Building AI Employees With Claude Code Sub Agents

0:00The world of AI is constantly changing and in this video, we're going to be deep diving Claude Code sub-agents on how you can become more efficient as a business owner and an entrepreneur by building these agents out to complete tasks for you. So, let's get into the video. All right, Brett. So, you did a great presentation the other day in one of the groups and I wanted to bring you on the podcast because you did a really good job explaining what Claude Code is, what it does, and how we should get started. So, I kind of wanted to just bring in a lot of the beginners up to speed as like what makes Claude Code different than some of these other, you know, things like Lovable and Bolt? Like, why should people make this the change to it?

0:46Yeah, that's a great question. Um it doesn't have to be Claude Code, let's start with that. Uh I personally love Claude Code. I've been using it a lot. I was started off using Cursor. Um no real complaints there. Just wanted to check out Claude Code because I kept seeing a lot of things about it on social media, mostly on X. Um and the more I dug into it, the more I realized how extensible it is. And it's sort of it's not the most user-friendly out of the gate. So, I'd say that's probably the biggest barrier for anybody who's a beginner in this space is Cursor is going to give you an interface. You're used to seeing an interface with, you know, file, new window, etc. And it kind of it not hand-holds, but it sort of like

1:32gives you more controls that you can see. Whereas, Claude Code is like, "Hey, here's a flashing cursor in a terminal and you're kind of on your own to figure it out." Um now that being said, I think the internet is going to get you more than ample resources if you're willing to go look for it. And that's kind of where it hit the sweet spot for me was if you I'm a I'm an explorer, I'm an experimenter. I love building stuff. I I pushing these things to see what I can get them to do. Meaning the large language models in these coding environments. And so the more I dug into it, the more I was like, "Oh my god, like you can customize this thing to do anything." So it's almost overwhelming, right? It can be too much. Like a tool that can do anything

2:17to a new person, you're like, "Well, I don't know what I need it to do." So hopefully we can go through some of that today and kind of go over you know, again, we're going to cover a little bit Some of the topics we'll cover are general to Vibe Coding or LLM coding. Um And obviously the use cases we'll be covering that we'll go over will be Cloud Code specific, but some of the things we're talking about are just like if you are using Lovable, Replit, Bolt, something like that and you're ready to sort of take it to that next level and graduate from the limited set of tools that they're going to offer you. Uh I think this is just sort of help you bridge that gap and figure out A, when do you need to make that transition? B, what's that transition going to look like, right? Like what should you

3:02expect? Are you going to take your code base from something that's very user-friendly and just be dropped off in the desert and expected to figure it out? You know, maybe maybe a little bit. So there's a little bit of um an adventurous spirit required here. Uh But I think once you get into it, you're going to really be you're going to you'll start to realize like A, you don't have to be super technical. And B, you're going to start to open up more opportunities that you didn't know were there yet and start to push your coding abilities further, faster. I think so. I think. For sure. So I'm a beginner. I just started the project. Liz's day 11, literally using Bolt, built an hour project. So for the people that are watching, if I'm a beginner, what is

3:47something I'm going to learn from you Sure. >> through this episode. Yeah, so I think the number one thing I'd like to instill with beginners is just a little bit of confidence to say we don't, you know, a more technical user doesn't necessarily have any more answers than you have. It's just a matter of learning what to ask and learning not to be afraid to just keep putting the onus back on the LLM to figure things out for you. So, if you don't know the answer to a question, it's not like you're expected to just say, "Hey Claude Code, hey Cursor, hey tool that I'm using, how do I use you more effectively?" Right? So, one of the things we're going to talk about today is sub-agents. It's one of my favorite Claude Code features.

4:33Uh I don't know if it's available in Cursor. I know they have background agents. I'm not sure if that's the same thing. But, we'll go into how I use sub-agents. But, the way I got started with those was I went into a code base where I'd already had a project and they just announced sub-agents and I was literally like, "Hey Claude, look at my code repo and suggest to me what the best sub-agents would be." Because at the time it was brand new. I didn't know what they should be and so you would think like, "Oh, I need to you know, the old way of thinking is I need to go find documentation. I need to find a tutorial on YouTube. I need to figure out all of the ways to make this work." And I'm saying like now we are enabled and empowered with these tools to just keep asking them

5:20to take us to the next step. Improve yourself. Like it's kind of a silly paradigm, but you're going into a tool and saying like, "Hey, you're great, but how can you be better?" And it's going to willingly answer you and improve itself. So, you kind of you do need a uh precursory knowledge of like, "Okay, what is what is a sub-agent?" Right? Like you have to know to ask for a sub-agent. Yeah. You don't have to know how they work. You don't have to know how to configure them. Um I'm sure there are power users who get in the weeds and really tweak these things. Uh, I'm in between, right? I like to go I like to know that I'm doing things more efficiently, more accurately, but I'm also not going to spend the time I'm not going to spend, you know, 8 hours configuring an agent. I I'm too uh ADHD for that. I like to

6:05dive in and just get to work. So. >> Yeah. Yeah, I love it. Well, you did this great uh kind of chart explaining it, breaking it down. So, just take us through the workflow. What are What are we looking at here? Yeah, so basically, I gave a talk yesterday and that's what what you're referencing and some of this will get a little bit, you know, maybe beyond beginner, but I think the concepts are good. So, we talked about sub-agents. So, three of the things that make Cloud Code very powerful are slash commands, hooks, and sub-agents. Um, slash commands are basically just predefined command prompts, right? So, I'm going to do something multiple times, uh create a commit, um

6:50you know, do a security audit, uh whatever. Anything you're going to do multiple times, you can make it into a slash command. You do slash, name of the command, and it'll execute that prompt. So, you don't have to type it out or say it every time. Uh Hooks are interesting. I haven't gotten into hooks as much yet. We won't cover that too much on this cuz I think that's getting to like the in-depth workflow, but essentially, anytime Cloud Code has a part of the process that it's going to complete, there's a hook associated with that, and you can add additional commands or instructions at any phase it. So, you know, if you want to say um for a while, JSON inputs uh inputting your formatting your prompt as a JSON object was uh

7:35gaining some traction. Yes. >> So, you could say, "Hey, every time I enter a prompt, intercept it, reformat it as JSON, and then pass it off to the next step. So, that would be a hook, like a user submit prompt hook. Um so, again, I don't think we'll get too too deep into that, but just know that you can really get customized workflows with this if you if you find the need to. So, like if you were working on something, you said, "Hey Brett, like you know, uh it would be great if every time Claude before it did something, you know, Like for example, when mine is done completing a task, it plays an audio ding. It just goes right ding. That's a really simple example of a hook, and it just means I can go to another tab and be notified when it's

8:20done. So, but the my favorite hands-down is sub-agents, and that's what we'll go over here real quick. So, sub-agents and why they're awesome. So, typically, you've got Claude code up here at the top, and it's trying to maintain a context window, and context window is just everything that it currently has an understanding of. So, the larger the context window, the more things it's trying to hold in memory. So, think of that like as a person, right? If you've got 20 things you're trying to juggle during the day, the likelihood of one slipping gets higher, right? So, if you have a to-do list, and you're taking on one item at a time, you're focused on that one item, and you know that the others will be there when you're ready to get to them, right? But you're going to give your time and attention to one task. So,

9:07if you give Claude code a task, and you say, "Hey, you know, I want you to build me a to-do list." It's going to go out and do some research and figure out what a to-do list generally consists of. It's going to try and figure out what the front end should look like, what the back end should look like. Uh I have API calls here. That's not always external. Sometimes, it's building API calls between the front end and the back end. Uh if you have a database, so say you're doing a to-do list, and you want to store your items, now it needs to make database calls. Um and then I just threw mobile in here as like, "Hey, imagine you're also doing this for Now, it needs to be a mobile dev." Right? So, it's it can do all things. Claude Code or any LLM for that matter is a generalist. It can handle most tasks. But, what you've got here is you've got a lot of tasks been handled by the one agent,

9:52which is going to use a lot of contacts, which A is going to take up more tokens for pay for uh billing purposes. >> Right. >> And it's also going to be more likely to confuse the LLM and give you hallucinations or less than ideal output, right? You might have more bugs because it didn't slow down to try and find the bugs because it wasn't told specifically to go test this code, right? So, you've got that. Uh Claude Code or any again, Cursor, they have tools like read tool, write, um I don't know what else they have. But, essentially, every time you call a generalist agent, it has all tools available to it at all times, which again is kind of you're throwing everything and the kitchen sink at this and hoping that

10:37it's going to come back with a with a good response. And that's where you get this output. >> I mean, and I think I'm there now. I've been using drool.new and it feels very much like that. Like it's got so much in-depth into the project that it's trying to handle. Not only does it take up a lot of token usage, but it just seems like it's kind of even forget stuff that we had just talked about and >> You have to remind it, right? So, you've got to be like, "Hey, that thing you built an hour ago, I need you to remember that you built that thing." Right. >> Uh the best is when it comes back and uh criticizes your code. And I'm like, "No, you wrote that, dude. Like, I didn't write that." Yeah. Yeah. Uh and then it works sequentially, which is very slow, right? So, it's going to go do its research

11:22phase. It's going to go figure out what it's going to do and it's going to go through each of these phases in order. Um and so, it just takes a long time when when you do it that way. So, they introduced up agents and this is uh my diagram kind of explaining how those work. So, now you've got Claude code working as an orchestrator, right? It's the project manager. And you've given it agents and an agent is essentially just another set of instructions for the same LLM, right? So, it's still going to make Claude calls to Sonnet, Opus, whatever. But now it's got In this particular case, I've got six different agents representing specialists on a project, right? So, now I've got one agent whose job is only to understand the mobile infrastructure I need. This

12:08guy's only job is to understand back-end dev, front-end, research, APIs. So, all the same things we just had. But now those have each been given to one specific agent whose only job is to do that task. Yeah. So, now they've each got specific context windows, which means they have less that they have to understand and more that they can specialize in, right? So, my back-end dev is only aware of its own back-end tasks and it doesn't have to worry about what the API coordinator is doing cuz that's not its role, right? So, you've got a separation of concerns now where and all of these preexist. You don't have to like they're now GitHub repos you can find the stuff we're like that you can ask Claude code

12:53how you'd like it to or you know, the to suggest its own agents, basically. So, specific context windows, specific access to tools. So, for example, my research tool may need the Perplexity MCP, but I database expert does not. So, the database expert is probably going to need my Superbase MCP. My mobile dev might not. So, everybody gets now their own tools, which ultimately leads to us getting more accurate results. And this is the big one right here. They work in parallel, which just six in this particular case six x is the time that it's sorry to reduce the time it's going to take to get your project done. So it separates and Claude does all this for you. It separates all of the tasks

13:39out in a way that they can be done in parallel where each agent does not need awareness into what the other agent is doing and then it's going to ultimately give you hopefully a better I was going to say the same output but hopefully a better version of the same output because of the split up approach. Does this making sense to you? Yeah, so say I wanted to take my project that I'm currently working on another platform and switch to Claude code. You could give it um you know, your files and then it can like you can ask it what are the best agents to >> Yeah, it's totally your project. When you first open up a if it's an existing code base like you've got like we'll use your project for example. So you alt dot news pushing things to get up. So you

14:24can clone get hub on your local machine. You now have the exact same access to the code that you had through alt. Um it like I said it just doesn't have the guardrails anymore which can be intimidating but it can also be liberating once you learn how to sort of flex your muscles inside of this environment instead. Um So yeah, you would start with there's a command once you're in Claude it's just slash init which is going to have Claude go through your code base familiarize itself with it and it's going to create a Claude markdown file which just has general information about the project and things that it finds relevant that it needs to know and it's going to use that every time you give a command the Claude MD file is going to be referenced. So if you want to put something in there that you're going to repeat or you know, something you don't want to have to type

15:10every prompt or whatever the Claude MD file MD means markdown the Claude markdown file is going to be it's sort of source of truth, right? That's it now understands your project because it just gave itself a something you can reference. And this goes back to what you mentioned earlier where you code something and it sort of forgets. Well, the Claude MD file is how it's going to always remember things. And you can actually give it a command. I didn't cover this yesterday, but if you give it a hashtag command, hashtag means remember this. So, you can say like, "Hey, that was great. Like the thing we just fixed, hashtag remember to always do that." And it's going to put that in the Claude MD file. And then every time a new prompt is given, it's going to reference that and be like, "Oh, right. You wanted me to remember to always, you know, format

15:56it this way or, you know, put a joke at the end or whatever." Right. Have it do whatever you want. So, this was an example I shared where I had 10 agents working at one time. This was more of a "Can it do it?" than an "Is this necessary?" Um but you can see it's got a generic coder, a back-end dev, a mobile dev, system architect. That's pretty interesting. It's cool. Uh somebody, you know, it's got an agent analyzing the code to make sure things are going well. And then a tester. So, this is something that as a developer I'm just too lazy to do on my own is writing unit tests and making sure that the things that I add to my project don't break existing functionality. I'm sure everybody's been through that, right? Like especially when you're doing LLM-based coding is it'll just come through and just knock

16:41all the pieces off the table to make that one and you're like, "Dude, this was working." So, um another pro tip, always make sure that you are creating commits. If you get your app to a working state and you're about to try something experimental or even like, "Oh, I don't know. I'm going to add this, you know, AI AI API call and I have no idea if this is going to work." Create a commit and that way you have a fallback. And we can talk about that if you want. Yeah. That's whining in the background. Um so, yeah, it's as simple as like you can tell Claude Code, "Hey, create a commit. I want you to create a commit." You can even build hooks in if you wanted to where you say, "Hey, every time we complete a feature, create a commit." I haven't gotten that uh crazy about it yet, but I do try and just keep that in

17:26mind. So, um But yeah, so it'll write power That's super powerful right there, what you just said, because I was wallowing through that right now. And then everything was working, Google sign-in authentication, and now I've spent the better half of two days fixing the Google sign-in because of the And it's not just you. Like, you go through X and you'll have This is sort of the I don't want to call it a misconception cuz it's not a misconception, it's absolutely dead-on. It's just avoidable. So, when I first started using Cursor, I would be like, "Oh my god, I one-shotted this whole feature." Right? And then I'd say, "Oh, but I don't like the CSS. I don't like the styling on this." And he would wipe it out and write the whole thing again. And now I've got a worse version of it, or it doesn't work, or whatever. And I

18:11was like, "This worked." Like, So, it was it was this like push-pull relationship where you're always trying to get it to do something else without breaking everything it had already done. So, I think the models have gotten considerably better since then. Um I don't know what I was using back then. But I still think like as a as somebody who's building something, setting those milestones that you can revert back to is an invaluable safety net. It's incredible cuz you can literally just be like, "This is terrible." And again, you don't have to be a Git expert. You can tell Claude, "Hey, roll back to the previous commit." And it'll just it'll handle that for you. So, everything Git is like the brain, but you want to be able to get that onto

18:56your local machine. And you want to work out of your local machine, and make changes. And then when you're like, "Oh, this looks great. This works." cuz I'm testing it on my local machine, then you would do a commit, right? Yeah, so if you've got things in a good state, uh and have Claude create the commit for you because it's going to give you a very detailed list of the last things that it's done. My commit messages have historically been like things are working. About to add security audit. Don't want to break the app. Uh that's so that if I go back later, I can be like, "Okay, this is where things were working." Uh cuz yeah, I've definitely run into that where I'm like, "Okay, I've built an app. It's cool. I want to launch it." Uh it's pretty common that vibe coded apps have security issues in them because a lot of times we as developers don't know

19:43what we don't know. And, you know, API keys get leaked. There's things like cross-site scripting. So, you need to like sanitize input text, things like that. And you don't have to know all of that. Ask Claude to do it. And they've actually just added a feature. It's a slash command. I don't know if it's security audit or security check, but it's slash security something. I'd have to find that. But create a commit before that because it's going to go through and rewire a lot of things to be correct in terms of security. But it may lock things down so much that like, "Oh, now there's, you know, row level policy row level access policy issues for your database." And things that were being fetched before are no longer being fetched. You should keep those security changes and move forward through the fixes, but let's say

20:30you went so far down a path that you're like, "Look, this is a mess. You know, I'm going to like you said, I'm going to spend eight hours just getting back to where I was." I would say take what went wrong, roll back to a previous commit, and adjust your prompt. Now knowing like, "Hey, this is the path we went." It's almost like you got to see the future a little bit, right? You go down the path, everything breaks, you kind of know how you got to where it was broken. And so, you're going to come back and say "Hey Claude, like first off, revert to previous commit. Second, here's my new prompt with some more detailed instructions on how how avoid that problem we just we just ran into, right? And you kind of get a chance to do it again. That makes sense. much time into. And I'm going through that and experiencing the growing pains, but yeah, it seems

21:16like you know, I think you get to the point when you're building something ambitious and there's a lot of, you know, APIs you're integrating and a lots of features. Um I feel like this is the next step in the evolution process for Vibe Coders out there. And you know, I expected Bolt to complete this project, but then I was coming up with new ideas um for the project and I was like, you know what? I don't I think this is going to be a little bit too big for um you know, using Bolt or Levelable. And I like in 10 days in I'm I'm in there. I mean, it even pops up every once in a while, you know, the project's too big. Right. >> tells me flat out like it's going to have a hard time making small changes. Like in your your um slide there, it just gets to the point where it's like

22:02having to analyze everything takes 200,000 tokens to to respond. So, it just um And you're going to hit limits. Specifically, you're going to get more fragmented results. Like the results won't be as good. So, uh it's what's funny is it's crazy that in 10 days you're like, "Hey, I've already hit the ceiling here, right? I've already I've maxed this thing out." And I think you're right. Uh I I don't have experience with Bolt. I've used Replit a little bit. Um so, I don't know how they all compare in terms of what their context windows are, but I do think you're going to get to a point where those are meant to be introductory tools to this world, right? So, you can go I mean, you can spin up a landing page, you saw. You can do a landing page, you can do database

22:47integration, you can do uh authentication. You can get a workings of an app. You get it there. But once you get beyond that, and I've been following your your YouTube and some of the stuff you're adding is great. But you're starting to get where you're like, "Okay, I need something more advanced than this, right? I'm trying to add like multiple vision models to look at an image and give me And Yeah. you're you're going far enough off the beaten path that Bolt is like, "Yeah, that's kind of not it's not really what, you know, this was designed to do. This is designed to like, "Hey, do you want to spin up a quick, you know, chatbot or a landing page or whatever?" Perfect. So, I think they're great tools for getting your feet wet. And I think this is the next evolution. What did I post

23:32something the other day? I said that to me the evolution goes like Replit, Cursor, Pod code. All right. So, eventually you're just you're going to want to be in the weeds and just sort of own this experience. And I'm telling you like once you get over that intimidation factor, you're like, "I can just keep asking this thing to make itself better." Um so, we'll get to that here in just a sec. So, this is a an example of a flash command. So, you could have something where you say, "Hey, I'm I'm in a good spot, right? Flash review my app." And review my app would be something you would have gone in and said, "Hey Claude, I want you to create a flash command that called review my app. And whenever I call this, I want you to do an SEO pass, I want you to do a security pass,

24:17and I want you to do a a UI review." And these would all be the sub agents that would specialize in these tasks. It's going to call each of these and say, "Okay, is your site optimized for SEO? Is your site uh optimized on uh security? Is it up to up to date on security standards?" And you can even have it This one's really cool. I saw somebody on YouTube do this where they set up Playwright to take a screenshot of their app and pass it back and then grade it against a rubric. And then >> Yep. it would continue that loop until it gave itself an A+. And it was just based on design standards that they had laid out beforehand. So, you can say, "Hey, that AI you or that UI you made it, it's okay, but it's not an A+.

25:02Right? And you probably even ask You could probably say, "Hey Claude, what's an A+ design?" Great. Store that in a markdown file. And then now, as part of my review {slash} command, I want you to go through a loop and keep iterating on the UI until it gets an A+. So, let's make sure all of our alignments correct, you know, all of the little animations and micro-interactions are all correct. Um So, you can really get pretty dialed in in terms of a workflow. And again, here's your output. So, my >> Your R is like the SEO specialist. Two podcasts ago, Greg had um on the Startup Ideas podcast, had the Boring Marketer on and for the SEO specialist, that blew my mind. Yeah, you're going to eventually have your site ranking. Like locally at least.

25:49>> Yep. And there are MCPs for that. I've actually followed another video that the Boring Marketer did where he talked about the I think the company is just called Data for SEO. And they have an MCP that you can hook up to and we can cover what an MCP is if if you feel like that's within the scope of this conversation. Um and it will basically go out and do complexity research and find keywords. And then it'll use the Data for SEO MCP to go look at keyword search volume, uh competitor analysis, uh they call them like keyword difficulty rankings. So, if you pick, you know, Chinese restaurant, the difficulty ratings could be very high because that's a very generic word. If you say, "Vegan Chinese restaurant near

26:34the San Francisco, you know, bridge or what I don't know, whatever." Then now you're getting much more specific and it's going to say, "Okay, this is search volume for that. The keyword difficulty is a lot lower, which means you're going to be having an easier time ranking higher." So, yeah, I mean, it gets way beyond code. Honestly, if you really want this to be the sort of Uber solution for running a business as a solopreneur or even as somebody who's trying to just, you know, you work for a large company, you're trying to speed up processes or make things more repeatable. I honestly think that this goes so far beyond just how to use it to output code. So, we're focused today on, "Hey, you know, you're a vibe coder, you're graduating from sort of a beginner-level

27:19tool and something more advanced." But, it go like it can be your marketing specialist, it can be your SEO specialist, it can be an email campaign manager. Like, you could have this thing look at your Supabase users, look at their usage of your app, segment that audience into groups that need different types of communications, and then automatically email those segments with different content based on you, do you need a nudge to sign up? Do you need a a promo because you you, you know, we've analyzed and said you might, you know, be a high-risk return. Yeah. You could really set this up to help you run your whole business. Um Yeah. It's just a matter of like, okay, at what point does it become too much for you to

28:04hold in your head? And honestly, I think you could have separate Claude instances where you're like, "Hey, you are my marketing department. You are my sales department, you know, you are my HR department." Like, you can really set this up to handle any number of tasks. So, we're using it for coding, but it's really just a window to an LLM, right? So, it's a window to that AI intelligence and getting back answers. But, then you add in tools like research and screenshots and database access, Stripe access. Yeah. You know, you can really start to make these You can empower them to do a lot. And you know, I think getting used to it the way you're kind of exploring it is smart because there's a trust level there of I'm going to give you access to my

28:49database. Are you going to wipe out my entire database? It's a little scary, yeah. Well, so uh Superbase recommends that their MCP that you use in read-only mode. I recommend that as well for at least a couple couple days while you see what it does. I've gotten to a point where I'm like, "Yeah, I trust you. I'm going to say uh you can have write access as well." Because to me the convenience is like, "Oh, I just want this." Now, I haven't been burned yet, so that I could give a different talk if one day I get burned and my database gets deleted. Uh I might I might change my tune, but uh this is something I touched on just a little bit earlier. Uh just ask Claude. Replace Claude with whatever LLM you're using. Just ask. So, the biggest tool at our disposal right

29:34now is the ability to on demand get any information you want, right? That's kind of the the revelatory moment of these LLMs. Is we can now, you know, I use Google Gemini deep research to do deep dive custom articles on any topic I want. And I'm like, "That's the most empowering thing ever." Like, this It's crazy to me. So, keep in mind that just because you're now in this terminal window, you can just keep asking. Like, "Hey, add me a slash command. Hey, what is a slash command?" First off, like, I keep hearing about slash commands. Explain it to me. Let Claude explain it. "That sounds awesome. How can I use those in my project?" Let it review your project and say, "Oh, you know what? You might benefit from a slash command that does

30:19XYZ." "Well, that sounds great. Can you write that for me?" Of course it can. So, it's just a conversation where you're teasing out of it what can you do that's even cooler. Let it explain it to you and then say, "Yeah, I'll have that." Like, "I'll take the thing you just described. That sounds great." So, that's how I've done slash commands, subagents. Um I did I gave an example here where uh I had the hook that plays the sound when my tasks are completed, right? And I was like, I don't know, like, can you just do that for me? And it's like, "Sure." And then I have to restart it, but now all my things are done, I get a little notification. So, that's you know, that's a really simple example, but >> This is Yeah, this is pretty, you know, I've been in this space for

31:05since the beginning of the year just as from the outside in seeing it happen and we're really at a point where the trending word is agentic AI, right? AI that works by itself. And I really feel like now with Claude code sub agents and 8N doing amazing things, we are there. Like, we're we're entering the agentic AI I think paradigm, right? Right. So, I would say right now, the way I would describe it is fragilely available. It is available. You can do it. You can one shot an entire app. It's going to get more and more fragile the more complex it gets, but the fact that it's possible is like,

31:50we're right there. And then you mentioned N8N, which is funny. I've actually considered, could I put in a hook to Claude code that makes a call to N8N, triggers an automation, and returns me the result, right? So, like, now you've got Claude code's typical, you know, workflow, but at any point in that workflow, you could go trigger as complex an automation as you want. And have a whole research phase that pulls down, you know, the latest tweets on a topic and ranks them based on virality and, you know, does a whole bunch and then returns that and then just proceeds to the next step in Claude code. So, from your terminal, you see nothing, but I you could get into where you're saying, "Hey, every time you get to this part in this, you know, this phase of the

32:35development cycle, trigger a full N A N automation or I think you'd go the other way. I think you'd have N A N trigger cloud code headless instances and say part of my N A N automation is to have it custom create a component for me right now and return it to me. So we talked um or it's been mentioned we're kind of heading into this like software on demand age and I think you're going to start to see things where a website doesn't necessarily have a predefined component for something. A user has a need and the website will then go create that component and return it. And we're not going to have a predefined UI where you must follow this user journey in order to get to here.

33:20You're going to come to a website and it's going to be a chat box. You're going to say, "Hey, I need to update my membership. I need to you know, see my billing." And it's going to create an interface for you right there. Now some of those will be predefined and saved, but if not, it's not going to go, "Oh, sorry, I don't have that." So we're going to see things where it's like, "Hey, make a call out to something like a cloud code agent." It's going to operate fully autonomously and return you back uh you know, a piece of content that the user can interact with. Right. So I don't think we're quite at that point. Yeah. >> But we're seeing these inklings of that and I think we're I think we're close. We're really close. Um And so this I touched on just a second ago very briefly. What about things that aren't uh aren't code? Right? So cloud code

34:07is essentially an extensible way of talking to an LLM. It's not just how do I get, you know, a React Native component or whatever. That is one thing that it's capable of, but you know, you go to chat GPT, you ask it questions, you know, you're going to say, "Hey, help me come up with the a recipe for this or whatever." This could be that. What if you had a recipe engine that got feedback from top chefs? Right? So you say, "Hey, I want you to be Gordon Ramsay. I want you to be Anthony Bourdain. I want you to be Guy Fieri. Um I want sub agents of all of these guys." And then I say, "Okay, I'm thinking about putting together a recipe for sourdough bread with jalapeños." You know, what a You could have the top personalities in that space review your recipe and say,

34:54"Hey, you know what would be awesome? You know, Anthony Bourdain says, 'You should add a little bit of nutmeg to that because like he tried that in this other recipe. It's going to be great.'" And that's a weird example, but I love using weird examples because I want people to start thinking about how else can you use this, right? We're sitting on this agent. You know, you you used the word agentic and it's agentic for coding, but it's just an agent that we happen to be used for coding happen to be using right now for coding. So, you know, it you could have a car mechanic agent that like, "Hey, my car's making a funny noise, you know, and maybe you've got a Toyota Tundra specific agent. I don't know. You know what I mean? Like there's wacky stuff. Some of the examples I gave here kind of intentionally funny like first off I put

35:40my name in every category. Yes, of course. But, this one is like, "What if you had a Willy Wonka agent?" And I know that's ridiculous. I know it is. I want people to think in terms of like get outside of even what you're already thinking of like, "Oh, I could have you know, I I use Alex Hormozi as an example. He comes up a lot in startup world, small business world. He's very popular in terms of like how to structure offers and how to get sales and leads. So, he's a he's an easy personality to target. I summarize his views, his systems, and review my landing page in Alex Hormozi's you know, criteria. Give me feedback, right? Uh Greg Eisenberg, you know, we all in our community we we follow Greg. So, hey, have Greg's input on my app and you

36:25know, am I on the right track? One of my favorites that I've used recently is my Rick Rubin agent um from this sort of like hippie-dippy like, "Hey, man, like what would be the ideal use case or the ideal user experience for this app?" Right. >> They come in. I want them to have that like Rick Rubin Shangri-La moment of like Zen where they like, "God, this app is perfect. It just suits me and I don't know why." Mhm. So, have somebody like that who's just like a renowned creative look at your app. You know, I I again throwing Leonardo da Vinci. I have to imagine there's enough information out there on him that you could have the have your agent build a profile that represents his kind of thinking, right? So, it is crazy. Think about Willy Wonka. If you're if you're strolling through your

37:11app, how do you introduce these whimsy these moments of whimsy and delight and Right. you know, a user touches something and they're like, "Wow, that was really a magical experience. That's cool." And you know, I'm using everything towards like how it relates to an app, but I honestly think you start thinking about this in terms of what other areas of my life could I use guidance or feedback or, you know, finances um >> Right. exercise. If what if you could make a thing with the pop fitness influencers Yeah. >> and have them collaborate and give you a workout plan. That's pretty cool, you know? So, the sub really jazzed. I love it. This is the next step. So, would it essentially be another agent, let's say a Rick Rubin

37:56agent, or how does that fit into the workflow? Yeah, so I typically have been creating these as sub agents that I can just call on a project. So, um I think I did not cover this. Give me 1 sec. I I've got a little section for this. Um So, there are three levels when you're talking about how these work. You've got global and this is for slash commands, hooks, and sub agents. You've got the global, which is this is available to anytime you open up Claude code, it's going to be available. You've got things at the project level, which is something you're going to maintain and check into your GitHub repo. So, let's say you and I were both working on Easy Flip, and you created a series of sub agents that make sense. That's going to be at the project level.

38:41You're going to check those in, and then if I check out your code, I'm going to have those agents well, because we as a team have decided this is a a good subset of sub agents to have for our project. And then at local, this does not get saved to the project. It's still at the project level, but it's for me personally on my development environment. Maybe I really like for it to play a whistle noise when it's done, you know, something silly, and that would annoy you. So, this is only local to me. So, each one gets more specific and overrides the previous, right? So, global's going to be for everybody. So, I have my Rick Rubin agent as a global agent, cuz I want him to be able to review any project that I'm that I'm taking a look at. If I'm building a mobile app, I might have a React Native specialist

39:26agent that's a agent for that project. And if I'm doing something specific to my project, I might have an agent locally that just, you know, is literally only for me in my development environment. Uh I work by myself, so I don't have to worry about that, but if you were to if you were to start building out a team, you would do things at like a project and a global level, and then independent developers can do local. Um And so, yeah, I create sub agents for the people that are of interest to me. So, I've actually created a round table. I'd have to go look and see who's in it, uh cuz I had Claude like it you know, which is also Again, all my all of my answers always go back to I'm already sitting on the tool that'll fix it for me, right? So, I say, "Hey Claude, create me a group of sub agents

40:12that would be the end-all be-all round table in terms of like SaaS applications, right? So, a security specialist, a marketing specialist, a branding specialist, a positioning specialist, like whatever. Yeah. Create me a list of 10 people that if you could get them all in a room, you couldn't afford the consultation fee, right? It would be It would be astronomical. And now, when I'm at a point in my app, I say, "Have the round table review my app." And I'm going to get a combination of perspectives from all of those different people. And when I say people, obviously, I'm referring to some summarization of of what they're known for. So, it needs to be something where you know, you can have your agent go out and be able to find information. You know, I couldn't say, "Give me the Bo Johnson agent that's going to give me feedback." because it's going to be like, "Oh, well, you know, I'm not sure

40:57who Bo Johnson is." so you know, I'll just make something up. Yeah. I figure if there's a if there's a public profile out there, Right. have it go summarize that and build into a markdown file. It's crazy. So, okay, that's was my next question. You know, the sub agent of Alex Hormozi would then create go and research who Alex Hormozi is and then create whatever knowledge base in a markdown file within your project, so it can refer back. Yeah, and I've only asked Claude to go do that. But imagine like you wanted some more control over that. Like Alex Hormozi has different systems that they follow, different, you know, he's he's great at systemizing things that you can apply to your business to

41:42you know, increase leads, increase sales, you know, right? The grand slam offer is a big thing. So, like take a look at my pricing page. Tell me what kind of offer I could give my users, my my clients that's going to inspire them to sign up, right? And he's going to He's going to look at it and go, "Okay, yeah. You're offering Easy Flip AI, right? So, boom, you give them 100 free credits off the top." And it's like, "Whoa, that's cool, you know, it cost you nothing because you're gaining a customer. So, the actual usage with the data is pretty small. Um you know, it's going to say create a a quick online course for how to flip and how to make a profit flipping. And give that a value. Say, "Hey, this

42:27is a $1,000 course I would charge you for. It's a 2-day weekend, you know, it would be $1,000 to attend. I'm giving it to you for free today if you sign up, right?" And so, it's going to help you think of those types of things that are going to have the psychological triggers to encourage sales and encourage conversions. And then that's obviously that's one piece of it. So, you take that concept and you abstract that out to like, "Okay, well, what else can I use help on?" Copywriting, SEO, um image generation Right. Yeah. Anything. There's so much. Um I mean, we can probably do a whole episode on that, but it's cool because yeah, you got to think outside the box. I never even thought of that. And just having almost like a like a council, you know, from some of the

43:13leading experts in that field. Um yeah. Dude, you you you can create a you could create a Vibe Coding Coach. And you could say, "Hey, I'm stuck right now. Have the Vibe Code Coach look at my app and give me a recommendation on what I should do next, Just whatever problem you're faced with. If you think an LLM could answer it for you, have it giving it access to your code base is that next level of like you could you know, if you go ask ChatGPT, I watched again, I was watching your most recent video where you're taking your Bolt.new code and you're putting the errors into ChatGPT, and then you're trying to take the code that comes back out and put that back into Bolt. Yeah. But

43:59by doing it this way, it is already immersed in your code base and can see everything. So, it has a full picture, right? You're When you're giving this to the chat GPT chat window, it only has the context of this is the error. Right. >> It doesn't know how the app is also built. So, it's going to give you its best guess. >> Yeah, yeah. >> Let's say there are five ways to do something. It's just going to pick a first one that comes to mind and say, "Hey, like, I don't know. This should fix it." But it doesn't know that that would only fix it if you'd also made these other five decisions, right? >> Right. >> Putting this inside your code base and saying, "Hey, now you have Now you know everything I know." Right. Yeah. Fun So, I I do this all the time with pod code where I I go paste an error

44:44and it gives me It's so confident. It's like, "Absolutely, we fixed it. This is incredible." And I go, "That's it." And I go, "Same error." And it goes, "Oh, yeah, of course. I found the issue. Now it's fixed." Right. >> Same error. And so, what I'll ask you is I'll say, "Hey, slow down. Do a comprehensive analysis of the of the all of the code base that surrounds this problem. We're focusing here, and we keep trying to fix a problem. It keeps coming back. There's something more systemic here that's causing that." So, being able to tell it like, "Hey, one, pull back your view your view a little bit. Look look broader. Two, assign sub agents to even troubleshoot this, right? Like, the same way they can output better code, they're also going to understand better that's a

45:29React Native problem. Put a React Native agent on it whose job is to specialize in that area. And also have a a code in it code analyzer that's going to go through and try and find the problem and then assign the problem to the React Native code agent. You know what I mean? Like, get these guys working in tandem. And so, I find that that gets me unstuck a lot is especially with React Native. I don't know if you've done anything with mobile yet. Uh I'm I'm experimenting with it now. I'm by no means an expert at it. But I find that a lot of times it's like there's uh compatibility issues with different libraries. Like you tried to put in an animation library that was for React, not for React Native. They don't They don't go together. And that's the type of thing, too, that

46:14I can be like, "Hey, remember in the future that if you're going to add an animation, don't use this library. Use the one that we just found out that works." And then it's going to store that kind of as a lesson learned, right? So >> Yeah. um Yeah, dude, it's pretty wild. Um I get real excited about it. It's almost like overwhelming to try and think about all the things it can do. So I would say for somebody like yourself getting into it, this is all meant to be exciting and cool. Don't try and do it all at once. Don't try and dive in and like overcomplicate what it's doing. At the end of the day, it's a chat window. Ask it for things that you're asking right now, and just it's going to be better because it can see all of your code. Right? And then from there, ask it like, "Hey,

46:59how can we speed this up?" Right? Like and it'll go, "Oh, I can make sub agents that work in parallel." It's like it's almost like I wish you would have already done that, but it doesn't, you know, it doesn't have that level of awareness. So you just hit say, "Hey, how can we speed up this development process? How can we make this process less bug intensive? Like less bug laden?" And it'll say, "Oh, I can probably create like a unit testing agent." You say, "Yep, that'll be great. Create a unit testing agent." Like, "Do it." Yeah, so with it let's just use the what I'm going through right now. If if I have downloaded my project files from GitHub, and I'm ready to start with Cloud Code. Maybe you're watching this right now, and you're in the same position I am. Um what would be the next

47:44step? Would be like obviously installing Cloud Code, and then giving it access to my files, and then start analyzing them to start optimizing the workflow and you know, all the sub agents. Is that like the next step? >> Yeah, honestly, the first the first thing I would do is I would open up the project up Sorry, open up Claude inside the project and do an init {slash} init. Yeah. And that's going to create a local to that project Claude NV file. So, you Claude has the global NV file, which is its general instructions of, "Hey, how to be Claude? How to What to do?" In that project, it's going to then map out what it can learn about your project. And it's going to have more of more localized context. So, start with that {slash} init. Cuz now any commands you ask it are

48:30within the context of this code base. So, start there. Um I would absolutely say that the next thing you could like just leapfrog to without even having to fully understand is say "Analyze my project and create as many sub agents as you as you need that will help speed up the development workflow, uh fix bugs, you know, whatever." It's going to give you back a list of what it thinks. Just give it the go ahead and say, "Yeah, create those." So, you'll then have close out the instance and open it again because once it's created those, they're not pulled in until it loads on when the when the app initializes. Right. >> Um so, if you try that and then they don't work, then you probably just need to remember to cancel out like close it and then open it up again. Okay. But dude, that would be enough right there.

49:15That would be enough to get you so much farther along than I think honestly even like your typical live coder is you now have an agent or an army of sub agents that are going to specialize and work in parallel. So, your work is going to be faster and more accurate. I will say, let me caveat that by saying the token usage is the same or less because of the managed context windows, but you're going to go through it faster because you've got 10 agents working at the same time. If each of those uses 60,000 tokens, you've gone through 600,000 tokens. Yeah. Now, my feeling on that is you would have gone through those 600,000 tokens anyway.

50:00It just would have taken you an hour instead of 10 minutes. Yeah. But, Claude uh limits by 5-hour windows. So, when you're saying so, it's almost like developing slowly, you're less likely to hit your contact like your limit, your usage limits. By doing this quickly, you're going to get your results faster, but you're also much more likely to hit that time-boxed window limit. So, I started off on the $20 plan. Uh and I was hitting the limits for that immediately. Um I did a $100 a month plan, which is 5x the usage. Right. I'm a really heavy user of this. I mean, I'm normally doing two apps at at a time. Uh I normally have multiple

50:45uh terminal windows open, and each of those apps is running sub agents. And I was finding that I would hit it anywhere from two to three hours into the five I don't know why they do 5-hour windows, but it's kind of nice cuz it by comparison with Cursor, you can use your entire monthly allocation in a day. And be like, well, now I've got 29 days of free usage where I can get back to using like the premium models. Cuz I kind of like the way they do it. It's a little bit more micro-managey, but you're way less likely to accidentally shut yourself down for a month. Uh But, yeah, after a couple times of going, "Oh, man, I've got 3 hours to wait till I can use this again." I was like, "I'm just going to bump it up." So, I'm on the $200 plan. Um And again, it's just where you're at in your journey. It depends on where you're

51:30at in terms of what you expect to get out of this. The amount that I'm using it, I think $200 a month is a steal. And I realize And I realize that's purely intellectual. That's purely objective of like or yeah, subjective is that to me, that's a deal. If you're trying to get the most you can out of a free plan and you're like $200 a month, that's insane. You're probably right. Like, don't don't do it. But, just know that these tips and tricks all somewhere in that spectrum of the more you leverage them, the faster you're going to hit those limits. But, again, it's not more expensive. So, I want to be clear on that. You see a lot of stuff online where it's like, "Oh, this model's expensive. That model's expensive." You're using the same about, you're just

52:15allowed to do it way faster. So, yeah. You know, you be prepared that you could hit limits. Uh for somebody like yourself, you're trying to build a profitable app that you're going to put on the web, I think you're getting thousands and thousands of dollars worth of development effort for 200 bucks a month. To me, it's a no-brainer. Absolutely. >> But, if you're just, you know, if you're just tinkering with this stuff and you want to see if you can make a to-do list app or, you know, a solitaire game or whatever, then yeah, 200 bucks a month isn't going to be worth the cost for you. But, you'll need to figure out then where you're comfortable at price-wise and what your expectations are in terms of what you get out of it. Um I found Heroku's pricing to be more frustrating. I don't think it was any

53:00more expensive or cheaper. But, I would go, "Oh man, I'm on a roll. I'm cruckin' I'm cruising." And I'd be like, "Oh, you've hit your limit." And it'll reset literally like 28 days from now. And I'm like, that's just crazy. So. That's crazy. And that was only I had a 100 color limit, so I could have bumped it up. I figured I would have hit that in another 2 days. Yeah. So. Um All right. Well, I think that just giving the basic concepts and understanding of what Cloud offers to a person that has an idea, but doesn't know what to do with it. I think that you know, starting off with Bolt or Lovable is like a good step. That's what I did. Build the landing page, the workflow, the sign-ins, the Superbase, and and you can

53:46connect all that. Get your >> Well, I think just somebody like you, too, you realize what you're capable of, right? So, having a tool like that that gets you over that hump of what you can actually do, even as a non-technical person, Yeah. is worth the upfront cost because it's liberating. You're going You're like, "Oh, I'm just a dude, right? I'm just a dude at his desk." Like, but now in 2025, I can be a full-stack developer? Like, I've never heard of that before. So, yeah, I think I think if that's a a toe dip in the water for people to check it out, absolutely. Do it. >> Yeah. Flex those free plans as much as you can. V0 is great. Yeah. >> Lovable, Replit, Alt. I mean, there's so many of them out there. Just go try them out. See if something, you know, really just tickles you and it's like, "Oh, man,

54:31like I like this idea. I want to take it further. I want to push it." And push it till like you do. Push it till it's going, "Hey, you know what? Like this project is too big or you know, I'm not You're not You've now learned what you need of your app and you know when you can't get it. So, you've gotten to a phase where you're like, "Okay, I now know what I'm not getting out of this. So, I now know what I want to get out of the next tool." Yeah. Yeah, the the thing that I kind of confuses me is when using Bolt, it's not like working on a local level. Mhm. It's cuz now it's telling me it can't access files. It can't edit ENV files. Well, I can do that. So, what am I accessing when I'm using Bolt? Is it

55:16this all in the cloud or Yeah, that I'm not 100% sure. I don't actually know where they store things. My guess is that's part of like the payment, is that they're storing files for you. Cuz they can't be live referencing I mean, it can be stored in GitHub. I can't be making like live changes back and forth between that. So, yeah, my guess is if if I was I'm going to 100% speculate, is that they're pulling that down, spinning up some sort of virtual machine that's going to host your code, Yep. letting you interact with that virtual machine by the LLM, making changes as that's how they can host them and preview them. And then when you're done, that VM probably just shuts down to free up the resources for another user. Again, 100% speculation, but that's that's what would make sense to me. Is that they just give you allocated resources on a on a server that's going

56:02to run your code for you and let you interact with it, but Uh-huh. Okay. Which also then, you know, you're you're you don't really know what your what what are what is your usage look like, right? Like what's a project that's too big? You know, like, okay, well, >> Yeah. you know, you're probably going I mean, my project is growing, but I feel like it's in a fairly reasonable It's a fairly reasonable expectation for an app to like grow to the level that you're at. You're like, okay, well, then like, am I Is my code out of control? Is it too big? Like Yeah. Um yeah, I I think I think taking ownership of it is that next step of saying, okay, I'm comfortable, I've gotten started, but I now know there's more that I need to do that this tool can't do for me.

56:48Pull it down and start tinkering with it. You know, it's it's low stakes, man. Like all of this stuff is such low stakes at this point that for 20 bucks a month, you can get enough credits enough credits to just experiment with things. See what's out there. And if something catches fire, then it's a no-brainer. You put a little bit more money on it. But if it doesn't, then just have fun. Like tinker with it. Make make mobile games. I made a mobile game this morning using chat GPT GPT-5 because I just ran this code into it, too. And I was like, make a side scroller with the power-ups and obstacles that I have to hop over, and it went through a couple bug iterations, and then I had a fun game. I was like, this is silly. Like, I just made, you know, a pretty basic game, but like

57:35a few years ago this would have been an App Store smash. Go. Oh, oh yeah, 100%. It we're in in a time right now where even just a year ago I had the idea of what we're building now, but I didn't have the access to the tools like what am I going to use? I know there's cursor and all that and I didn't want to do it because it looked intimidating and it and it and I'm just like I don't I have no idea. So I even hired a developer to use a I think it's called bubble. Yeah, bubble. >> us to build it out and then I was like, all right, let's do this and just was hesitant because it's like, you know what, what if it doesn't work? This is going to end up costing thousands of dollars, tens of thousands of dollars for this idea. And I just kind of put on the back burner. Well, then also too for somebody like yourself, you don't feel empowered

58:21to do anything about it. Right? So you're like, okay, I've spent $10,000 having a developer get me to this point. And now there's a bug or there's an issue. So I can either suck it up and throw another $10,000 at this and hope we get to where I want to be or I have to eat that sunk cost and bail on the project. Right? So when I say this is low stakes, I mean like you're putting in, let's say you're doing $200 a month. In by comparison, the stakes are so low. And you own it. So you're going hey, like I can work through this bug. And when I'm done, I feel more empowered the next time to work through the next bug and through the next bug and you're more learning the development process as opposed to the syntax.

59:08So you don't necessarily have to know how, you know, how a Python method is structured. Right. >> to know like, okay, I want Python to handle this because it's got some library that can crunch numbers or do machine learning or whatever. But you're going to learn more about You're a process overseer now. Yes. >> And just implement. You're going to have agents that can implement for you. And honestly, I think we're I think we're fully democratizing the coding skill set itself. Mhm. And you're going more into Can you ideate on a on Can you iterate on an idea? Yeah. >> And can you pushing it to let it to evolve it? Oh. You know. And using these tools in a creative way, I think it's

59:54going to be the the lever, sort of the moat to to separate yourself from like the dabblers the dabblers on on GPT. Yeah. >> Which is great. I'm not in any way diminishing that. I think these tools are incredible. Mhm. But to get to like oh, I'm doing stuff that people aren't doing yet. I think that's where you can differentiate. Think outside the box and ask you to do stuff that people aren't aren't doing yet. Yeah. And we get ahead of the curve. We are in that paradigm shift. You never know you're in that shift until the change has already happened. And I Just where I live and, you know, kind of my friend group and everything, I They ask, "Oh, what what are you up to?" Cuz I know we resell on eBay and we do live shows and stuff. And we we I was

1:00:39just so pumped with AI. And I just see kind of their eyes glaze over. So, like, it's it's very We're still really early. Yeah. >> And my My wife is very tolerant of these conversations, but she's not interested. She's like, "That is great. If you're It's like talking to your parents. You're having fun. Sweet." That was awesome. Yeah. >> And I do at times feel like this um sort of crazy Nostradamus like, "Guys, the future is here, you know, like But genuinely believe I said this many times that this is bigger probably than the than the advent of the internet. And I know that's like a ridiculous thing cuz people are like, "What? The internet? That was like the world the most world transforming technology ever." I'm like, "Yeah, it was." Right. >> And now you're literally going to have I I I

1:01:24like I said, I don't even think we can conceptualize yet the things that are going to become possible in the next 5 years. Right. >> another year from now, where are these large language models going to be, right? Like In >> In the last year, they went from like, oh, that's cute, it can kind of code Yeah. >> to like oh, people are building Scary. you know multi-million dollar apps and stuff. Like, that's crazy. It still takes a little bit of skill and a little bit of know-how to push it and get that. Yeah. Right. Consider too, we've now going to we'll have this recursive loop of we now have the tools to make the tools better, to improve the tools even faster, right? So, that's like, you can now use cloud code to make the next cloud code and it's going to make it 10 times faster and better. Right. Yes, but so, it's

1:02:11it's cool and it's cool talking to other people in the space like yourself. Just to know that I'm not the only one, you know, um and I'm I'm fired up. I love this stuff. >> Dude, I love it, too. I love your enthusiasm. I've been watching your videos and like I said, I love that you represent the non-technical crowd >> Yeah. that just wants to see what's possible because I still think there's this uh stigma that like, oh, it's for the techies, right? It's for the the uber nerds of the world to hop in there and, you know to do this, but >> Right. I would say it's the opposite. It's not it's not making the gap larger, it's making the gap disappear. And I love that you are representing that and showing that to people that are like, hey, are you a person with an idea? Like, there we are. Push it.

1:02:56>> Yeah. What happens? Yeah. And as a driver of the community Like, you don't have to mortgage your home, you don't have to do anything crazy, just try it. Put it out there. Yeah. Yeah, it does. And then, yeah, absolutely. And to kind of just wrap it up with the things that you're building, where can people like follow you along? I'll I'll put all your stuff linked down below, but >> Uh you yeah, you caught me unprepared there. I am awful. Uh, I'm on X at Red Juggernaut, so it's just same first and last name. I'd say that's where I post the most things that I'm working on that I actually uh, just share status updates or screenshots of. Um, I've worked on so many disparate apps at this point that I don't have any one online presence that really represents any of this. So, if anything takes off, uh, I will happily create some social media around the brands that do well. But yeah, just to

1:03:41get my take on stuff, I'd say that's where I share most of my development thoughts for projects I'm working on. Um, thoughts on the the landscape, the AI landscape in general. Yeah, that's probably the best spot. Yeah, I mean, if you have questions, reach out. I It's just like it's such a an open community. It's like there's no reason to struggle by yourself cuz we're all kind of going through the >> And I get so excited to talk about this stuff. I'm always telling people, I'm like, I'm not Ask me anything you want. Shoot me a message. Ask me a question. To your point before, it's like, my friends and family don't want to talk about this. Yeah. So, if you have a question, oh, I'll be so excited to to talk shop with you. So, shoot me shoot me any anything you want to know or just share thoughts on something or whatever.

1:04:26Tag me in a post. I love it because I can't get enough of this and uh, yeah, this is great. Um, so yeah, continue to just keep doing what you're doing. I'm going to keep documenting the journey. There's going to be more podcasts coming. Um, we'll have to have you on again. I mean, the space is moving so fast. Uh, you know, if we have you on in a month from now, I can't imagine what's the new model that just got released. >> Yeah, let me I'll tease just a little bit. The thing I want to work on this coming weekend or next week is I want to see if I can put um, GPT-5 into Claude Code through API access. So, right now I love Claude Code as the tooling, but I don't want to be married to any one model. I want to be able to flop to whoever's the the model de jour

1:05:13and if GPT-5 is better, I want to figure out how to wire that into Claude code's interface so I can cuz really it's the tooling I like and the LLM is great. Don't get me wrong. Opus 4.1 is spectacular, Sonnet spectacular. But if they if GPT-5 is better, I don't want to have to wait for Claude to pass them. I want to be able to use whoever's the best. So I'm going to try and figure that out. Yeah, I just saw a post this morning about it. Um it was it says Claude code pro tip. The GPT-5 API price is ridiculously cheap. Use the repo prompt MCP server and you can have Claude sending coding tasks to GPT-5. And I like oh Wow. Isn't that crazy? Yes. >> Like I'm telling you just just picture

1:05:59like just go in a perfect world this would work like this. All right, then see if you can figure it out because like the tools are there now, right? Just ask it. Just pushing it. So yeah. Crazy. Wow. I really enjoy your channel. I love I love watching what you're building. I'm super excited. I'm rooting for you and I can't wait to see some Claude code start whopping up on your channel. My next video my whole my whole head of hair will be gray. Hey, when I when my kids were born I had no gray in the beard so I feel you. There we go. I appreciate it Brett. Well, enjoy your weekend guys. I hope you have a blessed weekend and we'll see you in the next one. See you Brett.