Transcription
start talking here in a moment. All right. Well, hello everyone. This is our second official afternoon tea for all of our lovely premium members out there and of course VIP inside of it. Now, the main goal for today is to actually look at skills and plugins because I saw a lot of questions from all of you about that. But from a kind of higher level perspective and then I also wanted to talk about the affiliate program that we just launched with the links that you all can access. Um, but let's uh let's go ahead and dive in. I also am probably going to ask some questions from chat today, too. So, that's cool.
Again, for some of you, you may really deeply understand a skill. And this may seem basic. Just hold on your horses. I'm going to dive into some more advanced stuff as well. But essentially, skills are my folder architecture, but in kind of a different routing way. And it's not just for Claude. Claude definitely really uses it well. Um, but the folder architecture of skills is very different. Um, oh, blue services. I see you hear no audio. It might just be a you thing. I don't know. Can we Does everyone else can still hear me? Okay. I'm sorry, bud. That might be a that might be a you thing. Um I'm sorry. Well, if anything, I'll have it recorded after if uh you can't get to it.
But essentially, um one of the most important things with Claude skills specifically is they have a little YAML um or JSON uh initiation at the beginning. So, anytime Claude is looking for a skill, it's going to look for a name and a description first without reading the entire markdown file. It's how it saves a lot of tokens searching through documents. And it has its own routing structure as well to go search through all of those as well. Like, hey, you know, locate these skills in this area. You can even create your own routing structure. And I'm going to show it here in a second.
Um, one thing is if you relay it to my ICM. So again my inter interpretable context methodology framework uh that I've kind of created that paper for those of you who've read it and for those of you who haven't it's essentially I tried to take thinking the way in which we think about processes we break them down we organize them and I took it and applied it to the way that AI reads these and over time I started seeing patterns of really good efficiency and formatting a skill is essentially what I am doing in a different format. And that's why I always say you don't have to take all of my stuff as concrete. It can be a version that you kind of develop on your own or you tweak it. It's just the way that it makes sense for me in most cases and I formalized it. A lot of people in this community were already doing ICM processes, but they just weren't calling it that. They discovered it on their own because it was the most useful way to route an AI through context context engineering with very little infrastructure. Obviously, there's a lot of ways to create infrastructure and route context. There's rag pipelines. There's vector databases, lang chain, langraph, semantic kernel, the list goes on. I just really love the simplicity of a couple folders and a couple markdown files with the occasional script here or there.
Um, but if you really look at what skills do right the YAML front matter that little thing that the skill has at the top is its identity right the description of where and what it's used for is kind of like routing for the skill itself but back at the model the skill MD the whole body or the multiple of them also has workflows and steps and that body may also route to other markdown files and references which you'll see there and if you look these are layers in thought. It's what when do you need to think about something? You think about the identity. Then you go in route to why you're using that identity. Then you're reading the contract of that process, that instruction. And then you're going and referencing maybe certain things inside of that process and then you create an output and then maybe you need to rework that output and do something with it. Skills are using this methodology. Absolutely.
Um and there's a couple ways. So I I wanted to instead of just sitting here and talking head you just dive in for some people again referencing where to find them, how to organize your file structure. These are all really important questions, but we want to think about how to think about these things. If you noticed all of my videos, I don't really talk about features that much. I talk about processes. And I could make a whole bunch of videos about new updates that Claude or the new 5.5 Codeex, right? I have subscriptions to Codeex, to Gemini, all of these. I'm constantly testing them, but I'm trying to create methodology that works with all of them at all times. And that comes down to breaking the thinking about using the features, not the features themselves.
But if you're sitting in there and you're in the app, you can come to customize and you can actually see there's a whole list of skills that and you can add as many as you want. You can create a skill. Well, when you actually look at that skill, this UI is handling the routing for you. It's coming in and saying, "Hey, this is where the skills are. Here's an easy way." Instead of me having to go in here, find Claude installed on my computer, click into the skills, and go find those skills. It's saying, "Hey, it's right here. This is the routing for us." And it makes it easier to find that. But you want to do the same thing with AI. And if you open it up, uh, let's actually look at the skill creator skill. It's literally a skill on how to create skills. Uh, it's the same thing. Markdown files followed by organized folder structure for different types of thinking. What types of skills would it need to do? Are you going to compare skills together? Can you grade that skill? And it creates an quote unquote agent, but it's just again an agent is just another type of markdown file read by the initial model and sharing that across, right? That's essentially all skills are. And sometimes they have HTML.
Um, so this is like a little uh thing that they built inside of the skills so that it could add queries itself and um Claude can actually do a tool call inside of the skill to go and look at the evaluation of that skill. So it's just a very deep workflow. This skill creator is a workflow that someone did manually with the AI and created a process and organized that process into natural language and gave it a tool or two that maybe isn't uh just HTML. In this case, a pi, it runs this Python script every time to output something so that the AI doesn't have to create this script every time it's doing it. But they noticed when they were creating skill, they needed an HTML file. they needed a Python thing and they just put it in there so that the AI doesn't have to think about using it. It's already good. It's already where it needs to be. And that goes for every single skill out there. There's some sort of combination or process. But beneath under all of the skills is that ICM, that kind of folder architecture.
And uh you can also download like plugins which are essentially skills but a little bit more connected. Um so if we come Whoops. come back to the customize thing and I'm going to show what this looks like behind the scenes too with VS Code. But these Oh, I'm not in cloud code. Hold on. Let's fix that. Yeah, this the other thing. Plugins only work for cloud code. So, let me go do that real quick. At least on the desktop. Oh my gosh. I cannot This is why I prefer. Okay, there we go.
Um, you can come in and these plugins are also like skills but with connectors built into them. So it's that idea of okay, I've created the folder architecture. I've created these markdown files. Now I want to connect it to outside systems. Well, that's exactly what this does. When you look inside, this is a markdown file. Again, it's in full markdown. You can see it here. It's describing a method or a process of doing something. But inside of the skill, it might be a reference to something it needs to do like a connector. So here I can actually connect Asauna or notion or Google calendar or Figma. So if I'm designing something and then I need to check in with my project management tool like Asauna, I can just install that and connect my Asauna account or maybe I have an in notion big database and I'm using notion to share notes on a team. I can connect that directly still living locally with folders in here. And this is what I think is going to become easier to do. Anthropic, OpenAI, all of these people are going to make this process easier.
And even when you're inside of VS Code, it's the same thing if instead of having the skills in your so let's take uh let's take this for example. This is my writing and scripting um process, one of the many that I've made. I have my cloud MD that routes to everywhere, right? But if you notice, there's not a skill file in here. Now, I could put some in here, but Claude Code installed on my computer already has skills involved, right? I did add some specific ones inside of this one, scene generation, because it needed stuff specifically and I built them myself. These are my own skills that I'm using for Nano Banana Pro, Seed Dance, a certain style that I like. Um, because I'm testing a I do my animations, but I'm testing where the animations are actually done by like Seed Dance or Cling, which are visual models. Um, and it's a it's a it's a test and process. You'll see a video on it here. But most of the time the skills I need when I just open up and I'm saying working most of the time those skills are already installed across everything right and there's also a skill to find skills right so if I can say hey let me use a skill to find skills that skill is not in my folder structure here it's actually installed directly inside of claude on my computer right so if you go and find um the enthropic folder where uh claude is installed which you can see Here they actually have their own local skills. So this is across all the projects. Anytime I open up claude, these skills are all available. And if we look at the where's this find skill? Find skills skill, right? It's just a markdown file on how to find skills, right? Hey, find skills helps users discover and install skills when they ask questions like this or this. And then it goes through and it has a link inside of it already as if you're copying and pasting this to go and find that skill which it goes to this website which is a really great website if none of you are aware of it skills.sh SH again as I mentioned last week be very aware of prompt injection because as all of these things are markdown files and prompts someone could be malicious and hide you know if let's let's look at a really intense uh and deep skill here um which one has a lot I think yeah so this one has a bunch right they could have a prompt that's inside of the Python code that injects into your cloud session and tells it to navigate to a website and then maybe upload some sort of work that you're working on or share av file. Same thing in here. These are markdown files, but they could have code or a prompt injection hidden somewhere in here and it might not notice it. It's just going to know, oh, these are my instructions. I'm going to run these instructions. And that's something you want to be very careful and aware of. However, these kind of open agent skill systems, people are usually monitoring it and going through it. Um, but there's some amazing skills on this website. Let me actually throw it into chat for you all.
Um, yes. And Erin, I know I me and you, we need to get together. Erin has been huge for me um in a lot of these, but am I moving too fast for anyone by the way? Is this all making sense? if you could throw into chat whether or not I'm moving too fast or um oh it's good. I thought you said yes I am moving too fast. Carlos is like, "Yes, Carlos, I this is actually a lot of this is in response to what you asked in Discord." And this is the whole point of of kind of premium like obviously I'm sharing I'm going to share a lot of these markdown files constantly uploading our my the vault with these things so you have them, but I think it's more valuable for you to ask specific questions. And my time is extremely limited. Um, but this is what's not scalable is these kind of back and forth. So, uh, Carlos, is this mostly answering some of the questions you had in a lot of ways? Uh, I'll let you share what you're working on here shortly as well. Um, but I just wanted to make sure we're on point.
Let's see. Janette, curious to know what all uh your workspaces are. I currently just work in one master where I do all of my work and some personal stuff, too. Curious how better to think about multiple workspaces. So, there's there's there's actually some divided opinions on this. Um, one of my favorite opinions is actually from Andre Cap Kaparthy. Um, and his wiki, his a AI wiki. Um, yes, right here, this one. And I'm actually making a video on this as well because it's very good, but his idea is that you create a kind of self-learning environment where um, let's see, does he have an image of it? Um, I'd rather show this visually um because it's much better. But the idea there is imagine you're sitting there and Oh, no. No, no, no, no, no. By the way, this is one of the founders of OpenAI, but um he basically had this idea of if you have all these workspaces, there's things you can pull from it and create memories on it and basically give your AI a brain of where the workspaces are, what the workspaces are, being able to search them. So, the organization doesn't matter as much as allowing them to kind of go in there.
Um, and then another good one, yeah, David, that's actually another great marketplace um for skills is Lowhub. Um, there's a lot of different curated collections, especially for Obsidian skills. And Obsidian is actually one of the main things that Andrew Kaparthy is using to organize uh a lot of his models knowledge um, or it's one of the references and other ways people are doing it. However, the other thing is just very informal structure. So for me, I just put everything on my desktop, right? These are all workspaces for me and they're just right there on my desktop. When I'm done working with it, I can toss it into a folder of past work and then it's all there. But they're also not technically in the same folder. They're kind of separate and I can just come in, log in, and and call it a day. Another option is to hide all of these on something cloud-based. So if you have a one drive and you could do this methodology with co-pilot, you could put all of your workspaces into one drive. I've seen people do it inside of notion. Um it it really comes down to again your personal preference. Where where do you want to keep your information? What is simple for you?
Um, is in one drive. So I moved it to the root as I was concerned about files with credentials backing up to one drive. Okay, this is a very great thing. Um, there is something and that's actually the final place which I see which is people are actually using GitHub a lot more because GitHub is already designed to help you navigate file structure. So well and when I upload maybe a workspace like I have a whole bunch of repos out there and I'm shared repos where I'm working on Adoba stuff with my company right we can add something called a git ignore and what this is is a markdown file or in many cases it's actually its own file a git ignore git ignore um and it basically says hey ignore these type of files um and one of them can be aenv so anytime you're uploading or sending stuff to a shared workspace certain certain files can be said, hey, don't upload those. Um, you could do a kind of fake git ignore inside of like a markdown file. So, let's say we're let's go back to our simple one here. Let's say I wanted to go in and I wanted to create a new file. Um, ignore these MD. Um, this won't always work, right? This is this is where AI becomes kind of a problem. But you could say you could literally describe these are the types of oops types of oh my gosh I cannot type today. types of files I want you to ignore. And then you could describe them, share them, even link them. And you would just need to make sure that in your claw.md right at the top you have a ensure to always check um ignore MD. Right? This is the most broken down simple version of what you know a git ignore does. Problem is it's much less programmatic. you're relying on whatever model you're using to read that and then read this and then verify before you're pushing or it's reading those things which is another part of it.
The other thing which actually Toby this is a really great point is just keeping it on a separate drive or a virtual machine. There's a lot of YouTube videos out there. Again, I find that simplicity is king for me. If I'm working on a project and I don't need to upload anything, I won't upload it. when I do need to upload something, I usually upload it to GitHub and then share it with my team because then I can just use a git ignore file. Um, I hope that kind of answers some of your questions. Uh, Windows Credential Manager is a really good one as well. Um, because you can flag certain files or or even folders from having access. Um, even in your one drive itself, let's say you do upload an ENV file, that's okay as long as that one drive is secure and people don't have access to it. It's password protected. You actually have a pretty good bit of security with one drive. It doesn't mean it's perfect, but Microsoft does have a history of being, you know, with the US government. It's why they get kind of well, one vendor lock in. That's another big part of it, but they do have pretty decent security with a lot of things they do. Um, so that's that's another side of it as well.
Um, OpenBrain Project. Yes. Yeah. No, Nest Openrain Project is really cool. Let me actually pull that up here. Um, I've messed around with that a lot. Again, I always break down to the fundamentals and I end up rebuilding stuff myself. Um, but let me see if I can find uh yeah, with Oh, that's because I I want to also I would love So, Nate B. Jones is one of my favorite YouTubers. I don't know if any of you know him. Other obviously, my favorite YouTuber should be myself, but Nate B. Jones is someone I've watched for a while and kind of dove in. And one of the things that he looks at is this kind of idea of a uh open brain project. Oh, this is my notion. Um of this open brain project. This idea of like hey we have these workflows, we have these processors, there's infrastructure layers for thinking. He it's using the same concept that I'm teaching you here. And that's why I'm always saying like be careful how concrete you take from me. Yes, my stuff is great. It is a great fundamental, right? this this paper this GitHub wait interpretive interpreted context methodology we need to find the paper um oh what does Gemini have to say about me interpret is a fuller architecture man it organizes work in spaces yet developed by Jake Van Clee in there I see him uses three that's okay that's not dec that's not bad I like that look at that Gemini talking about me anyway um this paper is useful and is great to give you an idea of the thinking behind all of this, but it doesn't mean this is the only way to do this. There is going to be a multiple ways and whichever one clicks with you the fastest. That's the best way. Something that you can understand faster and better is always going to be the most useful to you. Even if there's a more efficient one hiding on GitHub somewhere, right? I find that engineers take so much time trying to find the perfect most efficient solution when in actuality the solution that works the best is the one that most people can use as quickly as possible especially in consumer culture.
Um so on that you want to search OB1 um on Git Oh yeah. No, that's that's a good spot to look as well. U get open rank. Let's see if it actually pulls it up for everyone. Um, yeah, same exact thing there. Let me actually toss that in here. That one. And then this one is a different type of project. Actually, it should be the same. Is this not the same? No, this is the same. Yes, I was correct. Um, so yeah, Open Brain's a really cool project. Uh, hopefully I can do a collab with him. I need to reach out to him. Um, I don't think he would have any problem with it. I just haven't spent the time. Um, he also has massive amount of subscribers and stuff, so I'm sure he's super busy. But really recommend this guy.
Um, anyway, that was a lot of information all at once, so I hope you uh you got it there. But yeah, that at the end of the day, skills, going all the way back to the beginning, skills are my folder structure, but for smaller workflows or parts of workflows. And that's the idea, right? Eventually, that workflow that you've created out of folders can be condensed into a skill. It depends how much you want a human in the loop though because for me I actually don't always download skills. I often just look at what other skills people have made and then make my own, right? Like there's skills out there for these models, for these processes, but I wanted one specifically for what my process was. And that's why I teach this process because nine times out of 10, your thinking, your opinion, and I actually made a slide for this because I knew I was going to get here. Um, ends up beating the downloads half the time, right? You yourself can really learn from the patterns of those skills and then write your own to match your work, pull from those skills. And that's that's software development in a nutshell. Before AI, I don't know how many of you were software developers before AI, but Stack Overflow was basically people finding scripts, Python scripts, C scripts, HTML, asking people question, hey, what have you built? We would take someone else's codebase, rip out what we don't need, put in what we wanted, and integrate it into our own codebase. That that's software engineering most of the time. Very rarely are people building every single part of the software.
Um, but let's see. uh in regards to skill. Oh, wait. No, everyone's on comments on Nate video. Oh, yeah, dude. That's great. And everyone go comment on his like, "Hey, you gotta work with this guy." I haven't even messaged him yet, so let me message him. But go ahead and do that. In regards to skills, how much context also so you guys can see this. Um, in regards to skills, how much context have you seen usage increased as you keep adding skills? Yes. So, this is this is hugely important. don't want to just add a whole bunch of skills to something, right? Because look, look at just these few skills. These are lots of lines of markdown files. If it was reading every single one of these every single time, you're increasing context a lot. How much from that skill do you need actually? And you'll see some people have already what I call um made their skills more efficient. So, if you come in, right, you'll see uh that's not a good one. I'm trying to think. Is it the MCP builder one? Yeah. They'll break their skill into multiple markdown files depending on what they're doing for each step. And you could have sub agents only read one part of the skill while another sub agent's reading another. And it lowers how many the initial tokens are.
Um, but I do will say I mean skills literally are just a massive amount of tokens that are going to increase it, which is why it's so important. You only use them when you need them, which is where my ICM comes in. When I'm creating pillars and brand ideas, I don't want to use every skill. But in the same project, I might need 150 skills. So, I'm organizing the folders by structure so that if I point Claude at understanding my pillars, it's only going to read the ones it needs for that task it's put on and then I can open up another instance of it. Or maybe I want to open up codecs or maybe I have a local model. It's the same concept. You're trying to break apart and make it so that the AI doesn't have to think where to find the task, but it can read the task as it needs to.
Um, let's see. Can I make this bigger? Let's see. Um, I take what I learned from you and prove my message. Do you keep skills outside or in projects? Both. Both. Both. Both. Both. Both. Both. So, if I find that that skill I'm never going to use outside of that project, toss it in there because then it doesn't have to do a bash command into another folder. It can just in its process read it in there. Right. The skill. That's why I put these skills here. I'm not really going to use these skills for any other project because these skills are dedicated to specifically this folder in this project, right? But there's a lot of skills I use all the time and those ones I like to have installed so that I can, you know, access them without having to install them into my product and then I don't have to make my workspace flooded with extra stuff. Right? So there's certain skills that you might use across workflows.
Um, let's see. Your skills create a script for D. Yes. Yes, absolutely. This is one of the best things you can do. Um, and you'll see most skills actually have scripts in them for the deterministic thinking. And this is why I made an argument months ago on a YouTube video saying that AI are the current level of models are kind of like compilers as they're taking logic and compiling it into a new semantic language. They're not exactly like compilers. Compilers are very deterministic, but they're also it took a while to get them there. If you look at a lot of skills, some skills, uh, is it front-end design? No, docs. Yeah, doc. Docx creating a word document as a skill. It has PI files, Python files, so that it's very deterministic on what it's outputed and how it outputs it. But on the fly, it can also edit that Python file. So, it's kind of the best of both worlds. And you can even describe in the skill how to use those scripts.
Um, so yes, absolutely. Um, you want scripts for deterministic outputs. I I highly recommend that. Um, and it's really good for token. And you can see eventually you you maybe you might start out with no scripts at all and that's your skill and there's just a whole bunch of markdown and things. You might notice parts of it where you're like, wait a second, why am I describing this and making the AI do this? I could just write a script that tells it to do this instead of waiting for the AI to figure it out and write the script for itself. That's the whole idea behind skills and ICM. And my um my 60 3010 method is built on that. Most of the time 60% of your workflow is probably going to be traditional solutions, right? Most of my workflows are folders and files. people were trying to use AI to do all these crazy routing logics and stuff, but I could just use traditional solutions. 30% might be database. The magic of the 10% AI really 10% of your workflow being AI is where the magic really happens. You're you're really lowering how much tokens and processes there.
Um, yes. No, I'm very excited to drop about that competition. We'll um I'm having Matt and I kind of look through it and we're definitely, again, this is our first competition. Um, for those of you who who don't know, we we ran a uh a competition um right now and we I've been giving them I realized some of the best ways to learn is just to create, right? And so we created a whole series of fake kind of don't worry these aren't actually real things. We're not offloading work on to you all. Um but it's uh let's see where did Matt put it. It was essentially like hey you have this client and you're trying to do work for them. Um, and it was a dog washing client. And the idea is what can you do to help them with their branding workflow process. Um, here we go. Here it is. And the idea is you just get a basic brief. Now you have to figure out, can you build the structure process, the brand voice, something like you see here, right? Like can you build this type of process for them for someone else and hand it off to them? And we're kind of excited to see. And then of course whoever wins is going to be $200. We are gonna do community vote because I figured it would be kind of bias. Um, and I don't also have enough time to really focus in on it. But if that doesn't work well, we're doing one of these every week. Um, and we're going to refine what the voting is like when it's pushed there. So, please just work with us on this.
Um, but I want to and don't worry, Jenna, that you missed it. We're doing one every single week because you all have been supporting me. Every premium member is getting immediately allowed to come into this. You guys are giving me $27 or for those of you who joined early, $7. um you all are gonna be having access to win 200 bucks every single week because I just again I make good money with my main company. I'm not here to take money from everyone, but I need to put a pay wall behind some of my time. So, I figured a good way to make me feel less crappy. Um even though, right, you all are hopefully getting a massive amount of value is like, well, let's make competitions where you all can earn money. Um and that's that's something there. And I'm sorry that you missed this, Matt. Um, but again, literally every single week. So, for the next year, every week, we're doing a competition. So, if you missed this one, there's going to be more. I'm realizing now we could have done better marketing on it and letting people know. So, I'll work on that in the future and talking about it in Discord.
Um, but that's the idea there, too. So, uh, just wanted to share about that. Um, yeah. No, dude. I And this is I checked on it this morning. I'm literally You know what? Let me just give you an idea of what like my normal kind of day is. My Instagram and my um my chats I when when I respond to you understand I'm responding after a massive amount because not only do I have crazy general but my request it says 15. It's not 15. This is in the last 24 hours, right? And I love it. Like don't get me wrong, this is a great problem to have, but like I am I am putting payw walls in around stuff because if I tried to answer every single one of these, I just I would never get to any of you all.
Um, could the competition get pinned? Yeah, I'll repin it right now actually. Um, so, uh, so that could be there just in case too. Let me share the link here for you all. Um, so that you kind of get into that. Yeah, I know. I feel bad. Aaron literally puts massive amount of work for me. I'm he's amazing and like also a devil dog brother Marine and I still barely have time for him and it's super bad because he has two really cool projects I want to work on with him. Um and he also has a friend that he sent a thing there. So I'm it's it's I need to get a second phone and a virtual assistant. But I also struggle automating this because then I'm missing the interaction. The whole point I started this was the interaction for things and thank you Erin, you're you're the best.
Um, but yeah, I'll make sure that the those are there. Um, skillbuilder MCP also great by anthropic. Um, and I'm glad that you really like this kind of method. My research paper um on the history of software development. Yeah, let me let me go ahead. So, it's on it's published on rxive um I also have an ethics engine research paper where I go into um psychometrics in AI that I published which is cool. So, that one and then interpret interpretive context methodology. We'll go right here and I'll share that. I do make posts around it every so often, but I'll make a big post. So, those two are really good papers if you want to dive into like my more academic research. This one's really crazy. I um I I basically gave a whole bunch of I created a data pipeline uh where I took a whole bunch of psychometric tests, personality tests, and aimed them thousands and thousands of times at some of the top models and collected their answers and basically found patterns on how they answer. Uh which is is really interesting.
Um, and then of course, yeah, the the context methodology. I go all the way into kind of the history of like, okay, um, the Unix tradition, what we're doing there. I talk about David Parn's ideas and there's a lot of good links like I have even if you don't read most of my paper going back to some of the original content uh back from the 1970s 1980s is really cool because again these are super important processes that were developed way before I was even born. That's what I'm doing. I'm learning from the past to build the future and that's that's where a lot of this comes. Same for that. I realize I'm talking very fast so I need to slow down. I get very excited about these things. But yeah, I uh I I I very highly recommend checking that stuff out. This is why I make videos because it forces me to slow down.
And Woosaw um outside of the competition too. Again, I wanted this to be I realized I'm getting so many people in here and a lot of other communities will charge a lot more money than what I'm charging to get people in. They won't even make it free to join. And right, I I I I think that's because they want to make money in the front end. But in my opinion, making other people money makes you more money in the long term. Um, and so we we figure it out. We're going to test it and see how you all like it. If you do, it's a um a referral program. So, if you refer someone to premium or you refer them to VIP, you make money. Now, I want to be very careful with this. This is dangerously touching on multi-level marketing, and I do not want that to be here. I just wanted to give you all something of value. This is not a full-time job. This is not me saying, "Hey, here's a course on how to make more money with more money." I just wanted to say here's a link if you want to share it with people because you're getting value. Here's some cash because I'm getting value out of all of you being here by being a part of this.
Um, and that's to me is important. It should be I think um let me see. Whoops. Going to open all the Discord. Oh, there's my desktop background. For those of you who are Dragon Ball Z fans, well, this is Dragon Ball before even Dragon Ball Z. Um, is very good to Where did I Oops. School. Here we go. Um, one thing that you I think you just go into settings and then go to invite and it should have an affiliate link. I'm an owner of this so I can't remember where it is or I don't see it. I don't have that option. Um, Matt kind of talks about it right here. Where is it? Yeah, open school. Click invite button. Okay, you click the invite button. That should be at the top of the page and you should see an personal affiliate link. Feel free to share it with those people and uh it it's pretty good. Oh yeah, dude. I've dra I've read and watched all of the manga of Dragon Ball, Dragon Ball Z, Dragon Ball Super. Uh I'm like uh what is it? Dragon Ball GT, which technically isn't canon, but I like that, too. And then for those of you who don't know, uh but maybe you should check it out. Dragon Ball Z Abridged is one of the funniest shows I've ever seen in my life. Basically, they dubbed over all of the characters um ideas uh and voices, and it is so funny. Uh highly recommend it from Team Four Star. Little bit offensive. probably wouldn't go well um in in every, you know, walk of life, but definitely definitely check that out.
Uh, let's see. Okay, so uh let me stop sharing my screen and and kind of go. I I think I promised Carlos that I would let him talk about his project and I took up a whole bunch of time here. So, um I know we're six minutes over, but that's okay. I I like hanging out with you all. Um you got to start at Dragon Ball if you haven't. Oh, you know what would be cool is if I made shirts based off of Cleaf Notes for Dragon Ball Z competition winners. got them. That would make me happy because I would wear those too. Anyway, um Carlos, would you let me let me make it so that you can share your screen and talk. Where are we here? Um there we are. So, you should be allowed allow audio and allow video. Um, but yeah, let's uh share your project questions, things like that. I'll give you give you some time here. Um, like I said, I know we're over time, so anyone who doesn't want to sit around, you're not you don't have to, but I'd love to hear from you, Carlos.
Yeah, thanks man. Can you hear me there? >> Absolutely. Right. So pretty much what I'm trying to build here is I've been going around um you know AI agencies for you know go high level and all those automations but now I moved apart from the companies I work with and I'm trying to build a personal AI assistant which can then create for business owner a a personal assistant which they're able to manage data from their business or maybe create a new workflow without breaking the structure so the agent won't be hallucinating or having all these difficulties. So it's kind of something I want to build as I already have my AI agent but try to replicate that. So between me and him scaffold the whole thing for business owners non- tech guys you know people don't want to mess around with programming don't want to be managing files and stuff. So me having the, you know, the admin permissions and then they'll have the owner permissions for the AI assistant, which practically when they say, "Hey, I have a new employee," they can call their AI and let them know the their phone number and just move with their business, you know, pretty much. I mean, I have I haven't gone through it a lot. My AI assistant works pretty great. is still managing a lot of stuff there. But I understand that every client, every business owner needs different MCPs. So I I found that Composio could do a custom MCP connector which then I would link it to manage agents from Tropic so that their AI agent would be living via Appy because mine would live from my subscription, right? So instead of me setting up the whole cloud subscription for them, I would just put them on an API for anthropic manage agents and, you know, try to make it work.
You're you're muted, brother. So >> thank you. Um, I think that's a really solid idea. One thing that I I've I've gotten some success with in the past is like organizational management uh is really annoying for a lot of companies. And if you can like make that easier through like a UI, um, being able to do that, people will pay money for that. I know I'm constantly see me on Instagram just absolutely crapping on agents and stuff like that. It's it's for most people. I will say there's plenty of situations to make money and and today's VIP session, we're actually going to dive deep into um kind of uh one of one of the uh individuals in there had a similar question like, "Hey, I'm building this stuff on Langraph. Is it going to get replaced?" And my response to him is essentially going to be, well, it might. Anthropic can drop something tomorrow that does what your company does. It doesn't mean that you can't make money with that company, right? People are still not, some people don't even know what chatt is, or they do and have never used it. I just worked with a uh a client down in Miami. Everyone in his office, all of his admins had never even used a chatbot, right? That still exists. And this is a multi-million dollar company. It's not like they were just some small. They're bringing in a lot of money, but they're in the construction side. So, they don't always maybe see that in their algorithm. And a lot of you might see a lot of AI out there, but at the end of the day, you have to remember a huge amount of people are not touching it and using it in the way you are. And that's value. That's shortcut that people are willing to pay for. Now, for your specific example, Carlos, I think um yeah, the things you want to look at there is um if you are building through anthropic managed agents, understand you're going to be locked into anthropic. That's okay. That's fine. If if you're able if you like the value in them and you see value, then you know, do that. Run with that. The only thing is if Anthropic ups their prices, if they change how manage agents work, you have to then and you have clients and businesses using it at scale, you now have to deal with what that is. So, I would always keep in the back of your head, it doesn't mean stop using managed agents. If it's the cheapest, quickest, fastest solution that gets you value, absolutely use it. But always think, okay, is there a way for me to do this without anthropics mandated agents or how can I create the structure separate from it that I could latch on elsewhere later? You don't have to implement it, but it's just one of those things where you want to have it in your back pocket. So, if you get a bill from anthropic API that's triple what you were expecting and you're at scale and you have 2,000 clients on this software, which I think you could, right? What you were describing is very useful.
Um, and you want to pilot it. Are are you already pil piloting it with anyone right now or um is it not? >> So what I have here is um pretty much the the front end for my clients right now is go high level because that allows me not to create an API for WhatsApp for Facebook, Instagram. So the front end as a CRM for them to manage their clients and then I would I would connect my AI assistant. So everything lives in the same repo in GitHub. And then when I have my client's folder only in their folder their their assistant can use uh GitHub API to manage business configs and it would be and that's why I'm using my managed agents practically so that they can touch the part of GitHub only the business part not the config but I'm still on top of everything with my AI assistant and he's trying to you know like actually correct mapping and all that ICM that you've saying I would not made something custom right now because
I think those tools like the Go Level and Manage Agents are far instead of me going creating something. You always say like, until then, you know, until you have that issue, go ahead and do it. I do not have any pilot right now. I mean, I was trying to do it for some one of my clients, but I think I've been kind of stuck on the on the part of the skills and all that you were saying at the beginning. This is actually something I can, you know, replicate for a new client without me having to go through the whole thing about onboarding and and all that.
So, it's I'm kind of stuck on that part where I can make it scalable, right?
Yes.
It should be easy. I mean, it should be easy the way you put things, you know? It sounds easy and it I understand it might be a lot easier with the workspace builders and client delivery and all that you already you already have in your, but I'm kind of stuck in that part where I'm I'm trying to make it manageable and scalable for myself without having to, you know, go through all the well, pretty much from not to start from zero, but actually go forward and and scale that.
Yeah, I think the scalability question is going to be a question you're going to hear me talk about a lot more too, because we're we're looking at how to scale ICM inside of enterprise a lot more. Um, the process, the learning, and uh, our solution was to build a kind of scalable learning solution called Scribe. Um, but for your specific situation, um, I think, and this might sound crazy, what might be interesting, have you have you thought about this? And maybe I'm maybe you've already done this. And have you thought about trying to have the process be manual and train and work with one of your clients or even a new client and upfront say, most of this is haphazard and manual. I want to work with you or one of your people to get the outcome. So you use your kind of disgruntled, kind of all over the place systems and you use people to scale it short term. And then as they're using it, as they find problems, you find just one little thing to automate and you chip away over a month or two months automating those small parts and then you integrate those together rather than trying to figure out what the scalability is now or what the integrations are now. Try to get the outcome that they would get with that software manually. And then as they're working with it, I guarantee you you're going to notice points like, "Oh, oh, this is what I could add in here." Or, "Oh, wait. Go High Level already has this integration to handle that. Let's just stop doing it manually."
Um, yeah, it's almost like key loggers. That's funny. Um, David, that's really good. And Toby, you said you're you're doing that for your company. Maybe I shouldn't be talking. Um, Toby, would you be willing to maybe describe to Carlos in in your perspective? Because again, I'm not the only one that has value here. This is the whole reason I bring these groups together. Uh, if not, you don't have to talk, but um Oh, wait. You have to be allowed to talk, don't you? Let me show you. You should be uh, let's see. Allow audio and allow video. There we go. Um, yeah. Toby, do you want to chat on that? Possibly. If not, it's okay. Um, I think I just gave you the ability to talk, allegedly. Is that making sense, Carlos, while we're waiting for?
Yeah, I mean, it it totally makes. One of the things that I also use as um, you know, infra is the same High Level because you can build workflows in there. So instead of me building like on the folder side, some workflows, some of them I just make them, hey, I have this workflow in High Level, what you can use instead of pulling it out, you know, not letting any like everything to the AI. But the problem is that what you said, because not every client's going to be the same. So I'm trying to be a more niche guy for some businesses that this can be, you know, a little bit more replicable.
And that's actually a huge part. Start niche, then find the scalability. Everyone tries to find the scalability and then niche down. I think in the AI world, you'll find the patterns if you solve the problems of something niche and you'll slowly take components out of it um with every single niche client that you get. And now this is hard, right? Like that's going to take time. That's that's a good moat. If it takes time and it's hard, even with AI, that means you found a very good business because now it's not something that someone can copy easily, right? You want to look for the things that are not best, but different, that are are opinionated or are niche because that's where you're going to find value over the next decade. Sure. Are most of the things we're creating now going to be automatable in two years, three years? Probably. We're moving at a fast pace, but the problem is, it doesn't matter if you automate everything. What matters is who you're reaching and if they see you first, right? There's a thousand copies of every company out there. Every single one of them is making money. Almost every single one of them. The ones that aren't are the ones that either give up or invest too much in the wrong thing. At the end of the day, you could make an Uber app in your local city that does something specific for your local city that Uber doesn't, and you'd probably get some good customers. Like, I guarantee you. But a lot of people won't try to build that because they're looking at too big of a market. Start with a niche market. Start with one company, one person, then build up. Because I promise you, once you get it right with one person, you can get it right with 10. Once you get it right with 10 people, you know what you need to get to a 100 people. Once you get it right with a hundred people, then you can get it to a thousand. But until then, you will, if you try to skip to 10,000, nine times out of ten, you will just be stuck in a loop of figuring it out and getting stuck in barriers versus just brute forcing it. Plus, with smaller niches, they're more forgiving. They're more willing for you to say, "Hey, I don't have this fully solved. I'd love for you to work as a partner to help me break through this productionizing thing." Uh, to see if I should even productionize it in the first place. Uh, right? Because your goal should be get the outcome they want today regardless of how manual it is and then work backwards on making it automatable. Like a lot of clients, a lot of software I've built, I did the process manually for a client with a couple of AI tools scattered and then did it more and more. And out of annoyance, I found the production because I'm like, I don't want to do this anymore. And so I'm always trying to automate myself. That's always the goal. Every day I'm like, how can I replace what I'm doing today? Because I feel like there's so much more I could be doing. Uh, there's always people that I'm trying to help, always people that I'm trying to think. And then also there's the artistic side. I'm always trying to make music or art or these things. I think the idea that if this skill I have gets automated or this process I have gets automated, I'm replaced is is crazy. Humans aren't replaceable when this is very hippie-dippy. So some of you might not believe this, but humans aren't replaceable in the same way I think people put it out to be. I mean, human creativity is infinite. I believe that. And yeah, okay, sure, we replaced carriage and horses with cars, but guess what? In St. Augustine, I see more carriage horses going down my street than cars. Like, we have a whole bunch of them still because people care about those things. And that is a market and someone is making a good living off of that. We're we're so focused on this crazy unreachable goal. It's not unreachable. Everyone in here could be a billionaire and I'd have no idea. This this idea that's like, "Oh, I have to I have to change the whole world." Not necessarily. Sometimes changing the world is one person's world. You've changed their whole world, right? And that could be enough. And then you can scale that later. Right. I'm a capitalist hippie. I don't know what to tell you there, but that's the way I see it. And your niche is much easier to handle.
Exactly, David. Um, Toby. Oh, yeah. Sorry, Carlos. I went on a freaking rant there. Sorry.
No, that's great. I mean, that's pretty much what I did. I did wrong that I try to scale the whole thing, you know, to every business I can I can get in contact with. So then I I run into the same problem, right? That it's not maintainable for me. And now I I went back steps. I I I backed up my steps and, you know, I I just I'm trying to restructure the whole thing, but I'm still learning the ICM. I'm still learning your p mythology. It's kind of like uh I'm stuck with both things, learning, but also produ, you know, going to production. But I guess I I went a few steps back and and and instead of building like a whole thing, I'm just going to start by a niche, which I already have an offer for that and I'm going to be hitting that, you know, probably this week. Uh,
Good.
So, yeah. Good.
Thanks for that, man. Really appreciate it.
Few steps back is still moving forward, right? Because sometimes if you were kept pushing through that wall, guess what? In six months, you would have come to the same conclusion that you are now. The all business is is pivoting. Right? My what Aduba was I went this is something that all of you need to know. My company, right? I I would argue I'm very successful now. Four years ago, I was in a business competition at one of my older schools. I lost. I went to it three times. I didn't even place in the top three every single time with what I am doing today because it was different. I wasn't perceiving it right. I didn't. All of the things I'm telling you aren't because I'm smart. It's because I screwed up. I took the steps back. But guess what? Those steps back were steps forward. They only stay steps back if you don't learn from them. But when you actually look at the whole thing, those steps back were you learning. If you look at it as a huge progress bar, it's not that you're falling down back on progress. It's that you found a new way to navigate the learning, right? It's a different path to the same direction. So, yeah, dude, focus on that one niche. You'll figure out a way. Just trust that you'll figure out a way to scale it and try to make it so good for that one client, so killer that from that process, you watch. I guarantee you're going to come back in like a week or two, you'll be like, "Oh, while you were doing this, I had this brain explosion and I figured out how I could scale it here." Because I'm telling you, you will. Like I'm glad that you're learning from ICM, but what you are doing now is creating your own solution, which will be as valuable, if not more valuable than mine, right? Mine is a general solution to the way of thinking, not to the way of implementing, which is where it's limited. And I'm trying to figure out how to do that. But yeah, just just focus on one, then move up again. Right? If you bite off more of your chinchu, that's okay. Take a few steps back and try again. That's the whole that's business, man. That's business. You're it's never never going to work out perfectly and that's the point.
That's right. Thanks, man. Appreciate that. Really.
No, of course. That's why I do these. These are my favorite things. Um, well, on that note, I am 25 minutes over, 23 minutes over, which is fine. This is the kind of stuff I like. Um, I will make sure everyone has this recording. I'll be putting um a bunch of extra stuff. I think um the markdown files and stuff I shared at the beginning, I'll share a whole bunch of that. I'm going to make a new um digital asset in the vault for everyone based off of that and based off the questions I saw. Um, for those of you who are here, seriously, thank you. This is this is my favorite thing to do is just talk and learn and teach. Um, so stay stay careful out there. Be careful. I'll have this out there. Don't rush, but rush as much as you want. Just make sure to give yourself the grace when you when you trip. That's fine. Learn from how you tripped and where the crack that you tripped on was. So much love. Oh. Uh, shoot. Hold on before I go. Ness, how do you join this Discord? If you're not part of the Discord, huge, huge thing you want to be a part of. Um, basically, let me just share this real quick. Uh, there's a lot of value in the Discord that I recommend because I also go in there and answer questions outside of this. So, if you're, you know, if you're in here, you're a premium member, you go in the vault, you come in and it says Discord setup, you're going to click this link if you're premium. If you're a VIP member and you're in here, then you want to go into the drawing room. Um, and click that link. These links are designed for um to give you the role automatically and then it'll throw you into the uh into the Discord and there's a lot of people talking. I sometimes come in and I'll do um kind of big dumps like this responding to people who are asking questions. We've got voice chat in there for for collabs. Um, I have a collab in here that I've I've been super ignoring this channel, which is bad. Um, but there's a lot of ideas for pushes and ideas. So, I highly recommend getting in the Discord because it's nice and also Shishro and Aaron are actively in it a lot, which is super nice, too. Um, so yeah. Cool. Now, now I am ending it. Um, because I have to get ready for the VIP call coming up in the next five minutes. Um, so cool. Much love, everyone. Bye a second time. And, uh, if you all want to chat and talk in here, go for it. But, uh, I'd recommend just going and checking everything else. And happy learning as always.