So, in order for cloud code to actually work with your code, it's actually literally reading the files and has to do these commands to try to like get pieces of your file out to build an understanding of it. Part of the magic sauce that Cursor has doing, and especially with this latest update, is that they have been building this type of like actual sauce about your code in a whole different way with different databases and different techniques and different file system techniques. You know, even like file-level like hashing all the way up and like that type of data like relationships at a level of things that you and I can't think of, but that's their full-time job. Like that's what they've been doing for the last like couple of years, less than more than 24 months, right? This has been something they, each of these people, have been obsessing about at a level of detail that is what I call cracked engineering. And they're able to turn around, like once that is already set up and working, like we're seeing, they're able to do move way faster with a really intelligent model that is very steerable with tool calling and everything in a CLI. And this is smoking, this is going to smoke Claude code, bro, like clock rip, bro. I know, I know, I'm being so hype. I'm a hype boy right now, but for a reason because I see this. It may not be like, "Yo, Ray, I just did that and it didn't make me like the next McDonald's, you know, or something like that." Bro, let me cook for a second. Let me cook before you go off on whatever your rant is. Let me cook. Let me cook.

Right now, what we're seeing is the beginnings of the next steamroller for code, bro. Like, these guys, Microsoft, I don't know where you're at, bro. Like, y'all have engineers. Like, show up to the game. Like, the level of craftsman and detail that these guys are doing with AI, bro, is at that level with cloud code too, right? Like, these guys are up here, right? Cloud code, Cursor, the Cursor CLI. Like, I feel like just Cursor just took that next step, bro. Like, they're coming in with this heat. What? Oh my god.

So, look at all of the stuff that it's done so far, right? It searched through the codebase. It did some grep, similar to what Cloud Code was doing. It scanned relevant files in the background. It was doing these types of reads as well. It's like still running in the CLI, right? And I'm going to integrate the new draggable card feature. So, like, it already figured out all that context in, you know, using GPT-5 as a model. And it's like, "Okay, I just got to do this this type of thing to pull in this component and then run this command." So, I can just say, "Go ahead and reject," and say, "Um, reject." Hit enter. I'm going to say, "Reason for reject is like I use, uh, I think in this repo I'm using pnpm, pnpm, right? I'm using pnpm. I use bun to install." Let's see. Here we go. Oh my god, this is so crazy. I just built DoorDash, bro. I mean, yeah, y'all are too much right now. Let's see. Run. Uh, allow list run. Uh, yeah. Okay, cool.

The Cursor CLI did the whole sub-agent thing Claude Code has. I think so, but that's what we're going to find out right now. Right. So, like, I haven't done some really crazy, like, super crazy research task, but I'm pretty sure you should be able to. Is GPT-5 or GPT-5 chat? I think it's just GPT-5, the AI, um, API because they write their own like context and everything like that and they probably just send it up to the API as calls, but I'm not entirely sure, you know. So, uh, it's doing a bun run build now. Cool. Cool. So, I added a draggable animated Tinder-style card with real-time movement, rotation, contextual badges. Damn. Here's what I changed. New component, and I can click this thing, uh, to drag swipeable horizontally. Okay. And in here, what I have is basically my, my swipe UI, my Tinder card. So, it like literally just added all this. That's cool. Updated dashboard issue page. So, they replace the basic swipe handlers with draggable Tinder card. Okay. Wow. Epic. Okay. So, try it. Drag the app and review the page. Let's go. Let's see. So, theoretically, localhost, right? Is it over here? Yeah, this is the one. Let me actually run the server. So, in here we are. So, this is running this. And then let me just kind of restart the convex stuff. And then let me restart this too. Clear. Oops. Oh man, I killed the terminal. Bundev. Move this up here. Oops. Okay, so we have this. We have the Cursor agent CLI. And then we have my app now running. So, my app should be able to go here to the dashboard. And then we go to issue review. And then we should be able to use the mock data now. And then we should be able to swipe the bro. This is so funny. Bro, let me refresh. Load mock data. Bro, this is so wild. Okay. Okay. Okay. So, I'm dying right now. This is so wild. So, there's some more work to do, but like I like this stacked UI. This is hilarious. Like, this is so cool. I love the like, all they did was like obviously import a component and kind of know about that, but like even then, like, like this is so fun. So, the issue is, once I swipe the card, the previous card below it doesn't have this. So, theoretically, I should just be able to like drag this into here, right? Like, as an image. Okay. So, image. Once I swipe a card, it, the cards below it don't show any text anymore and I'm not able to interact with them. So, let me just go ahead and see if I can fix that code there. I'm just really curious. Yeah.

So, Claude is $75 per million and GPT-5 is $10. I was using around 3,000 per month with Cloud Code paying the max plan. Doing the max, I would get that at 400 bucks plus limits. Plus no limits. Yo, man, you crazy, bro. Like, you're you're actually three comma club and like AI tokens, bro. You're cooking. You're cooking, bro. Like, hopefully you got this amazing startup. Uh, GPT. Okay. Yeah, I think we're talking about that. I think Ray uses Super Whisper. I actually use one called, um, Whisper Flow. So, Wisp F L. Um, you can you can use my, I think it's probably easier to just do this W I P R, I think. Yeah. So, this is my referral code thing and I'll just put this in the chat so that way you all have this. Um, so, yeah. So, like, I, I basically put my little whisper thing. Yeah, Whisper Flow is awesome. Like, it's probably my, my favorite thing there. Um, in the new, uh, Whisper Flow, if you like give it the at thing, it's actually going to add the files for us too. So, that's pretty cool. Yeah. Um, all right, cool. So, it looks like it's trying to fix this, and it's basically an issue with the, the key. Yeah. So, it's referencing a different ID. That's why it's not pulling it. That makes sense. Yeah.

GPT-5 is only for lead devs. No, bro. GPT-5 is for everyone. That's why I'm so hyped about it right now because it's going to change the way that we prompt and I think people aren't going to realize that until like much later. You know what I mean? So, this is the new paradigm. I think this is the new era that we're in. Uh, GE is talking about 2015 MacBook Core i7 maxed out 250. Damn. CLI installed. Cursor is definitely sending real-time notifications to our apps. I got the same popup. Yo, that's crazy. That's awesome. That's so awesome. Uh, wait for Anthropic to get mad, bro. Yeah, let's see. Whenever you get that, like, those nerds mad, man, it's gonna be a whole different level here. So, let's gonna run bun run build. Let's go. Making sure that the changes look good. Uh, I have to do this little dance when everyone does like little thumbs up stuff. And this is what we call global warming, bro. Yes. Yo, Chris. Yeah, this is this is what we're kind of getting into right now. I'm so Wow, this is this is amazing. Okay, let's just take a look. I think it should be my stuff is live now. So, I'm going to rebuild it. And then, uh, so this should be fixed now. And then just making sure that I rebuild here. So, yeah, let's refresh. Potential spam. No thanks. All right. So, so far, load mock data. Oh, I see. It's kind of like a funny thing because it always looks like there are stacks of cards. And the other issue is that like the stack in the back doesn't animate forward. So, I think that's kind of why it produces this weird effect where it just kind of snaps in. So, I think what I'll say is kind of describe it that way because I want, I really want to nail down this UX where it feels very like, you know, just beautiful and stuff. So, um, I love this. Like, this is so funny. Request changes. Approve. I like how the approve is in like red too, by the way. This is so much fun. Okay. Okay.

So, this is a really great step forward. The next thing that I want to fix is that as soon as you swipe the card left or right, the card behind it doesn't animate up. It just kind of snaps into place. So, it just feels a little irregular. Can you help me fix that? All right. And then let me also check in these changes so far, cuz I just want to make sure that we capture some of those changes and if we have anything, we can always revert back, uh, that's any bad. So, like this is like my way of trying to get stuff building in real time. It's always good to check in your changes a lot, especially if you have something that works right away. And then it's easier to kind of flip back to a previous time because sometimes you can have like five chat windows open. Especially if you're going in the CLI, I think it's more important than if you're doing in the CLI than if you are actually in the chat window because when I'm using the chat with Cursor, it's easy to just go back and say, "Revert all changes," and it will basically revert to that snapshot. But here, it's just, you know, you just check the stuff in. If you like it, it works, cool. Let's just check it in. Just keep get in the habit of checking stuff in. So that way you can always roll back later. Um, just FYI. So, yeah. All right. Let's see what's going on. So, it's going to take a look at our code. It's generating.

So, what we're taking a look at now, ladies and gentlemen, is Cursor has released their new CLI tool. And just with one command, you can install the tool and it's currently running inside of our IDE and it seems to run extremely fast using ChatGPT. When I compare that to using like Sonnet or I use the Opus model with Cloud Code Max. Is it as intelligent? You know, it's kind of hard to say right now because I'm having to like actually iterate and add smaller features. But I think where it'll be really interesting is like if I actually go, uh, and continue the plan, which I might do for the next step, is like add the next plan, right? Um, so that that would be really interesting to see how it does in terms of performance for adding the next part of our plan. So, here we are basically reviewing our issue page and it's doing an animation. So, it's added added this type of thing for like a spring animation to bring it forward and theoretically I should be able to run this now. I'm just going to recompile the app and then refresh it here and we should be good, good, I think. Right. Load mock data. I'm going to get rid of the sidebar for now. Oh, I see. So, the card, like, there's one that loads up from the back, but I think the card in the back still stays the same size, which makes it look weird. So, I'm going to just give it this type of thing. And so maybe we can kind of blur out the back and then make that really interesting. So, I think, yeah, let's go ahead and run that command. So, what's happening here is that like it is taking our instructions and it's kind of doing it well. But what we need to do, I feel, is like maybe create a blur in the background or create like a depth effect because if you have like real cars, if you ever done photography, like kind of like what you're doing right now, right? So, that background's kind of blurred out and that gives a sense of depth of field. So, that's the type of feeling I want in my apps. And so you kind of have to describe it in that type of way and say, "I want some depth of field," because right now, and we can just give it that screenshot, it looks like the text is stacked up against each other, right? A hard line, a hard text, you can read both. And what happens with your eye in terms of design is it's going to want to read both, right? Like naturally your eye sees something, it's going to want to, you know, go to one place first, especially if there's color, they'll go there first and then your eye kind of follows like this, you know, pattern. And there's certain ways that your eyes follow patterns. You may have heard this term in photography called rules, rules of thirds. So, it's just how you actually structure information to make stuff look interesting. So, if I move myself to like this third of the screen, it looks way more interesting, but it's going to look even more interesting if I'm on this side, cuz now there's a little bit more like interest, right? You look at me first and then you look at the rest of my scene. and then the rest of the scene gives more space for your eyes to open up. So, like these types of theories are really important to understand, especially if you're making apps. Um, when you're doing photography, when you're doing art, you start to learn a little bit about them, but then this is how you can apply them into your app making because now we can just tell it instead of trying to figure it out in code. And that's what used to take me weeks to do. Like it used to take me weeks to figure out, I know what I want to do, but I don't know how to write the code for this. And that's kind of what like ripped my hair out. It's like, this is not that we're, we're, we're in a very different age now, uh, with the age of AI. So, yeah, that's what's up.

Uh, do something that, uh, that exceeds the GPT context window because in CC's, this is Cloud Code's kryptonite. So, yeah, I'm really curious because like I don't have access to the context window down below in this type of chat. So, I'd be really curious as well, like how far can I keep pushing this before it like thinks I should reset the context window and is it going to even do anything like that as well. So, um, yeah, that'll be interesting. I, I'm with you on that actually. Yeah.

Yo, shout outs to the 408. We have 48 people here on YouTube. Let's go. Almost 2K people on, uh, X and then we got like six people over on Twitch. Yo, that's what's happening, man. We're cooking with the stream. All right, let's go. Let's go ahead and do this now. So, I want to give those instructions because now we got our little mock. I want that like blur in the background, you know? I want that depth of field. And so, I have an image. Can I just do, like, at desktop image or something like that? So, yeah, I think that was really interesting because like I want to see if I can actually like take a look and grab something from my desktop, but I just literally just will just drag it from my desktop like this. Just dragged it in. I just say, um, there is a way that you can set so that whenever you take a screenshot, it goes into your clipboard. Um, so there you can just paste that in. But, um, for me, I'm kind of old school. I like to kind of drag it in. But it can be super handy if you're doing that a lot. Um, so the problem I see now is the fact that the card behind doesn't have any sense of depth of field. And so the text looks like it's there. And when you swipe the card away, it's still present. And while the new card animates in, that previous card in the background still stays the same shape. So, it doesn't feel like it's moving from the background forward like you would if you had a stack of real cards. So, I'm going to just describe that issue and hit enter there. So, I'm really curious. So, if this is kind of my rule here, if we don't get what we want in a couple different tries, I might just clear out the chat and start a new chat. The other thing I might do here is try to see if I can take a look and run GPT-5 high. So, if I run high, does it, does the longer thinking time and the extra tool calling maybe allow it to really understand what this problem is and kind of understand where we should be going?

So, there are a couple things I want to find out is like how does the distractor theory work here and does that have any problems? So, for those who aren't familiar, I'm going to kind of put you on some sauce right now. Uh, Chroma DB, Chroma research, Chroma context rot. So, they released a paper on context rot. I'm going to go ahead and put this in here. Yeah, I forgot to tell it to think hard. So, this this URL will take you there. And this is a research paper they did on context rot. And basically the consensus is that if you like these models are really good at what they call needle in a haystack. You say, "I want this specific thing," it can go find it amongst this really big context. Right? So, this is something you'll see, oh, if I have a million token context window, what they discovered in their prompting and everything is that there's these things called distractors. So, if it's very similar to what it should be looking for, the language model can get distracted in the sea of everything they should be doing. And this is even more prevalent, especially if you're coding because there's a lot of stuff that looks the same, especially if code is repeated all over the place. And what this means is that like if you have more than one distractor, the effectiveness of your context window gets nearly cut in half or more as far as accuracy. It gets so bad, like up to like 40% if you have like the more distractors. So, like this one says that if you have four distractors in your codebase with like a, you know, over 500,000 tokens, uh, the performance just gets keep, like the more input length that you have, like the worse it gets. So, like if you have like a 50% token window, you're pretty much almost like at a 40% effective, you know, efficiency rate. So, all those distractors are causing problems. But you can see up here, if you just say, "I only want you to do this one thing," look how good it is. It kind of stays pretty well, like up into the 90% range. So, if you have these small changes, it's really good at doing that. And so this is where you start to like have, if you introduce a distractor, like one distractor, and you introduce four, this is where the idea and concept of context rot comes in because once that starts to happen, now that it's in there always for every single request. Whenever you're doing a chat, anything that you've previously said is in that history and it gets sent back to the model. So, that stays in its context. And what that does is you start to introduce one distractor. And if it's not doing what you're doing, that distractor stays there. And then while it's looking for code to do something and it doesn't go the way you want it, you could introduce another distractor. So, now you have all these familiarities. You're still trying to achieve the same thing, but now you've thrown it away. And as that context window keeps growing because you're trying to solve that problem, it just keeps getting worse over time. And this is the study that was done by Chroma DB and we're seeing that that's exactly what happens here. And people were saying like, "I just want you to change this," you know, and this is the point where we have to like literally clear out the context window and come back fresh, maybe take some of the things that we learned previously and start again. So, that's kind of why I wanted to give you some sauce on that. That's a little bit of history there on context window. And this is kind of what we're going to try to do now, uh, to see if it can actually fix the issue for us if, you know, in this second try here.

So, it's talking about a React thing. It is talking about on drag progress and then it's basically building more UI state for this and so, yeah, let's go ahead and see if we can, it's going to run a build command and then down here I'm just going to quit it and restart it and then we're going to take a look here. Y'all cool? Let's go to dashboard. Okay, let's see. Issue review. Load mock data. Oh. So, ooh. Yo, this is pretty fire. Oh, wow. Look at that blur. That blur though. That blur though. Yo, I like this. Oh, come on. Look at that. That's hot, bro. That's so bouncy, man. That's that's that's nice. That's so nice. Let's take a look at this. Hey. Hey. Okay. Okay. Okay. So, we're we're getting closer. You know, I like every little step kind of counts here. I, I think it's really cool. So, the first thing I like so far is like it's done this cool blurring effect, right? So, and then when I when I slide it, do you see how it kind of animates up a little bit? And I let go. So, now we have a problem because the back card is stuck in a specific state, right? So, the, the state is not resetting for the back card where we should say, "Hey, we need to reset the state for that back card," and it's literally going to go into that component and just reset its state variables. So, it snaps back and then the data reloads fresh from the database basically. Um, but so far, like the, the interaction of whatever it's supposed to do is is super nice. Oh, see, there you go. It snapped back. It snapped itself back into state. I, I think I know what the problem now is. Like, you have to swipe it back and it clears its state out because now we went from approved, disapproved. This is nice. Wow. Let's see this this UI here in dark mode. That's hot. That's hot. I don't know about you, but this is this is this is the stuff I live for. See, it has more depth to it, right? It has that depth where like you're playing with a card and kind of slides right up. Oh, this is fun. This is this is so much fun. I love doing this stuff with stuff. Awesome stream today. Yeah. Thank you, Mr. Frank. Thank you so much for hanging out, man. This has been an insane day for us to play with GPT-5 and it it's just been so crazy as far as like what we've been able to do. Yeah. CodeX CLI is not working due to high demand. It just says, "We're currently experiencing high demand which may cause temporary errors." No. Yeah. Bokeh. Yeah, that's actually that's just a word that you can use the bokeh. If you have like the bokeh framing, you know, you have that type of thing, right? Oh, we got to be careful here. Cursor is on fire. Let Cursor cook. Dude, Cursor absolutely cooked. Joining the stream super late. What are your thoughts on GPT-5, bro? GPT-5 is my daily driver, bro. It's I don't know if you saw that meme, Alex, that I posted online. This is what I feel about GPT-5 right now. Bye, Cloud Code. Bye, bro. Absolutely cooked. And like this is what's been this is what happened. Like I literally set everyone up. I was like, "Y'all don't even know what's coming right now." Like these guys are cooking. And like we're just getting started right now. So, GPT-5 as a coding agent inside of Cursor is where I feel like this model intelligence is unlocked. And Alex, the reason is because of the tool calling and the thinking ability. So, there's some thinking modes that you want to get familiar with. And so it's going to be high and medium. So, the default GPT-5 inside of Cursor is going to be medium thinking. That's more than enough for your daily tasks. I just enable max mode. You get 272 token context window. We barely even get close to that because Cursor is extremely good at grabbing all the context from our code files. And then two, it's leveraging the model's intelligence to to think and then think again and then really truly do some planning. We were able to take a template and then give it some ideas and information. It understood what the template was about. It grabbed all the stuff. We were able to generate this nice detailed plan and then have that generated as well and then start working on that plan. It generated the first milestone, which is milestone zero. Got us up off the floor. Now we're working on the UI. It's been a really enjoyable experience. Um, as far as like, you know, Cloud Code, it's going to be a very different experience, right? Very command-line driven. If you have a lot of processes and things already started with Cloud Code, you know, that is what it is. But I do find that Claude Code, like the context window, I have to pay more attention to it where with GPT-5, I don't have to. And like the long story short, GPT-5 and Cursor are like that power couple that have so much aura, but they're cool. Like they feel like they're with me, you know? They feel like they've actually had a real job in the real world and know how to solve your real-world problems. Where Claude Code is coming in like a little too proper right now. It's like, "I've never worked my day in my life. I just know how to do these cool things." It's like, "Yeah, you're very helpful, but like, let's get real with it, you know?" So, that's kind of that's the word. That's what I'm feeling with it right now. Uh, Alex, if if you're asking for Yep. Also, thank you for the super chat. Joining the stream. Uh, how much did it change the game? Uh, the, okay, so the second part is the CLI, which is kind of what we were taking a look before we, we, uh, kind of wrap things up here, is like the CLI. There is like the first step into like kind of running over Cloud Code, really, because one is like they have the hooks and everything into their own app so they can call their own private APIs and leverage all of the intelligence that they've been building up for the last like 24 months plus. And so that's a huge advantage, right? Cloud Code has to run shell commands and run everything in the terminal. Everything terminal-wise. I could be wrong here. I don't know. This is me guessing how those things work, but I'm kind of thinking like, bro, there's, there these guys have only up to go, bro. Like, this is just the very beginnings. Two is we're going to have to rethink the way that we prompt the model and it's actually going to be a lot easier. So, you can, I just verbally dumped an idea. I didn't have any structure to it whatsoever. And then that's where it really kind of shined. So, GPT-5 shined because I just had so much like I just said, "This is what I want to do. I don't know how I'm going to do it, but just go ahead and figure it out." And then that's like the first step, taking the plan and then telling it to go do the other plan. So, that makes me very excited for background agents because I feel like the background agents can now be trusted a lot more for me. And I have a feeling that the model is not going to want to just say, "Hey, I did the code for you know," like, and then not gaslight me. And I think that's kind of where Opus 4.1 is going. Opus 4.1 with code is wanting to gaslight me more often than I've seen before and I really kind of don't like that, to be honest. So, um, that's kind of what I'm thinking about. Uh, great stream, Ray. We're excited about all the new announcements today, bro. It's been crazy, you know. Yeah. Uh, better code analysis similar to Cloud 4. No joke. Daily for sure as of, yeah. 200k tokens as well already, like, you know. Yeah. You're you're definitely going to be blazing for sure. That's what I feel. Uh, funny. GPT-5 is, uh, telling people the coding aspect is just GPT-40 with slightly more reasoning. Bro, I don't think so, bro. It has a lot more intelligence than that. I, I can tell you that for sure just from playing with it. You know, the back card animation should be the opposite direction of the swipe card. Oh, yes. Yeah, yeah, yeah. As far as like the depth, yes. OpenAI will buy Cursor, bro. I think OpenAI CLI is next. We'll see. My fear is that when you're building, uh, where the stack is complex, multi-lang, Docker containers, O3, and Cursor didn't get. So, Alex, I think this is where you're going to be very impressed with this model and then even if you turn on high, which you may not even have to, GPT-5 with Cursor, I think you're going to be very blown away with. I think Cursor as well, as an IDE for their update, um, has really like crushed it. So, I, I'm, I'm really curious. The other thing, uh, product I'm really curious for is AMP. So, AMP code is similar to like, like, it's a, it's a coding agent similar to like other agents that you've used, like client, or whatever. But AMP code was kind of like the first to come out like back in February with like a lot more intelligence to other models. The only downside for that that people are saying is that they pay out of pocket even for Claude Sonnet. But people have preferred it for larger code bases because it seems to have a better understanding for the larger code bases. So, check this out. GPT-5 and AMP coming together and they had this new thing called the Oracle model that just got released a couple weeks ago. Like, it's about to eat up. And I'm, I'm very bullish on them. No one's talking about AMP. And trust me, like when that integration comes together, it'll be really interesting to see what that looks like, especially since the price of GPT-5 is not crazy, right? Compared to Claude Sonnet and Claude Opus, you're like, "Get out of here with those prices." I'm not really trying to pay that. AMP code plus GPT-5. Like, I don't need those fast priority requests. We were just doing the regular priority. We didn't go fast priority. We're chilling. We're so chilling right now. So, there's a lot of, there's a lot of positives to this whole thing. And you know, OpenAI is back for real. For real, for real.

Does Cursor keep history? Changes made? Changes by GPT? The Cursor keeps a history of everything that's happened in the IDE. So, like, whatever you're changing, all of the changes are kept. So, like, they understand, you know, what stuff has been checked in, what stuff you hit deny, all that stuff, and they, they, that's how they're like keeping the IDE very fresh and they kind of know, they almost feel try to predict what you're doing next. So, I don't know. Cloud Code still feels good after 2 hours of testing GPT-5 inside a Cursor. Yeah, I'm, I'm curious like, I feel like once you have a flow with Cloud Code, I know people have like a sub-agent flow or they have very specific things that could be it. Um, so, I'm, I'm really curious to see like what, what, what your stack is and would love to know more about that as well. Also, if you're interested in joining the community, feel free to do that. Just hit the join button on YouTube and then there's a post that says like, "Members only." It's only $6.99 a month to join our Discord. So, you can be hanging out with other builders and stuff as we kind of talk about this. Uh, it's a pretty light, nice, lively community, lively community of folks who have all kinds of backgrounds, which I really appreciate that. So, if you're interested, definitely check it out. Um, total newbie. Start with Cursor plus GPT-5. Yeah, for sure. GPT-5 plus Cursor, you're going to go a very long way in that agent coding. You're going to have a lot of, for me, it feels like you're going to have a lot of fun time interacting with the model, but also have it teach you some stuff too, which I think is going to be really the fun part for me. That's where I have a lot of fun. Bro, I had GPT-5 work with the Trading View indicator and it totally rewrote the logic and made me a cash machine. Dude, do not sleep on Trading View. Especially those indicators. They're so, they're so fire. Uh, that's such a, I love, I love that thing. Yeah, I actually, it's funny because I use GPT-4 with Trading View a lot and it's just, it's so good, man. Like, I've wrote so many cool stuff with it. Uh, given Anthropic's luxurious pricing, I'm delighted to see Cursor embrace GPT-5. Facts, 1,000%. Uh, are you running on? Bro, I'm running on every plan, bro. I got. Are you running ChatGPT? I have ChatGPT Pro, 200 bucks a month. I just canceled my Grok 4 heavy plan. That's 300 bucks, right? That's supposed to renew in a couple days. I'm like, "Nope. Get out of here, bro. Y'all need to deliver some value before I sign up for you again." Never again. But it was worth it. I mean, that was like the, the biggest stream I had. I think this is the biggest, this this stream is going to be bigger than, uh, the Grok 4 heavy testing. I did like three and a half hours of Grok 4 heavy testing to find out it was just not ready yet. Um, I have now Cursor Ultra, 200 bucks, right? I have the CLI, I have the Claude Pro Max plan or the Claude Max plan, which I may cancel. Like, I may cancel or downgrade it, right? Um, and it just, if I'm going to do anything, I might just pay the API cost because I do actually have a bunch of credits in the API. So, I might cancel. I have like 300. I like, I, so, in order for me to get to tier four, I think I had to put in like 400 bucks. So, I have $400 and it expires in October. So, I need to start using it. Uh, I should probably just give it away to my members because I hate to see it go away. I don't want Anthropic to keep my money. I'd rather have my members use it. All right. So, like, hit me up. Um, and then I got some apps and stuff that we're working on too. So, I got to do that. Um, Rod Humble, Cursor pricing isn't much better. I mean, GPT-5 is really good. I mean, just this is an indicator. Grok 4 helped me make and GPT-5 revamped the logic. Yes, Grok 4 is the true math at it. So, you can use both. And that's kind of what I was thinking here too, is that like you can hand off your plan to Opus and have it to cook, or Grok 4, have it do some things and send it back to GPT-5. GPT-5 like refactoring is so good too. So, don't be afraid, especially inside of Cursor, bro, that's where I think it's really taken off. I mean, $75 per million tokens, Opus, which is silly. It is a big ass model though. O3 sucks on tool calling unlike GPT-5 and instruction following. And I don't know if you saw this in the keynote previously with tool calling, they had to have like these JSON things that you would spit out. It's even smarter than that. You can get to the little details similar to what the AI SDK with Vercel says. Like, "Here's a specific structure of data I want you to output for it," and it does it. That's what's crazy. And people aren't going to be talking about it until the next couple of days because I don't think they really understand what's really there right now, bro. It's, it's so crazy good.

Yo, what's happening, man? OpenAI bought Windsurf so would not make sense to build its own CLI. Dude, they didn't buy Windsurf, bro. They, they, they took an L on it, bro. Yeah. Uh, what does it, what tool does it suck at tool calling? Uh, I don't not really sure. Well, it's not worth learning coding as self-taught from a third-world country. Time to move on to something different with my life. I don't think.

All right. So, quick life lesson. You are whatever you think about. If this is how you feel, that will become true in your reality. And like, I don't believe that to be true for myself. And I think there's a lot of value to like learn how to do software engineering, right? So, coding is one aspect of engineering as a whole principle, and then there's like a whole business that sits on top of that. So, literally, there are people who make business decisions, like, you know, Sam Altman and those guys who work with engineers who work with these other people, and those guys like who make these business deals, literally are moving economies in the trillions of dollars, trillions, trillions, bro. So, like, don't get discouraged. You start somewhere. Start somewhere. The only thing that counts right now is passion. So, if you're passionate about this and you like doing it, just keep going. I mean, I'm on this live stream for, I don't know how many hours we've been on so far, three hours, on my passion. I love this stuff, right? I streamed every single day in July because I love this stuff. This is what I love to do, AI. And like the fact that AI even entered my engineering realm, like I could have been afraid and say, "I don't want anything to do with this." But I embraced it and I realized that this is going to allow me to get that freedom to move to the next step, to up my skills and take everything I can and like build together. Find other people who you're messing with right now who are into this AI thing that are near you. Like, go find them. Go seek them out. Go change your environment. If you're the environment you're in is very negative and it's getting you to think this way, you need to get out of that environment and be somewhere else where it's going to help you get to that next level. Because I'm telling you, this is going to help you unlock these ideas and get those things out of your head so you can better communicate these things. And the only way you're going to get better is to use these models. OpenAI dropped an open-source model that you can use for free on any computer. You can even run it on the phone. That conversation back and forth for free can help you to learn how to code. Go do that right now, bro. Like, lock in. Go do it. This is your time to do it right now, bro.

You guys got me to sub the Cursor. Bro, I swear this is not a sales call, bro. This is not a sales call. I still love Cloud Code. It's so amazing. I just love my bank balance better. Hey, did anyone else test Codex with GPT-5 yet? Nah, bro. I'm waiting on that right now. I'm going to wait till that thing calms down because it's so burning hot right now. Uh, would you use IntelliJ with Cursor CLI inside for Java? For sure. I love IntelliJ. I love IntelliJ so much. I would try both. See what works for you. For debugging, IntelliJ is amazing. That's the thing I miss. I'm trying to get back in, uh, out of Cursor. Like, I want to see like what's happening at the object level, right? What's happening with all my code and what, how like actually do the debugging, like stepping through the code and stuff like that. It's really hard to do in VS Code unless you have a bunch of plugins, but IntelliJ totally nails that. It's like so, such a nice experience and I wish that was there, but maybe it's coming. I don't know. But I, I would definitely just try out any flavor to be honest if you're using IntelliJ. Yeah.

Cloud Code $20, Cursor $20, free student. Co-pilot student. It's 2x cheaper to sign up for student than paying for subscriptions. Yeah, man. I'm not a student though. You know what I mean? Like, I dropped out of school so I, I can't claim student status. Like, I wish I could, but I don't know. At the same time, like, as a prof, as a person who was in this profession and made some money off of it, um, like this whole stream and everything's supported from members like you. So, uh, big shout out to my members, by the way. Everyone who's been able to participate and join in. Y'all are definitely crushing it. I've had 900 bucks in the OpenAI API since January this year. I'm actually good now. Bro, you're chilling. Had to drop from the stream a bit. Cursor CLI, Cursor on desktop. Um, I would like Cursor desktop right now. If you're, if you're into the UI, it's so damn good. And I think I would prefer, I think I'm going to go back and forth on this, but I think for me, I'd rather, I like to use the agent up here. It's, it's so smooth. It's so fast. And we haven't even tried, we went GPT high-ax. We haven't even tried the fast requests yet. And I bet you those are going to be just blazing hot. Um, so, yeah, let's just see what happens because I think I was supposed to fix this swipeable thing and we'll, we'll see from there. But I think that that's kind of what I'm working with. Like, I feel like you have a couple different options. I feel like the really good nail-down experience right now is on the desktop and the CLI is right there. It's coming. It's here. And in some cases, I think it could be a lot faster than than the cloud. But yeah, I think that's kind of where we're at right now. So, uh, will Anthropic cut off Cursor because the GPG-5? I don't know. We'll see what's up.

Yo, Luis, thank you so much for becoming a member, man. You're the real MVP. Feel free to slide on through. There's a post for the members only. You can go ahead and join the members only there in my, uh, channel. And then there's one post that describes how to get, uh, the Discord information over, um, so that I can get you on the Discord as well. So, yeah, that's what's up. Learn to become a software engineer. Passion will drive you, or you will crash sooner or later. That's my guy. Exactly. Exactly. Don't like get into this field because you see a lot of people making money. That's actually the fastest way to depression and the fastest way to just like not take care of your body and do these types of things. Um, the passion will come and like, fact that you're even curious about it and showing up to the stream and hanging out. There's a community of people here who are doing good. Connect with us, man. Like, that's what we do. We show up. So, if you want to do this, this is what I do here. Follow me, Ray Fernando 1337. You know, just by like hitting the subscribe button, you'll get your stuff. One of the things that I'm working out right now, maybe I'll make this right now, um, as a project live because a lot of people who follow me and they hit the little bell icon, when they select like all notifications, they don't even

Get notified when I go live, even though it says all. And so, it'd be kind of cool when I go live, like if people want to subscribe over text message or WhatsApp that like I just have something that notifies them. So maybe I might build this app live here using the stuff and you know, just hook it up to the Trello API and it should be a pretty fast build. Uh, I would think and then kind of so people who subscribe and want to be notified. I think that'll be kind of fun.

Uh, sales call man, you got me on that. Chat GPT then bro, like six maybe. Eddie, I should just have you pay me bro. Oh man. I feel like Eddie, I I'd love to hear your experience, Eddie. You're you're a loyal member of mine. I feel like you've really taken to heart and you're very curious to you've you've tried these things and I want to hear you on the Discord too. Like, what has been your experience? You know, try Cursor today. Like, if you even want to cancel Claude Max and then cuz you can always resub to any one of these by the way. You can just cancel them and then resub later. Like, it doesn't hurt. They It's like a Netflix subscription. people expect to cancel and then kind of come back again when they write things up here. Um, like, let me know, Eddie. I I want to hear from you. I I respect your your views as a member for sure.

Uh, neither. I'm going to drop out as well. Just did the math. Yeah. Oh, man. All right. Cool, cool, cool. So, let me see. Let me cook, fam. Let me cook. Let's go. Damn. All right. This is an epic stream. I need to get some water, too.

So, we have this here. We have the plan generated. We have the issues. We have the swipeable. We have a lot of stuff so far. I can't right now. We have this. Okay. So, right now when we load the mock data, we swipe it comes in, but the the state of the back car does not get reset until I do this. Right. That's kind of where it resets back. And then the other part I think that we were talking about is like when we swipe right, it doesn't like continue to animate forward and disappear. um, it just seems to kind of stay there as a card. And so that's kind of where we're kind of me getting messed up a little bit. And so one of the things we talked about was context rot earlier and how if we keep going down the CLI chat rabbit hole, it's possible that we won't fix the issue because we've introduced like some rot here. And there are a couple different things that could be going on for the CLI. It's like, is it just the limitations of the CLI? And then two is like, I don't know exactly what model it's running. Like, it says GPT5, but is it GPT5 high or is it G GPT5 like medium? And is it just automatic right here? So like, these are some of the questions I have right now because it's not really clear what specific sauce of the model is running for me. Um, where over here I can be very explicit about and say, hey, uh, this is exactly what I'm doing so far. So, um, yeah, maybe I can just kind of do that. And I kind of want to try the fast request. By the way, I haven't done any fast requests yet. So, I'm really curious to see how fast like the priority scheduling is because that'd be really cool because apparently it's like we have some free credits to play with. So, why not, right? It's launch week.

Um, all right. So, we have this here. It's It's done its thing. I want to check in this code right now so far. So, I'm going to go ahead and do that as well. Go ahead and commit this thing. Sync sync the changes. Man, dude, this stream is wild. 2,000 people uh on X and then we got 451 on YouTube. Let's cook. Let's cook. Yo.

All right. So, let's go ahead and create a new window here. I want to I want to So, we tested the CLI. It runs pretty fast. Is it a cloud code killer right now? Probably not. It will it be in the future? I think so. Um, we're testing it with GPT5. Maybe I can try Cloud Sonic because I can select a different model. Model. Yeah. And I could say Sonnet 4. Let's go. So, switching models is only possible at the beginning of a conversation. Oh, okay. So, maybe we could do clear. Okay. Model and then model ls. Let's see. Oh, so I have different models. So, I have GPT5. Okay. Model slashmod ls. GPT5 sonic phone. Okay. /model and then sonnet. Okay, cool. So, I'm really curious to run Sonnet inside of Cursor CLI and we'll get a sense of the speed, right? This is kind of what I was talking to you about like maybe the CLI is a lot faster for this. So, let's just check this out. I'm really curious. So, I want to give it the issue that we have. So, when I swipe the card, the back card state doesn't like How do I say this? It's even hard to communicate what this bug is, but okay, let me see if I can do this right now. When I swipe the card in the GitHub issues using the mock data, I notice that the card swipe is successful and the card in the back seems to unblur and the state kind of gets stuck there when the new card comes in and it doesn't seem to reset. Also, I noticed the size doesn't seem to mold into the new card. So, as the back card is animating in, it just feels very unnatural. I'm going to need you to think harder and look through my codebase and truly understand kind of what's happening here to see if we can make this a very fluid experience. It should feel like you're picking up a card. And I do like the bokeh experience that's happening where it's blurred out and it kind of comes back in, but it everything kind of needs to be solid as we remove cards or go through the different cards left or right. And so we're going to be using Cloud Sonnet in this really long request. And I'm just really curious to see how the CLI is going to do. And so this is Cursor's CLI. And because I did the resizing there, let's see, it's freaking out. Let's see.

All right, cool. So, it's doing its own to-dos list here. So, it's working on four to-dos, and it's now doing some globbed search. So, a glob is probably their own version of like gpping and and doing different stuff within the IDE. And it doesn't have to be in the IDE, by the way. It could just live in the terminal. So, I could pop open this terminal window and or just open terminal and run this. So, this is Cursor agent. So, Cursor-agent, it's their Cursor CLI. So, their CLI tool is the one that actually is running here. And you know it's grabbing these different pieces of code. And what I'm also very curious with is like they don't yet show me what the context management is or how much tokens they've used. One of my favorite things of what's happening in the agent here is whenever there's a request and it finishes, it actually tells us how much tokens it's used like 20% and so forth. So, one of the things that we noticed when using GPT5 inside of Cursor is how little tokens Cursor uses for the requests which is kind of nice despite like you don't need the full context window like you don't need 400 token context window because Cursor is extremely efficient at using their own tools to grab the files and the stuff they need and it just seems to be like at another level. So, in some funny ways, we're comparing Claude Sonnet as a code model inside of Cursor as in in the actual CLI itself right? So, Claude code versus Cursor Cursor CLI right? And so, it's still the same model and to me, it just seems to be responding a lot faster for writing code and changes. So, let's improve the background card animations and so identify the blur reset issues. Now, let's improve the background state animations. Now, let's improve the front card animation. Now, let's improve the Tinder card. So, like as far as output, I'm preferring right now the output from Claude Sonnet in the Cursor CLI, right? So, like it just depends on whatever the issue is. You can try different things. Um, I find it this is really fascinating. Now, let's do a reset when the cards are loaded to ensure a clean state. Now, let's add the final improvement. Fire. All done. Bro, this thing's fast. Let's go. I don't know what it's generating here, but we'll see. Yeah. Wow, that was pretty fast. So, fix the reset state, set drag to zero. Great. Enhance the card animations, improve the front card entrance, and then enhance Tinder card component, and then added the third card layer, so creates a deeper stacked effect. Nice. So, no stack blur states between cards, smooth transitions, better visuals. Let's go.

All right. So, let's take a look here. So, this is still chill. And then I'm going to reset this one. Bundev. All right. Bam. Go to my dashboard. And then it's got to rebuild because of the the way the app was built. Okay. Issue review. And let me load the mock data now. Ooh. Okay. I think what I want to do is like there's a little bit too much of a delay. And that's kind of cool. Yeah. I mean it. Yeah. I I I mean, I don't know. I I feel like it can continue to be improved. You know, is it enough to like as a V1 prototype? I think it's pretty cool. But I feel like it could definitely be improved, but we've been kind of trying this out. This is so funny. Yeah, that's so cool. Oh, so there's a regression now. So, when I hit approve or hit request, it doesn't flip make the card fly out. So, this is a regression. So, that's why I'm hitting the buttons. So, let me see if I can fix itself. So, the new animations could be tightened up a little bit. When I swipe left or right, the resizing kind of looks unnatural. But the regression that we're seeing is when I click the buttons, it doesn't move them out of the way. It just seems unresponsive when I click the buttons. Just going to let it do its thing.

I have a question. Does OpenAI $200 plan work the same as Anthropic's Max 200 20x? Can I use Codex and have unlimited requests or do I need to use the API to use Codex? I think I think it's definitely different. Let's see. Uh, OpenAI Codex usage limits. See ChatGPT. Oh, I got I got the UI. Yeah. So, let me ask the same thing here. Let's see. Um, do I get unlimited Codex requests with GPT5? So, we're going to ask GPT5 about GPT5 so we can use GPT5 with GPT5. Yo, dog. Okay. Okay. So, what is Codex versus GPT5? Okay. Are Codex requests unlimited GPT5? No, they're separate. GPT5 access. Okay. Only Pro and Team gets unlimited GPT5 model access. And but that doesn't mean unlimited Codex task. Codex usually. So, I'm not automatically likely constrained. Refer to open is okay. Let's take a look at May 2025. Um, there were a bunch of announcements today, August 7th. Can you look up the latest announcements and look up the information around Codex usage as well for that? Interesting how it's like referencing older information.

So, yo, GPT5 for searching like this is nice. like the speed, but like it'd be interesting to see how this turns out for like day daily usage because I I used to use 03. 03 was my main model for everything. Oh, they got rid of 03 everything. Oh, no. I love 03. 03 was my beast mode bottle. It just has thinking. Oh, it doesn't support search. Oh, I want my 03. I just want my 03 fast search. Oh my god. Oh, that's that's that's weird. Okay, TLDDR officially live Codex. What's the deal? So, limits for Plus users aren't publicly stated. Some users report not hitting any after 25 tasks. Okay. Codex has it own operational constraints like a hard time limit for downloading your repo. I know that. No unlimited even with GPT5. So, there is no new update. So, I guess it's whatever it was before. Um. Huh. That's interesting. Yeah. Quinn is still the best and it's free. Okay. GBT3 is GPT5 thinking. Yeah. I also like it as a search model as well. Yeah. So, you pay per pro to you can still use 03 for real. Like, where where can I do that? Cuz I don't see it, bro. Like, I don't I I just see this. That's it. Like, if you can if you can get me the sauce on how I can use 03, bro, other than the API, that's what's up. Did you check the new bro? Where you been? Bal, we've been getting in this.

Uh, yeah. So, we we used the Cursor CLI with GPT5. Results were kind of mixed with GPT5, right? Fixing the bug and everything. Cloud Sonnet using the Cursor CLI is actually pretty damn good. It's really fast. It moves really quickly. And I like that speed over Cloud uh code. As far as like everyday usage, would I use the CLI every day? Probably not. I would probably stick more inside of GPT5 with Cursor and because we have way more variability. So, the biggest thing I can use there is the high mode. So, I can do GPT5 high which is basically the term for their like ultra thinking. And to be honest, I won't even need that at all. I would just stick to the medium, the regular GPT5 for most everything because of how good it was able to like search through the codebase, spend a lot of time looking through stuff and getting through those details. So, the first step is usually generating my plan in a lot of detail. I might use the high model to help me generate the plan. But once I do that, just pop over to the medium, which is the just the regular GPT5 and just let it just go through the plan one by one. Really, really good. If you need to basically refactor or solve a really challenging problem, pop in that GPT5, pop in that high um, yeah, GPT5 high, which will give it that extended thinking window. And I think you're going to be pretty good on that. As far as token efficiency, GPT5 inside of Cursor is really, really good. So, I've been able to see GPT5 just hound through a lot of files, shrink through lots of things, and it doesn't even eat up that many tokens. And I think that's because Cursor has done a really good job of understanding what those requests are. It's likely that they're routing to other models internally to grab that information for you and then feed that into the context. And the the tool calling and tool following of GPT5 inside Cursor really really shines here and it's super impressive. It's like mad impressive.

Yo, so uh that's kind of where I'm rolling if you're kind of just catching up to the thread. Yeah. So, okay, cool. Uh, let's see what's happening here. So, we want to check out to see what the latest changes have happened and we want to see how it's doing here. So, Whoops. Okay. So, I think we already did it, right? We already did load mock data, right? So, this says it fixed some of the issues with the Tinder card. The flag was never reset. Animation smoothness. Let me just kind of rebuild everything. I think we're I think we're good. I don't I was just kind of testing stuff out kind of Oh, when I click the buttons, that's what it was. When I click the buttons, it didn't do its thing. So, yeah, there we go. Yeah. Now, when I click the buttons, they'll actually work, too. And then we also have the the phone UI. So, if you're on the phone, this is actually how I expect to actually make it work. And so one of the things that I wanted to do is like I have this like design doc. And so I want to have it build this design deck and see if it can maybe um we want to build basically our own design spec to be like a design reviewer. And that was something that was demoed in GPT5 where they basically get like a design deck and say, "Hey, this is my spec and I want you to go ahead and update the stuff." and it didn't look so impressive, but I have a really long design spec and I want to see if it can actually like help us with that because as we can see there's some issues here in our UI especially on mobile like when you're doing this type of thing see how this little overlap is happening and then when I do it on regular desktop view it's it's spread apart. So, there are little things like like the design hierarchy that we want to take care of. And so in my repo, uh, if you go to my GitHub, Ray Fernando 1337, uh, in here I have a thing called LLM Cursor rules. Oh my god, 515 stars. That's crazy. Y'all are crazy for this. Okay, so one, the thing that I have in here, it's called the optimization principles. This one is correct. So, in this optimization principles, basically what I No, it's not the optimization. My bad. It is called productivity. No, no, no. Fire design team. Come on, bro. All right. I promise it won't be hype. I promise this will be worth your time. I promise. All right. So, if you're locked in right now, drop a five in the chat cuz I want to know how many of y'all are alive or how many of y'all are bots right now cuz I got like six people in on Twitch and I haven't seen any messages come in on Twitch here. So, I'm wondering if my thing is broken. I need to check in on that. And then I see some people on X. It says 2,272. I don't know if anyone's actually on X. Yo, El Chess the five on it. We got Brett House on the five. I do some Okay. JK Software for sure on Twitch. So, I know my Twitch chat is definitely up and working. Rodrigo in the mix. Jose Brea got five on it. Got Marthon. We got AI. Dang. Okay. My My boy got four on it, bro. Get out of here. We can't have that, bro. We can't We got to get five on it. The GPT4 was so like 2024, bro. Like, we can't have that. Xbox for sure. A passive vibe. Come on, man. Let's go. All right. My chat's awake. Okay, I'm awake, too. I'm hopefully I'm awake. Y'all y'all are in here.

Okay, so this is the one I was talking about. It's called Fire Your Design Team. All right, I promise this there's like like if we put the RZ aside from this, like this is going to be some real sauce for you. So, I did some research and I said we need something that we can repeat over and over again as far as design principles. And so what we need are like a typography system with different weight sizes and stuff. So, when we give something like this to a language model, it's going to follow these instructions and make sure that all the code that's generated follows these principles. So, anytime you generate code, especially with Tailwind CSS, the AI model is going to prefer what's in its training set for specific buttons, for specific sliders. So, what can happen is that you'll have this design inconsistencies all over. You'll see like, well, that button's too small and that one's too big and like this just looks off, right? You just don't know how to describe it. So, if you have a design system in place, I have this for free, by the way. It's all on my GitHub, Ray Fernando 1337. We're going to take this document and see if we can put it inside of our code and let GPT5 cook. We're gonna we're going to do we're going to let GPT5 cook in Cursor with hi with five GPT5 on high mode Snoop Dogg level and we're going to do priority mode. We're going to spend so much money on this model to do this design system review and we're going to see that thing blaze fast hopefully. And if you're down to see this, this is really crazy. We're about to hit four hours on this stream. Damn. And if somebody drops the $100 super chat, I'll have to stay on for another hour. I don't know how I'm going to do this. This is going to be crazy. So hopefully Yeah. No, don't. I shouldn't have said anything.

All right. All right. Let's go. All right. So, check this out. Uh, 800 point grid system. Um, we got all spacing values divisible by eight or four. So, that's you'll see some Tailwind values coming in with number three and there is stuff of because of its training set. This is why I'm telling you about this 603010 color rule. This is probably like the most important one, but I don't necessarily need this. So, I'm going to do a thing called prompt authoring. So, I'm going to give this as an example to GBT5 and see if it can author a new set of rules without following my rules. So, this is how we're going to see some of the intelligence of the model. Is the model smart enough to not follow the rules but write its own rules and do its own investigation in the codebase and then see if it can actually formulate us like a similar design doc but something that will fit our design you know style basically. And so that's what we're rolling with and I'm just going to cook and you'll see how I do this right here. So, we're going to go ahead and go to code, and we're just going to grab the a copy of the raw file like this. And I'm going to kill these uh terminal instances for now. I'm going to check this code in because it's sort of works. Yeah, it does work because we hit the buttons. And then we're going to see the current state of things here. So, that's doing that there. And now what we're going to go ahead and do is pop back into like agent mode inside of Cursor. And this is where we're just going to let the sauce do its thing. So, bonx convex dev. Okay, cool. I'm going to close this window, please. Okay, get rid of this. Let's bring back this sauce here and we're going to open a new chat and we're going to do a couple things here. So, I don't think I have a tab. Okay, so let me go and seek the changes. All right. So, what I might do is let's see. I'm going to do some delineation. So, I'm going to I'm going to do a couple things. Oh, before we do anything, what I want to do is we're going to do GPT5 high, but we're going to do high fast. Bro, are you guys ready for this? We're going to get high on GPT5, drive a Lambo with a blindfold on, and just go fast, fast, fast, fast. Let's go. Let's go. It's Lambos or nothing, bro. It's levels or nothing. Let's go.

All right. All right. Okay. Check this out. I want you to think hard for this task cuz you're going to do an extensive review of my code. And I want you to generate a design system for my codebase that we will follow for various design reviews whenever we need to do them. By the time you're done, you're going to need to generate a design review document as a markdown file in the documentation folder. And attached below is a sample of a design guideline stock that I want you to reference and use for this task. All right, so we have this. I'm going to hit shift and hit enter a couple times. And then I'm going to just say hashtag. Actually, I'm just going to paste it in by low. There it is. That's the whole document that I pasted in from GitHub, right? I just went copy raw. And so that design review doc is down below. One of the good things about GPT like opens models is they follow markdown very well. So this is kind of delineated the top the bottom. We're going to be GPT5. We're going straight up high fast max. High being this is like the longest thinking model because we want the accuracy to be there. Fast being it's going to be like the high priority one which is actually going to cost us quite a bit of money. But because we are doing it here, it's going to give us some free access I guess for the first couple of days and we're going to have top priority for this request. So, if you guys are ready to cook, let's go. Let's go. I'mma send it. Oh my god, bro. I don't think I'm ready for this. I don't think I'm ready for this. Oh my god, it says cook with high fast. All right, so we're gonna This is going to be really fun. So, now we actually get to see a little bit more of what's going on under the hood and we're going to see the power of Cursor as an actual IDE to just just just absolutely I think it's going to obliterate this whole thing. I think it's just going to be But check this out. Now we have this document that hopefully we generate which would we're going to use going forward. So, anytime I have stuff that I want to do for making UI changes, I'm going to see if I can make a GitHub webhook or something like that. So, when I kick off a background agent to do a task after it commits it into code or do something, I'm going to want it to do a secondary pass where it's going to go ahead and grab this file and do a code review for everything it generated. That's the goal here. That's the That's what I would do as well if I had like a CLI type of thing. So, yeah.

All right. So, this is doing its thing here. It's basically now reading through my document. Look at all the files it's grabbing right now. This is this is going through a lot of sauce right now. This is crazy. All right, cool. What? This is nuts. All right. In my opinion, Cloud Code feels better. Yeah, I'm really curious, too. So, like this is the thing that we're still trying to figure out like how do we prompt with these? How do these models behave? What's their strengths? What their weaknesses? So, uh, but you kind of write what's who's that? No disrespect. Who's Oboom? I don't know who that is. Okay. Uh, token while flying sounds fun. Yeah. EVT5 isn't very good for mobile front-end development. You mean are you using React or using Swift? Because I'm really curious. I have a Swift app. Let me see if I can show you what I'm working on recent. Where is it? This one? This demo one? Yeah, I think I'll show you later. But basically, it's using the OpenAI Images API to generate images and it does them super fast. Actually, yeah, it takes a little while to run, but I'll do that there. Can I just pop in Xcode? If I pop in Xcode, am I just going to kill my whole stream because of all the memory it's going to eat? Uh, I hate Xcode so much for that reason sometimes. Like, I have a lot of beef with Xcode for a lot of different reasons. It's like reasons beyond the stream right now. And I I don't want to be an old man right now and just go on rants. Oh, man. Oh, so much so much beef.

All right. Okay. So, like Cursor. Oh, because we were high fast, we got the actual Damn, that was fast. Holy smokes, bro. That was really fast, actually. Wow. Um, dang. Okay, my whole life just got flipped upside down because I didn't realize how fast that was. Here I was thinking we're going to be chilling for like 10, 15 minutes, but it's already done, bro. Damn. Okay. Okay. Okay, bro. We cooked. Oh my goodness. Like, it even has a review checklist at the end. Uses only four font sizes. All layout spaces are divisible by this. Let's go. Let's go. What? Okay. So, I like that it's actually referencing some of the stuff in my Tailwind CSS values about how I should actually um proceed with the existing token. So, that's one of the powers of Tailwind CSS and like the the new V4 system is this like tokenizing thing. So, you can just say call like d- primary or secondary for your accent colors and stuff. And then whenever you need to tweak them, you just kind of tweak them with one change there in the root file. You don't have to like go into every single class and change like red, you know, value to 700 to 800 or something like that. uh for every single component. This is nice. And one of the other things that I liked about this whole experience, and this could probably also happen in Cloud Code. You can do this in Cloud Code. You can do this in any IDE, by the way, but we're just testing GPT5. Um, is like it's going to go through and because it has that reference document, it's structuring it in the same way, but specific to my codebase. This is hot. This is super hot. And this is called the three comma, bro. We have our own design system. It's called the three commas design system guidelines, bro. Like, drop a three commas in the chat if you're in a billionaire status right now, bro. We're trying to get to that billionaire status. Let's go. Oh, man. Let's go. Same. Lol. X code spec. Uh, especially when we run Unity for AR Dev Nightmare. Yes, bro. They did just add Copilot though. Okay. Okay. Okay. Uh, I keep talking like to GPT5, but Opus keeps producing better results uh, in most of the refactor tasks that I threw both. Oh, dude, Brad, please share in the Discord. Want to see some more of that. I think that's really cool for sure. Yo, three commas in the club. Buddy Williams as well. Let's go. Three commas Rodrigo with the three commas. T commas. Yo, Super Kraken knows what's up, man. He knows what's up. Tas, let's go. Oh, man. This is funny. This is too funny.

All right, so now that we have this like basically our own handcrafted design doc, now we're This is where we're going to let GPT 5 fly. This is This is going to be so fire right now. So, let's just see some of the comments that it did. It reviewed our actual document and authored a specific repo design, right? So, it saved it as here. I'm going to make sure I check this stuff in so I don't lose it. Um, and then we can kind of reference this going forward. Go ahead and commit. All right. Sync the changes. Let's go. So, this is what it actually added. So, compared to the doc that I have currently on GitHub, it added like the core principles, the tokenization that's aligned with my variables here. So, like that's kind of cool that it just grabbed those for the alignment, the 8oint grid system, component architecture, and container query usage and motion details. And then some of this other stuff specific to uh the actual parts like a switch that I have as well. That's awesome. Wow. It's going to prefer size 23 over 2.5 for the tool tip arrow if acceptable. So, apply spacing fixes to the files above. Oh, it even identified the files that need the fixes. Whoa. So, it took a look at existing files in my repo and it realized that it violated some of the rules as it was writing it out. I don't know what Open AI is feeding these people. Y'all are crazy out there, bro. And Cursor teams cracked. Cracked. Oh my gosh. Like, there's some crazy intelligence in this model because as the model is writing out the changes for us to like talk about it, it's it's like it's saying, "Hey, bro, like I took a time to review all of your code and I know you want this design spec and here's your design spec and this is like what we should do for your design spec. And by the way, while I was already digging through your files, here's also some list of files that have already violated that you may want to consider and change. Like, it's just being super nice about you. It's not like gaslighting you. It's not saying like absolutely. Wow. Wow. Wow. Okay. Okay. Okay. I'm so excited for this next move right now. This is about to be this, bro. This is This is fun. Okay. Design system guidelines. Let's go ahead and pop in a new chat and let's just say hi, friends. Let's just stay hi. Let's go fast and let's get ripping right now. Let's go. We got no time for this. All right. Design system guidelines. I just call it like that. Um, I want you to thoroughly think through all of the code and especially all the recent changes I've been making to make sure that we fit the design system guidelines. I think that's it. Uh, let's go. Let's just see. I'm just going to let the agent really just cook with this and see what it does and see how like it's going to go through all the files and everything like that. Conducting code review. Yeah, it's basically just doing a code review. And this is this is kind of what I want. Like I want like code reviews to kick off to do these types of things like for design, right? This is super powerful especially if you work with designers or anyone who's doing design stuff. Like they're going to want to set up their own branding system. Your own brand has a specific design style. You're going to want to generate a document like this. We can probably even generate one based off of screenshots. Like we can go get some screenshots, throw them in there saying this is how my I want my brand to look like or I want my brand to feel. Can you generate this design doc that can help me like enforce this throughout my app? Right? So that can also be super helpful because now you have a document that you have as basically your your central piece that everyone's going to go and reference from. And this is like what you would do if you're working at any company. There'll be some designers. They'll like make a design document or something that can describe what your brand is. You know, every time you use a specific font, you always want this font. You always want it for this size. You know, those types of things are like usually like in some brand document somewhere. And this is what you want to do to kind of like basically 10x your power, right? Just in this one document without code. It's all English, right? It describes instructions that a language model can keep grabbing and then figuring out what to do with your software. Super powerful. Super super powerful. Yo, Mr. Frank coming in with three commas. You're back, man. Let's go. Burn them tokens. Does Cursor CI have web access? I think it does. Yeah, it does because we're doing some web research and stuff. Dude, man. Well, I've been pretty good, man. I've been chilling. We've been chilling. Bye. Feeling fine for sure. GPT5 for free on Windsurf, bro. Yo, I didn't realize they're back. I'm curious to see how that does. Yeah, bro.

Okay, it's planning the next moves. Man, it read a lot of files. Let's see. So, it's evaluating text sizes. It's adjusting component sizes. So, what I find really interesting here, right, is that I'm literally throwing at the kitchen sink. And before, what I used to do is I would chunk all my prompts down to very small pieces. And this is literally in the last 24 hours that I'm just rechanging the way I prompt again, right? And this is why it's important to play around with these models. Like, throw all your assumptions out and then see if you can just cook again, right? This is you're watching me live in real time prompt these models and you're seeing what it's doing. And to me, like how agentic this is feeling is kind of insane. This is a level of detail that they're going through for like thinking and then reviewing files and then thinking again and then applying them like it thought for like a minute and a half again and it's going through all of our files and like making those little changes, right? Divisible by four or eight, right? Like the tabs like it's going to make them feel like Yeah, that's crazy. Interactive. There's noise. Wow. The file is being edited by another file. Oh, my bad. It's probably me. Accept. All right, let's go. Wow. Demo. Yeah. Extra small. Oh, wow. Running a lint check on the edited file. Intro issues were introduced. Let's go. Oh, finally got GPT5 in the UK. Okay. Yo, that's what's up. Ardra, how's it going as a member? Thank you so much. Any chance for a quick summary? Yeah, I got you, fam. Right now, um, GPT5 versus like Claude code and so forth. GPT5 is a very powerful, capable agent. It likes to be steered. It likes directions. It likes lots of plans. And so, we have a couple different options inside of Cursor to configure them. And so inside of Cursor, if you go to models, you can type in GPT5. You'll probably already see them enabled. So, we have these different modes. By default, the one that says GPT5 is going to have a reasoning of medium. So, that's all you need. This is your everyday model. This is probably all you need. You want to do high. It's going to use up um more of a thinking window. So, think about this in Cloud Code terms of like Ultraink, right? It's going to think for longer to do different things. it will take a longer time, but um theoretically that's kind of what you're doing there. Um, the other than that, you know, with all of these models, whether it's regular, medium, low, medium, or high, you have a fast/priority version of that which can run and you basically get your your request and everything is basically first. And it is really really fast. As you were seeing right now, it's it's pretty fast. If it was the normal request, which is what I was doing earlier in the stream, it takes a little while. We can just do things and come back to it. Um, you do have the other models here. You have the low version. You have like nano and so forth. But basically, you have GPT low, medium, and high. And your everyday model, just medium, you're going to be fine. You're you're totally fine. Uh, as far as context window, I find there's a lot of context window efficiencies if you run inside of Cursor itself. Cursor seems to be extremely good at managing the context window. So, you don't have to really worry too much about it. And it does give you a preview here. Um, as far as a CLI, the CLI is very promising for Cursor. It's even if you use the Sonnet model with the CLI, it's also very promising inside of Cursor. I'm not really sure yet because it's still in alpha phase what specific model I'm running with GPT5 in the CLI. So like, is it running the high, is it running the medium, is it running the low, is it running the nano model? It just says GPT5. So, I'm assuming it's just the medium thinking, but not really sure. Um, what we've done so far is we've been able to get a template and use Cursor agent with GPT5 medium reasoning and it was really good at understanding what my codebase is doing, grabbing those files, you know, generating an app and we the app that we made is called a Tinder UI. So like if you've ever done the dating app Tinder, you can swipe or swip swipe left or swipe right for, you know, if you like someone or not. And then if you match then you go out on a date. So, we basically build the similar thing but just for coding. So, we wanted to have code reviews done with Tinder and we called in a specific framework for that. We run we ran it against Cursor using GPT5 and had it build out it stuff. So, we we have the scaffolding right there for the app. Everything's up and running and working. It's deployed to basically prod in some way and the cloud code version is you know gave us a plan and so forth. Um, you know, Cloud Code has a nice plan. Like the Opus plan that it generated with 41 is is really nice. And um, like I could go either way, you know what I mean? In terms of speed, I feel like Cursor has that speed kind of down. And the Cursor UI and the enhancements and everything inside of that with GPT5 to me has felt like a really great option. It's going to be like my new daily driver. That's where I'm going to. And part of the reason why I'm going into Cursor again is mostly because I want to be able to give it all these really crazy prompts that I do and this like whatever I'm thinking just put it down on paper. It gets me. It gets me. That's kind of the feeling I was getting when I first started with Cloud Code. But for whatever reason that context window with Cloud Code fills up really fast and I'm having to spend more time managing it because now I give it more and more tasks, right? So, like the more capable these AI gets, the more tasks we give it. So, in here, like I've just been super impressed with how just nonstructured my prompts are. You know, it's just a blinking text box. It's just accepting input and it seems to understand that, which is at a level I've not really seen in a while and I'm really excited about. But the second part is the fact that Cursor has just really cooked with all of their hooks in their IDE and they're able to like run these types of agentic tasks, right? We're still running this task for the design review and this could have been done in a background agent while we were like somewhere else else out in the world, right? I don't have to have Cursor on and it's literally meticulously going through using the GBT5 high, you know, extended thinking, right? It's using the the high priority for everything. So, it's not like it's taking a long time because it can. It's the highest priority task and it's it's cooking. It's it's keeps cooking. After it does the code changes, it thinks again. So, it thinks about the files. It thinks more again. It says, "Okay, ensuring compliance with the design guidelines. I really need to take a close look at the code, especially any recent changes to make sure there's adheres to the design system guidelines. See, it seems the user is asking for modifications to address any compliances, right?" So, it just keeps going. Amazing. Amazing. Amazing. Like, wow. Yeah, that's that's kind of what we're cooking with here. So, that's what's up.

Oh, man. Wow. This is so fire right now. Yeah. Um, so crazy. So, I'm I'm really curious to see how this thing's coming out. Let's see if I can actually pull up the the design here. Yeah. Let's see. Load the mock data. Retry. Oh, let me restart the server. Let's see. Bundev. Cool. Let us see. All right, cool dashboard. And then going to pull in this issue review load mock data. So, as far as some of the designs, I think it changed some of the bolding in some of these other areas. That's interesting. Open issue on GitHub. See, it did fix this issue here, too. Cool. Super cool. All right, cool. So, now it's basically just running the build to make sure any linting issues weren't uh Yeah. So, it's all done. There we go. That was a long ass time, but look at all the stuff that we did. Look at all the code changes we did, too. That's That's absolutely crazy. We nearly changed everything. Let's take a look and see. So, just went through here. Oh, what's going on? Oh, there's a lot of errors. There we go. Okay, cool. So, all of it's finally done. And I'm going to go ahead and keep all just to make sure we have all the changes. And then let me go ahead and changes. Stage all changes, right? Cool. Go back over here. Drop down menu. All right. Cool. Bundev. Dude, my computer's about to die. I've been streaming for a long ass time. Damn, we're four hours. Let's go. We're cooking. I My stream timed out on LinkedIn because I was streaming for too long. That's so funny. That's what's up. That's so crazy. Hero. Kro. Nah. No, no, not right now. What's the verdict, boys? Yo, I'm cooking with Cursor all day. All right, so let's let's I want to ship a feature with my app, Ray Transcribe. So, I want to show you what that is. That's looking like I need to use the bathroom, too. Um, so I might do that when I get back. Let's see. So, local host. Let's go and do the code review real quick. And then we're going to load the mock data. And then we're going to like Tinder UI. This go request a change.

This is so fire. Can I? Oops. Yeah, there's an overlap with the approve. Okay. All right. So, all of that, all of those files and everything we did, like it only used 25% of the context window. That's insane. Look at this. Of 272,000 tokens for the context window, it only used 25%. It touched all of those files, wrote all those changes in, and only used that many tokens. If we were using cloud code, there's no way we would be able to go through all of those files and all of that thinking and still have that efficiency in the token window. Look at that. Like, look at all those edits. That's insane. That's kind of why I'm telling you that like Cursor and GPT5 is the new power couple in Silicon Valley right now. Like it's crazy. And this is just getting started, right? The model just dropped today. I mean, sure, they've been like trying to make it work better, but I think there's something here. And like, sure, it may not be able to do something for you right now, but I think, yeah, I think this is going to be this is this is heavy, heavy. There's some sauce here, boys. This is crazy. Uh, yeah. Yeah. Thanks, man. We are we are so cooking with this right now. That is absolutely crazy.

I just joined. Someone give me the TLDDR if GPT5 or Cloud 4 is better in Cursor. All right, you got to go GPT5 and then you could just do GPT5 straight up medium. You don't need the high whatever. I'm doing. I'm just I'm just being stupid right now. I could have done this medium and we could have just been touching grass and I would have just came back. I could even just ran this as background agent and I would have just been chilling and like literally at the at the beach like doing my thing, coming back, you know? That's that's what I'll be doing. So the setup right now is like GPT5 medium. If you get a background agent running cooking, make sure you get your design system set up. Get that design system set up so you have these types of rules that you want to apply after code changes, right? This also works for cloud code. It doesn't work for Cursor as well. Like you still get both. You're going to get the power of the the background agents though to run and do that type of thinking and the extensiveness and the lack of like eating up all your token and context window. Like I'm not afraid anymore. Like I can go ham on like one chat for a long time now. That's that's really impressive. That's so impressive right now.

Um, so Cursor has that advantage over cloud code, right? Context window management. I don't have to think about spawning sub agents. I don't have to define the sub agents to do those things. GPT5 is going to figure that out for me. And we were just getting started with this, right? Everyone's going to apply their MCP, all of their existing tooling set because that's what they're used to with cloud code. But I'm telling you, you have to take the time right now to play with it. While it's free, by the way. I think there's some free token usage if you're a paying member or I don't know exactly what they're offering, right? Go play. Like your time investment in this is going to pay off dividends and you'll come up with newer workflows now. And don't be afraid. Like cancel that plan for Cloud Code Max and just play with this for like a month. See where you get, right? You're going to have like a way better understanding of these models and what its capabilities are, especially for your use cases that you may realize. And this is kind of what I'm starting to see now. It's like it's like it's blowing me up right now. Like we're seeing this live. This is so good. Too good. All right. That's kind of where I'm at right now. I have the ChatGPT 200 subscription. Don't use it like I do with Cloud Max. Sounds like more sensible thing to do. Switch from. Yeah, I mean, exactly. If you're not like using memories in ChatGPT, just the plus plan for ChatGPT is chill. Like you're good off that.

Um, yeah, the TLDDR just tuned in. Yeah. Uh, Bradley, I think you just missed it, but basically GPT5 Cursor new way of prompting like medium model. You're good, bro. Like go go go ham on that. Start using it a lot. Uh, we're going to add some more features to a different app. So we're going to move apps here and we're going to try something else. Um, and I want to show you a little bit, but I want to make sure I check in these changes so I don't lose them. Uh, we have all these changes. I'm going to yolo them. This is how we do yolo software development. Let's go. And I also have to go to the bathroom too. Let's go.

All right. So, all right. Let's show you real quick. So, this is our we basically b made uh if you're just catching up on the stream right now, we're basically doing code reviews, but with a Tinder UI. So, like Tinder for code reviews. I want you to go through your GitHub issues as fast as you go through your Pornhub. I mean, your your GitHub. That came out wrong. That came out wrong. That that came out wrong. Okay. Okay. Okay. Let's just let's reset that back up again. I want you to go through your GitHub issues as fast as you swiping left and right on Tinder. And then the way you're going to do that is through this UI that we made literally live hooked up to a database. You can have authentication login. We basically took an existing template. We popped it into GPT5 inside of Cursor and we had Cursor cook and GPT5 with Cursor is the new power couple in Silicon Valley and I literally think it's that meme with that girlfriend that guy's holding on and he's holding on to Claude Code Max and he's he's turning he's like, damn, GBT5 just passed me by. Holy. Yeah. Yeah. But it may not feel like it at first because you have to prompt and think differently. And so throw everything else you had away as far as context windows and stuff because you're going to be impressed with how efficient the Cursor is using the context window to look through your code and send it over to GPT5. And they just keep having this iteration that they do this. That's just beautiful to watch. It's so beautiful. It's so beautiful. I really, really like it. I really, really like it. Like I'm just I'm hyped because that means other AI tools can build off this. The tool calling is really good and Cursor is pretty much the golden star example right now, right? Just wait. Like wait till like the Lovables, Bolts, and all these other people start like really cooking with this. The Vzeros, I mean, I haven't even checked vzero.dev. Are they still alive? Let's go. Oh, GPT5. Let's go. Try GPT5. Oh my god, bro. Do I have enough time for this stream? I'm four hours in. Let's go. Are y'all ready to log in with me? Oh man, if somebody drops the $100 in YouTube, I have to stay one more hour, bro. I don't know if I could do this because I got to go to the city like in uh let's see. All right, do it. Try it. Oh my god, bro. This is so crazy. I was worried about the context window. Even though the model's 400k, bro, you don't even have to worry about the context window right now. Cursor's got you. Cursor has got you. And there is just. It's so good, bro. Lol. Wow.

All right. All right. All right. So, I want to tell you a little bit about my app, raytranscribes.com. I accidentally vibe coded the fastest transcribing system um last year using clouds 3.5 and I discovered that by putting the transcripts in my videos, I actually get ranked first like whenever people do Google searches. So this was me vibe coding on a live stream literally like a year ago and I was showing the power and capability using Vzero to actually make the transcriber. But then I gave it the documentation for like web assembly and a bunch of other crazy stuff and in like one shot we basically made this transcriber. So I can get like a 2-hour live stream and like in 90 seconds we'll have the whole thing transcribed absolutely crazy using the large whisper model. I made an open source version of this and that's kind of really where it blew up. And because all these people were hitting me up to do um uh they're like, "Hey, what about the Windows version? What about the Windows version?" I was like, "Bro, I don't I don't do Windows, bro. I I'm a Mac guy. I know, right?" But I I got you though. I got you. So that's why I made this version. So we vibe coded it live and you can basically do this type of thing here. So, one of the things I have here is like an outro or something and you can have a custom dictionary and put in your like words and stuff. And a lot of times, like I I popped open this like tweet because this guy named I guess um was saying um where was it here? What was the reply? So this is what happens. This is why people's videos and stuff don't get ranked. I'm just sharing some sauce for you as a content creator. Um, and it's basically because of this. So, like this popular creator made a video and his actual captions like that YouTube auto translates gets them incorrect. It says GBT5. Like literally like bro, people are not searching YouTube for that specific keyword. They're searching GPT5. And so, you have to have your your captions and stuff in your videos correct. And that's why I made this because as soon as I started doing that, when people type in those keywords, my my videos show up over all of them, right? Even if those like a three-hour live stream, I get ranked. All my stats show majority of my traffic comes from like Google search. And so like that's super important if you're a content creator, which is the the biggest reason why I made this app. And like like in no time, you'll have all this stuff already done. So you'll have all the right words and stuff. And if there's a wrong word in there, you can just put it into the dictionary and just add it so it does like a the right thing. So um if you're talking about GPT5, it's going to have that in there and all that stuff. So, you just download the SRT files and you put them into your things. But one of the things that I wanted to do was I made a first attempt at keeping a history of all of the all of the stuff that was transcribed in my database and displaying it here. But then I was treating it like an iOS app because that's kind of the I come from the client side world and I was like, oh, I should be able to run this multi-process. Though like if somebody goes to a different browser or does something else, I should be able to render this while the because unless someone quits Safari, I should still be able to run a background worker and run this thread. So like even if they come back and refresh the website, I should still have that worker running and have it reconnect to its state. I was so chill because I vibe coded it like in one shot. I was like, "Yeah, I got this." And then people started hitting me up with all these bugs like left and right. And I tested it and it worked, but there were just too many bugs to fix. And I tried to fix them and you know what ended up happening? I'm pretty sure you've seen this before. Fix this bug. Fix this bug. I created one bug and like 50 more show up. I test around them and I fix them. I finally think I got it fixed. I push out the new update and then like it it causes even more everything. So I had to roll everything back. Literally one month's worth of vibe coding. Completely roll back to my fresh state here where I'm at right now without all that stuff. And so I'm back to square one. So I want to see if GPT5 can reimplement the entire feature and we can just kind of cook from here. Just go piece by piece, right? Like I was using cloud code and I really was cooking. It was really cool but I kind of went too far. So I want to see if I can reimplement it because the next step from here is to do something that I do a lot with uh cloud code. So I have a cloud code project and one of the the whole reasons of even capturing your your time your not only your timestamps but anything from YouTube is because you can repurpose things for other purposes. So I have a whole hook generation thing here where I have instructions so you can actually generate hooks for videos and things and my retention rates for all my live streams and videos the first 30 seconds are over 80%. My last live stream that you see, the GPT5 one that we thought was going to come out was 160% retention rate for the first 30 seconds. Ridiculous, right? So, this all comes from basically getting the transcripts and running them through something like this. I like to author my prompts. I do a lot of research myself. I read books. I hand that information off and I kind of work back and forth in a conversation just like what I was showing you earlier in the stream. That prompt authoring technique where we're basically grabbing the design system and having it mold into our specific app is what I'm doing for these pieces of software. And the reason why I'm telling you this is that if you can author system prompts like this, that's one of your many pieces that you can use to leverage the intelligence when these new models come out. So now that GPT5 is coming out, I'm going to take that GPT5. I'm going to run it through my prompt. I'm going to run the transcripts through it. And we're going to have this multi-step process. And that's the reason why I have Convex for my database because they have this agentic workflow as part of one of their components that I can do this multi-step tool calling from now. And if Cursor can do it, you better believe I can do it, right? Cuz I'm seeing Cursor going through all those code files at different times. It thinks and it does another step. It thinks and it does another step. Guess what I could do? I could do the same thing that they're doing for their techniques, right? I can run it through one of these hook generation prompts and then I could run it again and then I can run it again and then I can run different sets of instructions. So now I'm generating this pipeline. So check this out. Like here's one that I did I think for um like let's see what was the first one that I did um AI orders pizza anonymously, right? So what I did here is I literally just gave it my like a YouTube short or or Tik Tok short. I gave it my live stream, like the whole transcript from the live stream and I just gave it the specific YouTube short that I cut out from it and I just said go. And this is what Opus did, right? Opus is a really expensive model and I'm paying the max plan for it. But look at what it was thinking about. It it goes through the thinking process because those are in the instruction set. It grabs the context for the live stream. It does a hook point analysis. This is something I read from Brendan K uh Brendan J. Kane's book about how to actually make stuff to make people stop and kind of listen. And this is teaching me how to become a better storyteller. By the way, I'm not a good writer. I'm learning to become a good writer by using AI, by having it author prompts, by reading books, by having these conversations back and forth. And look at this. This is a full like full-fledged app. This is going to be better than any like AI app that people have to generate hooks and stuff, right? So like this is then now I had it output its thinking process because I wanted to read it for my own self, right? So it tells me about what's happening for the shock. It actually looks through my results and figures out what's the most surprising, what problem has been solved, are there brief challenges. This is the psychology you want to get into before you write anything. So you want to know what's the impact to someone else, right? What are they going to feel when they digest this piece of information? And then what's the thing that you really want to convey in this whole thing? So you don't have to like get the exact output, but this is going to be your like sparring partner, right? You're going to take this type of thing and take it to the next level. The audience psychology tells me about the pain point. People with dietary restrictions, like desire transformation from manual research to AI assistance, current beliefs that are challenged like AI is limited with to digital tasks. So like we thought the AI was limited, but in my live stream when it ordered a pizza, this is when it's actually cloud code was jumping out of the CLI and it can write code to basically talk to an 11 Labs API and the 11 Labs API can talk to Trello and actually place a voice phone call and place an order for pizza, which is scary, right? Like that's a cloud code hook and we made that like just in a like in that live stream, right? And so this is kind of what like it dug through. It dug through the live stream. It's talking about like that and you're like, "Oh yeah, I should talk about that." Oh my god. Yeah. But I'm I know I'm going off on a tangent, but I'm The reason why I'm doing this is because beyond the coding stuff that we talk about here on this channel, I sincerely believe this is how you're going to leverage these AI systems. No one's talking about how you can make good system prompts. That's your co goal to having your like nice sauce for your app, right? So I feel like the before 2025 the era was source code. So if you had the code and you copied someone's code, you can have their app and because it took you months or years to write the app like that was your moat, right? If someone copied your code, you can actually sue them and you know there's all that stuff, right? These days AI can literally copy the app. So there's like no more moat there. The next moat is sort of these system prompts and these instructions. So taking knowledge like I have no clue about writing. I have no clue about writing hooks, but just dude, I'm I'm like learning like, right? I had a a AI generate a prompt. I'm going back and forth and iterating with it. Now, I'm able just to slam in these transcripts. And look at what it's it's giving me this rich data that I can use to now display back to the users to come up with better hook suggestions, to come up with better tweets, to come up with better YouTube titles. Like, all that stuff is going to go into my app. And this is it starts out with the transcripts, right? starts out with getting those things in, taking those next sets of transformations, timestamps, getting if you get timestamps in your videos when people search for keywords that are correctly spelled, Google's going to prefer your video because it goes to the exact timestamp where that happens and like the user will click it and then Google will keep ranking your video higher because the user got the information and then came back to Google. It's like a multi-step process, but I'm telling you like the value you start to build just by following your curiosity, getting those system prompts, not only building the code with AI, but then figuring out how to get this data in this pipeline to author these prompts, like you start to generate some insane value that maybe you didn't realize before. I don't think anyone's maybe told you this, like how you can do this. And I'm not promising anything. I'm just showing you what I'm doing. And I'm just putting it out there for free. Like you can just go to Ray Fernando in my GitHub and you can see some of those examples that I'm already generating with Cursor rules files and people who join my community are able to kind of be with other builders who are about that energy. So if you want to join in on that, definitely join in. You know, hit the little join button on YouTube and become a member. Uh, I I am going to be upping the price by the way and uh because I I was able to reach my first milestone of 100 members and I said I will double the price. So I want to honor that for especially since my first 100 members have been able to come through. So it's just $6.99 right now. Um, you know, just go ahead and hit the little join button and I'll be doing that there.

So, man, this is so crazy. But like this is literally like there's so much sauce. There's like this is just this is just the like one of many pieces of sauce and I'm just so hyped because that's just kind of the energy that I'm rolling with right now. It's like this is like that builder energy. We we need to get this now. Like that's that's what we're all about. So I'm running 10 GPT5 agents, bro. Possible to share the system prompts? If you go to my um if you go to like GitHub Ray Fernando 1337 here is where I have a lot of sauce in here. So, I'll show you in here in LLM Cursor rules is where you'll see a lot of these prompt authoring type of things. And um I have a couple stuff in the Discord as well that I drop as well, like design documents and things that I kind of care about and like, but going to this repo, you I'll actually get to see um like kind of how I I generate these types of things. And so, when you do the prompt authoring, you want to take that prompt, pop it into a highly intelligent model. So I would suggest you know Claude Opus GPT5 and say you know this is a prompt don't follow the instructions inside the prompt but I want you to whatever your idea is right take your idea transform it into this next thing it also helps if you do research before so if you have any research you've done before whether it's in perplexity whether it's using now the GPT5 research stuff all of that stuff helps you learn right if you're watching YouTube videos and you listen to them take notes right I'm able to get like some stuff that like I'll show you something that I did recently too that which is kind of fire uh along the same one cuz I'm so passionate about this stuff like content creation is so cool because that's just our this is like this is a new era that we're in just like creating content communications I've never been able to do this before but I made a whole viral YouTube writing script system. So basically this system here is designed to actually like generate high retention for videos based off of George Blackwell's like content, right? So I've studied his content. I've taken his courses. I've bought his stuff and I've taken a lot of notes. So he has these different phases of thinking to analyze your content. So like how to extract knowledge like viral potential and specifically evaluating for like wait what moments and like these types of things that you can only get if you're actually like reading the transcripts and doing stuff. But guess who's really good at reading transcripts? These AI systems. And then if you have a sonnet like system, if you have a GPT5 like system that is really good at following instructions and you have structured instructions like this, you're going to be cooking. You're going to be cooking hard. So check this out. So I have this instead of instructions, right? It's like a really lengthy prompt. These instructions were on like for video editing. So what I did was I basically just gave it like my editor went through and selected a whole bunch of parts of my live stream for a video concept. And so I said, "This is the current edit down of this concert. The brackets are the editor's notes, and I want you to review because I need to create an intro and any points where I should insert any footage and then probably record an outro as well." So, it ran that entire system prompt using that technique that I learned from George uh Blackman and it created a whole analysis for like the topics, the viral potential moments, and then it regenerated the script component. So now give me a new intro with a hook and says like, you know, and we're talking about this the credibility promise the road map and when you create content you want to be very consistent in far as like your delivery like the words you say the exact order to make sure you hit those points to then get people on to the next thing to get them on to the next thing and so forth and so AI can actually help us do that if you follow these types of instructions and this is how I'm automating like day-to-day life eventually, you know, once you get your video transcript and you want to apply by your writing style like, you know, now with my app, I'm just going to keep extending and adding value. So you see how it starts with one thing and then I just kind of keep adding it up and this is just prompting and this is where it's going to turn in from prompts to like actual real world apps, right? So this is absolutely fire, right? So um in this type of thing I think what I like, yeah, this is this is kind of this is crazy. And like here's another one I did for like AI coding workflow. Uh, I basically just took the script about like sub agents because I'm I'm doing one on sub agents and this is kind of like my new script that I'm going to actually record, right? So, I'm going to talk about like the first segment of content window which everything I've described here is all come from my live stream, right? Like these are my live streams where I talk about all of this stuff. Like this is like a three-hour live stream, right? Where I talk about context windows and a bunch of other things. And then I'm having it basically generate a new script for me to talk about specifically context window only, right? And it's got a nice little segmented um how I should actually format everything like all in here. Super cool, right? Like the five segments I should talk about like how to get people in there. Like look, a year from now, AI will handle this automatically. But right now, if you want to ship fast without the AI randomly rewriting the entire codebase, you need this workflow. This sounds like me, right? Because it has my whole example of how I speak from a live stream all in the text file that I just throw in cloud goat or in cloud right so fire and so the first revision didn't sound like me and because I said now I want you to sound like me it just did its revision there so I'm really curious to give this to GPT5 to see how it performs with something like this because this is so much gold and if it could do it for like basically a third of the price I'm all in bro I'm locked in. I'm so locked in right now. So, yeah. That's what's up, man.

All right. All right. So, like this is what's up. Oh, man. I really do have to go though, because I got to be in the city and I got to run out and it's already been 4 and a half hours and I've basically been lecturing to y'all. And we've been cooking. We got 3,000 people who have been in the stream. Got a couple people on Twitch. We got like 370 something on YouTube. We got 2,649 on X. Wow. Wow. Wow. Wow. All right, I got to dip out though for real. And I want to thank all of my members and people who've just recently become members. Um, Henri, who's really recently become a member, thank you so much for becoming a member. I really appreciate you. I definitely want to give you a shout out uh for becoming a member here on the stream. Uh, y'all are real ones. Fallen RV as well for becoming a member. I really appreciate that. Also, uh, David as well for becoming a member. Y'all are real ones right now rolling through the chat. Um, a lot of people kind of coming through the chat. Like we're going to have a really we have a really great community uh of people who are kind of hanging out here. Also want to give a shout out to my folks who have been doing super chats. So Alex Finn, you're the first one to get on. He's like, "Bro, where you at? Get on the stream." That's kind of what we were doing with this was like four and a half hours ago. This has been an amazing mega live stream. I've been here pretty much non-stop. So big shout outs to Allan, big supporter. He's also a member as well. Chords Maids. If you don't know chords maids, chords maids came through from I if you go back to one of the anthropic chats where they were going to limit the rate limit everyone. Chords maids is the first member of the three comma club not for being billionaire status but for generating billions of tokens and he was the original person who was so locked in that I felt like he was like misunderstood. So that's kind of the energy that we're rolling with right now. Uh big shout outs again to Damage who's rolling through with the super chats as well. The C giving us a heads up on the CLI when that that actually came out and I was super hyped about that. Uh and then also talking about Alex as well with this the super chat for joining the stream late making sure we got everyone taken care of. This has been a like absolutely insanely fire stream. Y'all are really cooking with me uh and able to stay this far. I mean we have some MVPs who have been here since the beginning of the chat. So yeah, y'all have been so crazy with that, too. So, man, this has been like absolutely insane. Also, P uh Prague Fan234, uh thank you so much for becoming a member. I really appreciate that. Yeah, coming roll through. So, for all the folks who just became a member, if you haven't heard, um just go once you hit join and so forth, there's a members-only post and in the members-only post on my channel, um only members can see them and then there's some instructions there for you to follow so you can go ahead and send me the information so I can get you locked in the Discord. If you are already a member, just go ahead and look for that post and you should be able to get in as a member and I'll make sure you get in. And if I don't like from your email or whatever and I can't tell that you, you know, what your YouTube channel is or what your YouTube channel name is, uh, please let me know because YouTube only gives me the your name and your channel name. So, if those don't match, then I can't really tell who you are. Um, but I'll probably reply back to you anyways. Also, any folks who've replied to me, make sure you check your junk mail. Sometimes some of those have been routing over to the junk messages. So, I just want to make sure you've gotten them. Uh cuz and sometimes some of the Discord invite links um expire. So, let me know and then I'll just regenerate another one for you so that you can get back in. So, thanks for the stream. Had to hop in and out for sure, man. Y'all are real ones right now. Uh thinking for 8 minutes, bro. What are you doing? What are you cooking, bro? You got to share the sauce. You got to let us know. New viewer here. If you had to guess, GPT5's IQ considering that GBT that 2.5 Pro is around one I I IQ is such a really bad way of measuring these models. Uh for a couple reasons, right? I think one of them I'm sort of triggered on the IQ question because like growing up like a lot of people thought I was really dumb. I was just kind of misunderstood. Like I had a lot of issues with reading and understanding things and speech because like I understood the world but I couldn't express it. So that is kind of also what's happening with these models. Um, a lot of the ones that rank really high in these benchmarks when people actually use them, that's why they call them vibes and that's why I have these live streams too. We're like trying to see what this intelligence of the model is coming out. That's one part. The second part is the integration and hooks in a specific platform. Right? So GPT5 for coding can be amazing, but if you're a general knowledge, what if it's really bad? Like I haven't used ChatGPT for everyday lookup stuff like flights and other things that I normally do. It could be really bad there. So, two different people, two different experiences. Um, and so it's really kind of hard to put that together. And IQ is such a small sliver of way of measuring things that it's actually not the best way to represent any type of intelligence to be honest. Even normal people, like normal people who are high IQ can be like the worst people for everyday tasks, right? And do you want them to be doing that? And even if they have an IQ, what's their IQ at at specific things, right? If they're only good at, you know, some underwater basket weaving, maybe people won't think of them as like being really good because they're like it's not like actually something that generates financial value, right? And I also don't even like the way that people are measuring like impact that way, too. So there's a lot of different things, but I would just say as far as like general feel, um, GPT5 has a lot of like it's already performing very well inside of Cursor and they've tuned the model to be really good. So I I call it GPT5 like in Cursor is the new power couple of Silicon Valley right now where Claude and Claude Code was that that power couple and it's still there. They're are very different camps, but it's going to it's going to vary on what you want to do. So my preference right now GPT5 and the reason why I'm leaning this way right now for GPT5 inside of Cursor with all these different configurations um is because there's going to be a new way of prompting that I can already see right I can give it a lot of really vague information. it's going to try to dig through and take its time to really truly understand that and Cursor doesn't really penalize you in a context window for that where Claude Code you have to be very surgical with your context window and it does really well. So yeah, it just depends on where you're leaning, what your experience level is, how much time do you have, like all these different factors, right? All are very viable. So I definitely I definitely mess with that for sure. Yeah. Yeah. Yeah. That's what's up, man.

Um, brilliant stream as usual, man. Thank you guys. This is so fire right now. It's It's so fun. I love streaming. I love showing up for y'all. This is so fun. Uh, I'm struggling to manage all 10 agents cuz they're making a plan first and it needs approve. Can't keep up, bro. No, no, no, bro. Just chill. Just chill. Just chill. You're fine with one, right? Uh, GPT5 Pro loves to think. Oh, I haven't done Pro yet in the UI. I'm pretty sure it does. I have one basic one-shot prompt and I thought for 20 minutes. Yeah, I guess that makes sense now. You know, it's like the GPT3 03 Pro. That makes sense. Uh, sounds like me. I had the same thing growing up. Yep. Yeah. Yeah. People just misunderstand us sometimes. Thanks for the great stream. Looking forward to the next one for sure. Uh, oh, sure. I understand. Then would it say GPT5 reasoning is better than 2.5 clots for reasoning? Yeah. 15 minutes. Wow. IQ means nothing. Yeah, for sure. Bro, you guys are cooking. Man's been cooking for 4 hours, bro. Yo, lock in, bro. Alex Finn, let's lock in. Yeah, I gotta make this app. Alex Finn, if you have this guy, he has a lot of really good content. If you want to know how to go from zero idea to building something, go watch his videos right now, he's 300k MR like in a series of weeks, bro. MVP of the chat. Go check out Alex Finn. He has a So, basically the the long and short of his videos is basically first is like take a problem that you have every day. I my problem was transcripts, right? I had this problems with like getting my transcripts even showing up with the right words. And later on through this whole process, I discovered how valuable that was. When people were pinging me saying, "Yo, your winds surf wave 9 thing is ranking on the top of Bing." I was like, "No way." And like, "It's ranking in Google, too." And I was like, "Oh, wow. I just started uploading all my transcripts, like, you know, on my videos all the time for my live streams." So, like, that's how you discover these problems, right? Solve problems for yourself, use AI to solve them, and then start charging for it. So, I used to like not want to charge for it because I wasn't sure it was a lot of value. And I just said, I need to start charging for this because it is providing value to me and there's someone like me who's a content creator that doesn't want to do all the coding and doesn't want to go through all the headache that I'm going through and you know, they just want something that works, right? So, I want to provide that value. And two, it's teaching me how to be a better business person, right? I don't come from a business background. Being an engineer, I was just like basically not coding and not doing that. I mean, I was just coding but not doing like all the business planning, marketing type of stuff. So, that's kind of where where I'm coming from uh in that. And like Alex has a lot of content right now on getting started with cloud code, getting started with these different AI systems, especially Cursor. He's going to probably have one on GBT5 as well. And he's going to definitely show you how to cook for sure. So, like he says, lock the in right now. If you haven't done that, make sure you lock in. So, for sure, uh, in the next stream, GPT V0, I think that will be fire. I love G. I love V 0ero. I got to go bathroom so bad. But I want to show you this like couple of create. Actually, let me show you a whole set of creations I did with Vzero. And I uh actually there was a Vzero meetup and they they've asked me to become an ambassador and I'm negotiating with them right now to figure out what that looks like because there's some real fire stuff you can create in VZ. So, um, you know that idea I got? Let's see. Got it. Okay. So, this idea, something I helped with my friend, this React swipeable setup. This is where I learned about code reviews because this is the app that I built in V 0. So, like that idea for Vzero for the swiping all came from here from a screenshot of my friend from Convex. So, he's like he gave it to me like he says, "Hey, Ray, can you help me with this?" or something. I was like, "Yeah, let me see what I can cook." And I just kind of came up with super fast in Vzero. I love Vzero for like all these different reasons. Um, this was the one I made for Yeah, this is like the real app. You know, you can do that type of thing. The other one I made was pretty fun in VZO was this thing called uh the lock in app. So, you can we're talking about lock in. This is beginning of August. We're almost done with 2025. Make sure you lock in. So, this is like an inspirational thing based off of this UI and like it was a whole meme. So, like this is a clock that you can have as an app that literally runs and it just tells you what month you're in and has like these motivational quotes at the bottom with these little things at the top. And so, yeah, it's basically I I should actually do more with this app, but this was generated in VZ like pretty much in one shot. Um, you can do really cool stuff in VZ. So, with GPT5, can you imagine now all these really vague ideas I have to just be finally visualized into something that's actually usable? Bro, the world is not ready for me right now. and the world is not ready for you. So you got to go tackle it, start using these AI systems, start cooking. So this is my favorite app that I got working right now. So this is going to basically allow us to use the open AI API, but using this really fun design system. So I found this design system on Vzero that I really loved cooking with. So you're basically can just drop an image in, for example, like I could drop an image here. And what it does is like you can just um like describe an edit you want or like have a specific style or like a background that you want or enhance it and it's going to do the face consistency thing. So you can change out the environment say I want you know to wear a hat or whatever and so it's going to use the open AI face preservation mode which is the new high fidelity stuff and basically you you'll see progress and things like here. So it's it's using this like really fun playful UI. Uh and this is all inspired by this thing called uh Bos brutalism design. So this is a design system as a template. You can take these templates and fork them and you can do stuff. So like the app that I made here which was super fire is somebody on X posted saying that they wanted to have this like Windows 95 or this like um I think it was like in 1990s or the early 2000s with the internet um where um like what was I reading about? So I guess back in the day people used to do these types of things where like they have like this um like buddy system. They can see who's online similar to Discord. So like this is like the version of Discord for people who were like in the early internet and they wanted to flip it so that basically you can use it with X, right? So uh I was like dude that's so fire. Why don't I just do like an X status cuz you can't really manage your friends on X as like buddies and stuff. So I like that whole concept and I just built this using existing design system. So this is the existing design system. It's called Boss design. Um and so this is the this called Boss brutalism design system. So this was the template that was there is literally just a landing page. So it said open in v0ero. I described it like in one shot. This was like how I wanted it. You know it said like online status away messages. And so basically what you can do now is like select your close friends or celebrities and you can even add to your buddy list. So you can say I want to create a new list of of friends or sports stars or musicians. So I can say I can create a new list of like AI bros. And then like with my AI bros, I can go in here and like start to add my my friends like uh like LeBron James, not an AI bro, but Crystal Ronaldo. Um, let's see who else. Elon Musk is an AI bro. We're going to save these changes. So now I can track them on X using this UI. So I want to see if they're actually online and stuff and see what they've been posting, what what are they up to recently. So you can actually see them here. And if you want to update your status, you just hit update status and say you're like you're away or

Something like that and you're like, "Bro, I'm cooking live right now. Lock in." And then like you know, update and like you'll have your thing and then people can kind of see your little status updates like and you know this will be like your like more condensed version of X but for your close friends. And so I think that's a fire app and I can now take this design from Vzero and just put this into cursor now with GPT5 and cook. And I think I should do that now because I'm so hyped with all the latest stuff. And that's Vzero in a nutshell, bro. Like I cooked this before like when was this? God, this is a while back. All these chats go pretty far back. But um now with GPT5, bro, you you don't want to see me with this. This is going to get nasty, bro. This is going to get nasty. Oh my god. GPT5, you're not ready for me right now. But you can cook. You can cook with Vzero. Vzero is is a fire design system. So, we're Yeah, like I like it and I think we're just going to get really lit with this for sure. Like 1,000%. Oh man. Yeah, what a stream.

Hey, just joined. Any impression conclusions from earlier stream about GPT5 models? um cloud code especially for implementation against planned tasks and docs. Dan GBD5 is really good, very thorough. Um you can use the high if you want to have it do some implementations for you and I'm very impressed. I I would I like the user interfaces especially inside of cursor and I think it has a lot of great potential and a lot of potential it does perform very well. So um yeah that's what's up man. We need to get community build stream going. So yeah. Oh, that would be fire. Like we like actually build this app or something in the community or something or whatever we want to do. I I agree. Uh one is testing creating a repo that uses combination of playright videos, screen grabs, and code definitions to convert legacy frameworks into React in some agentic sass. Bro, go cook. Let's go.

Is this still alive, bro? Yes, bro. We are still in here right now. 3,000 people deep. Oh my god. A lot of people on X right now chilling. I don't even know if they're actually there. I think the X people are bots. Oh, you know what it is? X only shows you your views. So, it doesn't show you who's actively online. So, if anyone's from X, if you just drop a five on it, I I'll hit your comment live. I don't even know how many people are actually watching on on X.

Uh does it uses GPU to transcribe? um like the earlier versions did, but now I basically uh set up some websockets and I stream um I do some like really clever chunking on the client side, which is like your phone or your browser and then I tap into like your local media renderer to um run a bunch of processes to like like do the the the conversion and then I stream those audio bits up to like an endpoint and then uh that's where the conversion all happens. So I have a couple fallbacks for my endpoints. So like one of them is Grock, one of them is like OpenAI um and all these different providers and I try to do it as fast as possible. And um I'm actually I'm actually rewriting all that stuff right now. So that's all going to change because I'm trying to do this new agentic workflow where I can just send up a file and have it batch processed and then I can do this other stuff like I was showing you in parallel like maybe kick off GPT5 and have it do like you know the transcript stuff where I'm looking through different topics and then like I was telling you about like you know generating titles like I can do all those things in these really cool workflows I can do in convex that I haven't been able to do if I if the whole thing was just basically uh in my previous method. So, or I might split the app into I don't know like I have it just depends on the demand and what people are actually asking for. Um, but as of right now, I don't really have like like thousands of users, you know, I just have like a handful of them and um they're kind of driving the the features right now. So, I mean that's the thing like I could have a hold of a lot of stuff that I want to build, but uh I'm kind of going in different sections.

So, bro, this guy's cooking. Did you test curs? Yeah, we we got to test that earlier. It was it was pretty chill. Yeah, it needs some work though. still an alpha. An aim version of that design would be fire. Yeah. Yo, do you have gotten your catheter? No, bro. I need to get one right now. Yeah, we tested the cursor CLI. It gives free GPT5 for sure. Yeah. Yeah, that's what we've been doing. Like the even cursor, right? Yeah, there's a cursor CLI available. They dropped it like 3 hours ago. We've been on it, bro. Yeah, we tested it and it was it was it was pretty good. like it wasn't like it's still alpha and I prefer the output from sonnet in that CLI and the sonnet one is actually the one that's really fast. So if you use it's kind of funny if you use sonnet inside of cursor CLI it's actually faster than the cloud code because of the hooks into the OS that cursor team has with their craft engineering and stuff like that. That's my my quick two cents on that. Um, my daily driver though is going to be GBT5 inside of Cursor. And um, I mean that we, ladies and gentlemen, are in a new paradigm right now. And it is literally like this full-blown meme right now. Oh man. and drop drop the five in the chat if you're if if you're going to if if GPT5 is going to be your daily driver or or drop like CC in the chat if you're an incloud code. I'm trying to figure out what camp you guys are in right now. Have you guys try got to try both or where you guys at right now? Yeah. So, that's what's up. Yeah, I've been trying GBT 5 V0 and it's been cooking for a hot minute. Wow, I'm so hyped for that. Oh man, that's what's up. Uh, give me the cursor plan. I can't use Opus models, bro. Bro, what did you do, bro? Yo, Mike got five on it. RIPCC. Oh man, y'all y'all are too much right now. All right. All right. All right. Cool. I got to roll out though. I I have to drive to San Francisco and I got to go meet up with Riley cuz Riley's doing a meetup. I I don't know. Like, you know, Riley. Let's see if I can find Riley. Damn, my messages are blowing up. Riley Brown. Riley Brown. Riley Brown, content creator, super he founder of the Vibe Code app, is hosting, you know, he's the one that teaches people how to do AI stuff. I have been wanting to meet with him since he moved to the city and then we're finally going to be able to meet and so this is going to be really fun. There's a meet up with him and this other guy named Yoni and Yoni is an amazing uh content creator as well. He's been posting every day, bro. No way. Oh, that's what we should have made is a Flappy Bird game. Bro, GBT. Wow. Okay, I'm getting way too distracted. I got to go, too.

So, this is interesting. How well does it do for mobile apps? Wow. Create a retro productivity app. Huh. Interesting. Wow. Yeah, he has a lot of cool cool content around this stuff. So, definitely go ahead and check them out. I'm just kind of getting like all distracted. What's your main goal when it comes to using AI? Is it to build something profitable or you doing it for fun? general question not talking about today.

So, uh, Bros, bros number 12. Basically, my goal with AI right now is to to generate freedom. Like that's my overarching like theme right now. I'm trying to use AI to generate freedom. I used to work for a corporation and I got so sick with long COVID uh that I wasn't able to go back to work. Like I literally I was after being on leave for two years. Um, they just said, "Hey, let us know when you feel better and let you come back." I'm like, "Okay, bro." like I was with you guys for like over a decade and had huge huge impact. Um and so I feel like okay with the age of AI it just was born right as I was getting over my my long co and stuff and I realized that there's so much intelligence and what's going to be happening and so many different realms being transformed that this is actually my time to lock in and use my existing experience to try to do something with this. Right? I've never been an entrepreneur in my life. Right? Like this is my first time doing this content creation first time, right? Like I used to be a photographer like back in the day. I had so many different jobs, right? Like I I dropped out of school. I don't have any like engineering degree, right? Like am I cooked? No. And so my goal is to just use AI to help me generate freedom. I'm building these systems to see if I can generate value for myself but also generate value for like normal everyday people. everything that I've been reading about and I've been seeing people who are successful like even very young people with very little experience like they blow up right sell millions of dollars an app and for them it's really simple they're just like yeah bro I just take a picture of a thing and I just send it to GBT and it gives me the calories and that's what I use and I'm like well what about the accuracy like it's like super inaccurate as far as number of calories and you're like well they don't care cuz the FDA says like this is the variable and the variable for FDA accuracy is like really freaking large and I'm like you can just do that is like that's not like ethical or something and they're just like don't care. And I'm like hm but you made millions of dollars. They're like yeah okay that's cool. Like that's not how I roll. Like I have way more ethics than that but I feel like there's like kind of a mixture here. It's like okay so people who like don't really like people can make a lot of money just by showing up. And that's typically like you know like I feel like that's business 101. It's like show up, charge the money, and then like figure it out. And they ask questions later and kind of learn as they go, right? Like that's just entrepreneurial 101. If you are overthinking everything, I've learned, you kind of make less money in a funny way. So I sit somewhere in between, right? Like I have a lot of care about the stuff that I do. And part of the favorite thing that of where I used to work is the fact that there's a lot of people who are these like craftsman level type of thinkers and and doers and things, but that's a really isolated environment and you're having another section of the company do all the marketing and everything for you. So you don't ever think about that. And so I'm basically now tying the two together and saying how can I enter this era of AI very fastm moving but also like build products solve my own problems and then the problems that I solve offer that as a service to other people and and there is no better time than do that than just right now right like we're in this era so what's going to make a difference is not writing the code but actually providing the service and so in my previous jobs I was known for having extremely good customer service because I actually cared about people. And at the end of the day, what I learned was that if you care about people, you will just go a long way and people will follow you no matter what. They want to be with you because you cared about them and whatever stuff was going on. And so the brand benefited from me caring a lot from different people because, you know, people would show up to Best Buy and I would fix their computers and save their life or whatever different things, right? And, you know, I would just be there. I'd spend that extra time because I cared because I treated them like family. That's kind of the way I do around here. Like I just I spread love. I spread peace. Uh, I'm about that good energy. So, that's what I do. And I typically like to stay away from people who like to be very like, you know, who aren't about that energy. So, as soon as people start to be about that energy, I kind of like really just walk away because there's no time for that. There's only time for really good energy. So, if you're about that energy, definitely join the community. Uh, Ray Fernando 1337 on my YouTube channel. You can just join in. It's $6.99 right now. Get into that Discord before I double that price. And I will see you guys in the next live stream. Peace out y'all.