Transcription
Hey everyone. Welcome back to another episode on Open AI's Codex.
Now, when you hear the word Codex, you probably think, "Just another CLI tool, kind of like Cloud Code, right?" And like most people, you might have just given it a shot, generated some code, and thought, "Cool, but it's kind of incomplete. I'll just go back to Cloud Code."
But here's what nobody really shows you. Codex is more than just its CLI. It actually also has a full IDE integration, as well as its cloud web interface. And when you connect all three of them, that's when the magic happens.
In this video, I'll show you how you can combine all of this together to build not one, not two, but three full versions of an AI-powered Reddit sentiment tracker. And the best part is, is that I barely lifted a finger. Codex was cooking in the background while I sipped my cup of coffee. I could track the progress of everything right here on my phone, and whenever there was a bug, all I needed to do was to just tag Codex and GitHub and let it handle the heavy lifting in the background. This workflow with Codex is a total game-changer, and I can't believe that not many people have been talking about it. So you'll definitely want to pay attention because this is how you will truly level up your AI development workflows.
I've made the resources for today's video as part of my Insiders Cloud. This is a free insiders list where I share multiple things, including early previews of lessons, as well as access to private GitHub repositories. And if you want to take things a step further, you can get hold of everything, including the source code, the prompts, as well as the final three applications from today's video as part of my AI Builder Packs. You can get hold of all of these in the link in the description below.
Okay, so I was very inspired by this thing called the C-Clater, which actually tracks sentiment across, uh, various different subreddits related to ChatGPT, Claude, and Gemini. So I thought I wanted to build one that was focused specifically on Codex. And let me just show you the three versions that Codex was able to build for me over here.
Here is version one. As you can see here, it does something very, very similar. It's able to give you a summary of the sentiment across these four different subreddits here. You can also swap between the duration, and it also gives you top keywords and various different things about how it was able to grab hold of the data. It has a methodology section, and as a nice bonus, it even has a dark and light mode as well.
This is the second version that you see over here. And as you can see, it has very similar functionality. The same thing around being able to jump and dive into different subreddits. But then the nice thing about this one is that you can actually jump in and actually see what actual posts contributed to the sentiment.
And then finally, this is the third one. And my favorite. Very similar functionality once again. I think the overall design and user interface with this one is just a little bit nicer. One of the key things here is that not only does it allow you to select the range, but it has a range picker as well. And the great thing about this, all three versions of this app were created from one single PRD. And because of the power of Codex in the cloud, they were running in parallel, nonstop for 30 minutes, and I was able to get all three versions for me to iterate upon.
Now, one thing to caveat is that these initial versions were still using mock data, but the whole point is to demonstrate to you how you can get to a stage where you have multiple different iterations of your product created by AI so that you have a great foundation for you to iterate upon.
Now, without further ado, let's get started with today's video.
Okay, so first of all, I'm going to be creating a brand new repository. This is the way in which the Codex Web and Codex agent will be able to perform its cloud functions. And I'm just gonna make this private for now. I'm just going to clone this repository into here. Then the next thing I'm going to do is launch Visual Studio Code with this folder. And for those of you who are familiar with how I like to work, I always like to start out with a PRD or product requirements document. I've been actually making use of Vercel AI to do my PRDs for me. And the reason for that is that I've actually been using Vercel for the research and planning related to my episodes. For example, here's me actually using it to study the changelog from the Codex CLI. And because it has built-in Perplexity Pro, it was able to actually study the changelog very well and then give me a good set of pointers that I could then use. And because of the fact that you have access to all of the state-of-the-art models like GPT-5 thinking, it is actually very robust. And one of the key reasons why I really like how Vercel works is that it's more than just a chat interface because, as you can see here, you can actually use it to generate a step-by-step plan to help you plan how you're gonna make use of the information that it provides for you. And because I had all of this context as part of this project, I actually used this to create the PRD for today's app. I even gave it the URL to the C-Chronometer app as a reference, and after it asked me a couple of questions and a bit of back and forth, it was able to give me a finalized PRD for this Codex sentiment monitor that we are using for today's video. So Vercel is extremely powerful. I've been using it a lot, not just for my planning, my episodes, for doing research, but also like in this case, actually generating the PRDs that I'm using for the applications that I build. If you want to try it out for yourself, check out the link in the description below. And if you use this code that you see over here, you'll be able to also get 30 days of Vercel Pro completely free. So a big thanks to Vercel for sponsoring this video.
And now that we have the PRD for the application we're building today, it's time to hop back into VS Code. I am just gonna copy this and then create a PRD.md and I'm gonna just pop it in here. So what I'm gonna do here is add everything. And then I'm going to push this to main. If I refresh, and you can see it's added in the PRD document straight into here.
So now the next thing I'm gonna do is that I'm going to go to Vercel, add new project. The Codex-ometer demo project is not yet available because I usually like to make sure that I only grant access to specific repositories rather than every repository. So I'm gonna go into here and do Vercel. Demo, save it. Once this is done, you can see Codex-ometer demo is over here. I'm going to import it. I know that this is going to be a Next.js app, and then I'm going to do deploy. The deployment is going to fail because we actually have no code written. Now we are set up with Vercel and now we know that every time we push a new branch to this deployment, we're gonna be able to get a preview so that we can see everything running.
Okay, so now we have the PRD here. What we're gonna do now is I'm gonna fire up Codex, right? So let's just go straight into it here. Alright. And you will see another thing here is that I've got it defaulted to danger, full access, and approval mode: never. So what I'm gonna do here is review the PRD and then come up with an implementation.md where this will be implemented using Next.js 15.5.4 and then SHA CM 2.1.0.3 and then React 19. And I'm also gonna say, "We will use Supabase as our backend." Let's see what this comes up with because that will form the basis of an implementation plan. We'll let this kick off and see what it gives us as a result.
Okay, it's completed its task. What we have here is an implementation plan using Next.js, React, and Shad CN together with the Supabase backend. It also knows that we are using Vercel for hosting. I'm just gonna quickly look through this to see what it's come up with. You can see here, it's going to look through and understand which subreddits it needs to look at, and what Reddit items are needed, along with the metrics to calculate the sentiment scores. It's gonna give a technical outline over how it's going to ingest the data, and the server-side implementations, as well as the backend implementation and the frontend. So this is a pretty comprehensive implementation plan. And so what we're gonna do is that we are going to see how far Codex is actually going to be able to get with building an app for us, using both the PRD and the implementation plan.
So before I proceed, I'm gonna commit this into the repository. And another way of doing so, rather than using the command line, is that you can just hop over to this source code view over here. If you're lazy like me and you just want it to commit message for you, you can do that. Add this and stage it, commit it, and then you can push it. And if you're not familiar with how Vercel works, if we jump back into Vercel and go to this Codex-ometer demo, it's actually kicked off a new deployment because we just pushed a new commit over here. A comprehensive implementation plan. And this is why I really like Vercel and why I wanted to get all of us set up because now every time you make a commit, it's actually going to be able to try to build the application and deploy it onto the web. And you want to be in this mode all the time. Most people love to stick on localhost and never break out of it. And most of the time, the problem isn't when you're working on localhost, but when you actually try to put onto the web and deliver it to your users.
Now, this is where the magic happens because I'm gonna show you how you can create an MVP of this application using just the PRD and the implementation plan. The great thing here is that we're not gonna create just one, nor are we gonna create two. We are gonna create three different versions of this MVP in parallel, that will then give you the opportunity to review which of these you like.
I'm going to just go to Codex within Visual Studio Code. If you haven't already installed it, you can just go to extensions over here, search for Codex. And you want to make sure you get this Codex one by OpenAI. There's a bunch of other fake ones which you just want to avoid. This is the one you want. Install it. And the first time you get it, it'll actually be sitting over here. What you want to do is drag this and pop it over here so that it matches most other IDEs like Cursor and GitHub Copilot.
So I'm gonna show you something. In the past, what you would usually do is something like this: "Based on the PRD and the implementation, build a great-looking app that fulfills the requirements. Keeping it simple, and make no mistakes." Now, usually what you'll do is that click this off, and then it'll just build a version for you that you have over here. You'll just wait, and then when it's done, you're gonna get a version of it, test it, and then you roll.
Now, this is the power of Codex over here. What we're gonna do here is that we are gonna click on this and we're gonna do "Run in the Cloud." You will need to make sure that Codex has access to their GitHub repository. Go to chat.openai.com/codex. First thing you wanna do is click on this one and do "Manage Environments." Do "Create Environment," and then this is where you want to link your GitHub repository. As you can see here, I have Summitter and Insiders Landing Page. These are the ones that we've been using for testing. What you want to do here is click "Connect GitHub Org." Go back in here, just like what you did for Vercel, and add the repository that we actually just created. So this is gonna be this one. I usually like to be very explicit about which repositories I am granting access to. Once that's done, I'm just gonna refresh the page one more time and it should pick up the fact that it has that. You can see Codex-ometer demo, okay. And then code execution and everything. I would recommend that you do "Agent Internet Access" on so that it'll be able to run some commands. This is very useful because otherwise it gets blocked on running and installing some of the dependencies. And anyway, this is a test application. You can always lock it down later. So I'm gonna go "Create Environment" over here.
So once this environment has been created, you are ready to switch back to your Codex CLI. You can click on "Cloud," and you will be able to see that there's a Codex-ometer demo. Here is where you can switch whether you want to continue from local changes or what's already in the repository. I like to work off what's in the repository because you don't accidentally contaminate what's already there. But because this only contains two files at the moment, there's no difference.
Now, here's the magic. You get to select the number of parallel attempts for Codex to make. So rather than ask me to do one version of this, switch this to three versions, and then you'll click on this cloud icon over here and let it go. And you will see, this is the amazing thing that I can't believe people are not talking about. This is very similar to Cursor's background agents, but the UI and the UX of this is so clean. It's right in your editor. It's tied into a nice web interface, which I will show you right over here.
So you can see here, it's created a new task. I'm gonna click over here and you can see you are viewing a Codex Cloud task. And you can see here it has three different attempts it's going to make. And you can actually go now onto your web and you will now be able to start seeing all of this running right here in the web. It's amazing because now you know you have everything not just in locally running within the Codex plugin within VS Code, but you also can review and see how everything is running on the web. This tight integration between cloud and your local machine is just so clean. And basically, this is almost like you have three engineers going off to build three different prototypes for you right away. You can see the logs of things that are running. You can see it's created a Next.js app. In this version, you can see that it's also gone on to add in a bunch of other dependencies in this other version. And the great thing about this is that you can go and grab a cup of coffee now while you let Codex cook. Because this is the beauty of using background agents. You don't have to be physically standing there typing in and working and giving it prompts. You're gonna just let it go and run off and do its thing and come back later. I have the ChatGPT app on my phone. And the best thing about this is that this is going to be running on your phone as well. You'll be able to see the progress of this as it continues. This is a very underrated workflow that you have over here. You have a sync between your laptop or workstation, the cloud, as well as your phone. You are able to get something like this working in Cloud Code, but not that clearly. You actually have to have a bunch of third-party dependencies and different things. I just feel that this is severely underrated, having this tight integration between your devices and having things running in the background as well.
We are gonna leave this running for just a couple more minutes to see where it gets to at the end. And then right after that, I'll show you the workflow of how you will continue based off what Codex is able to give to you.
Alright, so this has been cooking for about 13 minutes now. I'm just gonna show you how things are looking so far, and you can see all three versions are still going. If you wanted to look at the logs, you can get a detailed understanding of how Codex is iterating and fixing issues as it goes along, while it then continues building the rest of the application. So you can see here that it's actually being able to self-heal and fix issues that it created. And this is great because I've just gone off and grabbed myself a cup of coffee while three different versions of these have been kicking off. I think this is actually the killer workflow here, folks. I could have easily switched this to four versions and had four different versions running. This is just amazing and I really don't understand why nobody's talking too much about this. When Cursor's background agents came out, there was a lot of noise, but when this came out with Codex, it almost as if nobody really looked at it. Everybody's been focusing so much on the Codex CLI, and don't get me wrong, there's been so much that has made it absolutely amazing. But at the same time, OpenAI have been really cooking with the amount of stuff they've been putting in here. And this is still on the $20 plus plan, right? But anyway, we'll just continue to let this run to the end and see where we get to, and we'll draw back once it's been ready. Then I will show you how you're gonna be able to integrate this into your workflows, test out what Codex has been building for you, and how you would then be able to take it to the next level.
So now Codex has completed with all three attempts at building this application. As you can see here, it took about 30 minutes to complete all three runs. And because it's running in parallel, if you were trying to do this sequentially, it would've taken you an hour and a half. So you've already saved yourself a lot of time by relying on Codex Cloud.
Now that you have these three versions, the way that you want to do this is that you wanna figure out how you can actually test this. All right, so if you simply jump back into your IDE, you will see that the Codex IDE has actually synced everything that is happening in the cloud and on the web with your editor. Over here, you can see the three different attempts. And for each of these three different attempts that have been made, as well as the summary of everything that has been done.
All right, so how do you go about testing these changes? It's very simple. So I'm gonna open up a new terminal over here. If you actually scroll down, you will see that you see something called "Apply local changes." Because I tested this before, it's now looking at "revert" and "reapply." So I'm just gonna do "reapply" here. And you can see it's now pulled in all of the files that were done as part of this attempt. So now what you can do is you can do `npm install`. This will install all of the dependencies that you have over here. Okay? And then now we can do `npm run dev`. Okay? And now it's going to be running on 3001. So let's take a look at how attempt number one looks.
Alright, so it actually has. So you can see here it says "Reddit Sentiment Pulse," right? It monitors the Codex CLI and IDE conversations across the community. And you can see I have a weighted sentiment. It's pretty neutral. The volume as well as the positive share. I can switch between the different time intervals. And then I can also switch between the individual communities as well. And then you can see as we go down, we get other things like a daily drill down, as well as the top keywords that are involved with it, as well as a quality and transparency log. So I would say this is actually pretty good, but there are several things I don't really like about this one. First of all, is using this very purple hue, and overall it feels that it's not very cohesive. It's fulfilled the requirements. I can even export the CSV going on here, but it just overall doesn't feel super top. So I'm gonna click "revert" over here. Okay. And you can see it's removed everything. But I generally like to do a clean, just to remove anything like the node_modules and the node_modules because they can conflict with each other.
And let's take a look at attempt number two over here. So I'm just gonna do "reapply." You can see it's pulled everything in. Once again, I'm gonna do `npm install` just so it gets all the dependencies. Alright? And then once that's done, `npm run dev`. And now we get to check out how attempt number two looks like, right? So let's go to 3002 in the browser.
Here is number two. This looks a lot nicer to me at first glance. It has given it a name called "Codex Sentiment Reddit Analysis." And it tells you what the purpose is. It has very similar functions. For example, I can switch between different intervals and see the sentiment changing. I'm able to see the overall sentiment. I can switch between different subreddits to drill down into each of them. And there's also a keyword spotlight. But the nice thing that we have here is that we can actually. I noticed that it has this day-by-day sentiment analysis, which you can actually click on "View," and then it actually breaks down how it calculated this and what confidence level it has, which is great. And it allows you to jump straight into Reddit to see what the community post was about. But remember, this is all mock data at the moment, so it's not going to be linked to actual data just yet. And over here you can see the methodology section, very similar to C-Clater. And it explains a little bit about how it calculated sentiment. Overall, this looks better than the other one, a lot more cohesive. However, there is a bug where the contrast on the pills on the selection isn't looking great. And I will talk about how we're gonna fix this up later in different ways. But for now, I think this is looking a lot better. But we definitely want to check out attempt number three.
Okay? I'm gonna revert the changes. I'm going to do a `git clean -x -d -f` so it clears up all of these temporary files. And then I'm going to attempt number three. I'm gonna do "apply." By this point, I'm gonna do `npm install`. Alright? So it's gonna get all of the dependencies and then `npm run dev`. Okay, so it's gonna be coming up on 3002.
And this is 3002. This is looking a lot nicer. It has a little bit of that purple hue, but we cannot worry about that. But we can see that it's a totally different design, but the core functionality is there. I can drill down into the different subreddits so I can see the sentiment versus the volume here, which actually I kind of like the way that this is being presented. And down here we can see the keywords that we have, the insights. And I think this daily activities table is actually a lot nicer. And if I click "Explore," very similar to number two, you can actually see how and what posts contributed to we got for that day. Great. I noticed a few things here. It doesn't have the explanation of the methodology, unlike number two, but it does have the ability for us to have a custom date range this time and to export the CSV.
I've shown you how you've been able to leverage background agents to create different versions of your app. And I think this is very valuable because if you had just focused on creating one version of the app, the first version was quite underwhelming. And then you have to revert that, or you have to iterate on that. And I don't think that's the best way. Whereas this one, you had it all running in parallel. Within 30 minutes, you had three different versions.
Now let's move on to the workflow that you can take from here in order to start iterating on your selective design.
Now I wanted to show you the workflow in which we can utilize GitHub, Vercel, and the Codex Web interface to continue iterating on our project. I'm gonna focus on attempt two over here. So I'm gonna "apply," alright. And I'm gonna do `npm install`. The reason why we wanna focus on this one is because this one had a bug in which the pills here had the wrong contrast when it's selected. So of course, you can very easily just go over here and say, "Fix the issue where the contrast, et cetera, et cetera." And you can also do it within Codex CLI as well. But rather than doing it over here, I wanted to show you a workflow which utilizes GitHub, Vercel, and the Codex Web in order to do these things in the background. 'Cause this is a small change and you really don't want to be dealing with all these small changes. You want to be spending most of your time on bigger changes that relate to the functionality of your app.
So what I'm gonna do first here is I'm just gonna create a new branch. So I'm gonna do `git checkout -b` and I'm gonna call it `attempt-version-two`. And then I'm gonna commit everything and I'm gonna publish the branch. Once this is done, I've actually just hopped back into here and you can see "Version two" is over here. So I'm gonna create a pull request over here.
Now what's going to happen now is because we have linked everything up, it's going to first try to build and deploy this project. And this is the reason why. This is the reason why I got my project working with Vercel because straight away you will be able to have this running in the background. You can always jump into Vercel to see the progress of this running. You can see here it is installing the dependencies and building the project. And once this is done, because this is happening on a branch, it will create something called a "preview deployment." Okay? And you can see this is now running live on the web. If I open this up over here, great. Okay, so you can see now this is running on the web here. It's no longer running on localhost. You no longer have to keep switching between projects on your local machine. Rather, you rely on GitHub, Vercel, and Codex to actually continue to make changes while you focus on more important stuff.
All right, so let me, just demonstrate this to you here. As we know, we have this issue here whereby this community section, when the button is selected, doesn't look right. So I'm gonna take a screenshot and copy this. So what you do is go over here in your comments and tag Codex and say, "There is an issue in the communities section when selected. The pill has white text against the white background, which makes it illegible. Fix it and make no mistakes in the simplest way." All right, I'm just gonna place this here, which contains the image. So I'm just gonna comment right over here.
And so what you will see here is that because I've tagged Codex, did you just see that the emoji just came up? And this means that Codex has discovered that you wanted to do something. So I'm gonna go back into the Codex Web interface. Can you see straight away here? There is the task that has come up: "Fixed pill text color in community section" with the right repository and the right branch. So I'm gonna jump right into here and you can see over here. It's going to look through and decide what it needs to do to fix it. This is all going to be happening in the background, and it's gonna be able to fix the issues without you having to do anything on your local machine. This is a really good way because you can actually be doing this on your phone. You just have to tag the Codex agent on your pull request, and it'll just be able to go off and work on the task. You don't have to handhold it in this case 'cause it's just gonna go ahead and do it. While that's happening, you're free to continue working on your project and on perhaps a slightly more interesting or complicated task that you have over here. You didn't actually need to have access to your Visual Studio Code IDE or the CLI. You could actually do it right over here.
So now we know that version two has been sorted. What I'm gonna do right now is reset everything back again because I actually prefer, alright? And if you look over here in your IDE, and remember when I say that I really like the tight integration that we have. Look over here. "Fix pill text color in community section." This is the cloud task that was kicked off, and now you have it right in your IDE. And so what you can do is that you can see here that it's actually made a change that was required. And let's just apply it straight away. Alright, so it's applied it straight away. We put those changes in, let's run `npm run dev`, and then we'll take a look whether or not it's managed to fix that problem we just saw. I'm gonna open it up on localhost up here, and here we go. This is looking much better now, doesn't it? It actually has fixed the contrast issue that we saw earlier, which is great. And this was all happening in the background. We didn't need to monitor it. We didn't actually need to stop our work in order to fix this silly little issue.
So I've just shown you how you could pull the changes in, but just imagine you were actually working on another task. For example, all things considered, I really like the third attempt solution. So what I'm gonna do is apply the changes to my repository. Alright, so now when I do `npm install` and `npm run dev`. Alright, so now what's gonna happen is that if I refresh this page, it's going to be showing us this version that I prefer a lot more. While it was fixing the bug, which I think is very useful, you may not want to be interrupted to switch back into the other branch to test out those local changes. And this is the reason why connecting your project to GitHub and Vercel is very useful.
Alright, so let me just show you how you would do that. So you can see it here. "Fixed pill text color in community section." It is a Codex Cloud task. Let's just click on it and this will bring up the entire task related to it. The most important thing to pay attention to here is that you can go to the top right-hand corner, click on "Update Branch," and this will make this change as a commit to the existing pull request. And there's two ways you can do it. You can click on "View PR," which will then pull up your pull request. You can see that now it's just made a commit. And you can see what's happened over here. Four minutes ago, it actually told you what the fix was. But you have to always remember to do "Update Branch" in order to actually ask Codex to make a commit that will fix up the issues you have over here. And now with this fix, remember right now we are actually testing version three over here. So how can we quickly check how version two is looking? Well, don't forget, this is the reason why we have it connected to Vercel. And don't forget, Vercel is now going to be building this. And you can see it is actually now built the change that you've made, and now you can preview it straight on the web with the fix in the browser.
So now you have a working version in the browser with the fix applied, and you didn't need to switch from your ongoing task that you were working on. And this is the reason why this workflow is extremely powerful. It feels like you're multiplying yourself. You don't have to handhold Codex through every single fix. You give it some instructions and off it goes while you focus on something else locally.
I hope this has showcased the power of utilizing this entire integrated environment with Codex in the CLI, in the IDE, as well as Codex Web. And when you pair it together with GitHub and Vercel, it just gives you a flow that multiplies your efficiency.
So what I'm really gonna do very quickly is that I'm going to do the same thing that I did over here for all three deployments so that we can actually see all three deployments running on the web. I've gone through all three of them and created different branches coinciding with the different attempts, and pushed it up onto the repository. And you can see that they're all sitting over here. Within Vercel itself, you can see that it's actually going ahead and deploying each and every one of them. Now, there is an error here with version one, which I'll fix shortly. But the great thing about this is that if I go into Vercel and click, you can see that everything is now running on the web. And this is great because this means that we can easily distribute this application to other real users. And we've gotten out of localhost. And let's just take a look at the other one over here. And you can see this is version two over here. This is the workflow that I really think is very useful. You get things up and running, it's all running on the web, and each time you make changes to your branches, this also gets updated.
Now let's take a quick look at what's going on with this one over here. So attempt version one has this error where there's no file directory, `lstack`, `path`, `one server`, et cetera, et cetera. Now this is really interesting because when I actually tried to replicate this locally, there is actually no issue. So what I'm gonna do is I'm just gonna jump right in and I'm actually gonna switch the model to a high reasoning model because this probably isn't going to be easily determined without looking over. So I'm gonna just copy these lines that we have over here, and I'm gonna throw it in here. And then I'm gonna say, "This build error occurs only on Vercel's build in the cloud, but not when I run `npm run build` locally. Identify the root cause and fix it." Okay? So this is what we're gonna be doing. Now I'm using the Codex CLI to go ahead and fix it. I could also switch this to GPT-4 and then type in the same command that we see right here. But for now, I want to rely on the Codex CLI to be able to do this because in a lot of cases I find that the Codex CLI can be a lot faster, especially when it comes to a very targeted error that we're seeing right over here. And the reason why we're using a high thinking and reasoning mode is so that we can actually see what GPT is actually thinking about as it tries to get to the bottom of this. Alright. And you can see that something to do with the routing and how it gets exported. So it's actually gonna be experimenting and looking into several of these issues over here. And it's now come up with a plan, and then it says that there's a tracing bug found triggered by re-export the group page. So it's gonna try to see if we can resolve it over here. All right. And it says that a root cause has been found. And so let's actually see if this actually is working. So what we can do now is that we can just pop this in and then I'm gonna ask you to generate me a nice commit message. And then I'm gonna simply change this. And then it's gonna push it. Great.
If you jump into Vercel over here, you will see that now a new preview deployment is gonna be coming up for the version one of this application. If we actually look through the build logs, we can see that it's making some good progress. We'll just wait for this to run till the end, and hopefully that would have solved the issue that we saw earlier when it was broken. And it looks like it's actually completed now. Great. So it says the deployment is done. Let's jump back in and you can see we have version one of this application running. Let's open it. And here you go. This is the first version of it. Like you can see, even shows you the methodology, alright? And the guardrails that were taken. You could think about this, and I dunno whether you notice, if he actually take a look here, he even has this light mode and dark mode as part of its implementation as well. If you actually just look through the other two, they don't actually have this. So let's actually just take a quick look. So this is attempt two, definitely no dark mode, but it does have a methodology over here, which looks really nice and then a very different layout to how it works. And then this is the third one, which is my favorite by far. It doesn't have a light or dark mode, it just looks like that by default, but I the way that this one has the ability for you to have a custom date range that you would like to apply as well.
I hope that you can see from here the value in having different attempts. I hope you can see from over here what the value is in giving Codex and GPT or any agent model, several attempts to try to do the same thing. We gave it the same PRD documentation and implementation plan, and they've all managed to get the core features going in the way that I wanted it, except that some of them actually extra stuff like light mode and dark mode, which is really nice. And this one has actually gone and done something where it actually gives you some signals to watch, as well as the ability to do a day-by-day audit of how it came to that conclusion. And then finally, this one uses a completely different theme and in my opinion, is the best one with this custom window that it has over here. I think all in all, these three designs are actually really good starting points for you to begin implementing the rest of your application.
Just a quick reminder that all of the data that you see right here on these screens are all dummy and mock data. The whole point is to get a look and feel of the application. Once you've picked the design that you want, the next step is to actually go ahead and hook it up such that it's using real data. But the whole point of today's video was to show you how the Codex CLI, that you see over here, plays in nicely with the Codex IDE, within Visual Studio. And then finally, how it ties in all together with Codex Web. The integration between these three platforms, including how it shows up on mobile, is just crazy powerful. I'm really surprised that not more people are utilizing this to its full extent, especially once you've wired it up with GitHub as well as Vercel. You now have the full ability to not just be working locally, but you can actually have parallel agents and parallel versions and parallel streams of work that will allow you to really level up your coding development workflows as well as to move a lot faster.
So that brings us to the end of today's episode. In a future episode, I will be showing you how I will go about using Codex to actually wire everything up so that we have a full working SaaS that uses not just mock data, but the actual data from Reddit as well. So now my favorite is this third iteration. But if you would like me to do the wiring up for any of the other two versions, let me know in the comments.
I hope you like today's video. Give this video a like if you enjoyed it, hit that subscribe button, and don't forget to turn on notifications so that you'll always be the first to know when new content like this drops. If you haven't seen my video on how you can get the best out of your Codex workflows and the eight different hacks that you should be aware about when you're using Codex CLI, you definitely want to check out this video that you see over here. Well, that's it for this week. Till next time, I'll be seeing you.