I Built a Free AI Vibe Coding Starter Kit So You Don't Fail (ya free)

Hey there, I'm Parker X. I led tech for a startup that sold for 23 million bucks. After that, I tried Airbnb for music and a bunch of different other SAS startups. Those didn't work because they're silly ideas, but what did work was me learning how to leverage AI every single day. And on this channel, I give a daily update on my way to 100K a month in profit. And that's a split among AI services and a SAS product that I'm going to be launching soon. And so, we cover a couple different things. We're on the road today, by the way, if you couldn't tell. But we're going to cover a couple different things today and you're gonna want to stick around for the whole time because it's really good. So, first of all, this vibe coding thing. Here's a $3 trillion CEO. Man, that's a lot of work. So, let's say I I just want to vibe coat it. He says vibe code. It's kind of wild that that's a thing. He's literally talking about vibe coding and it's such a trend. And you can see with Vive coding and Vibe marketing over the last 30 days, Vibe marketing is starting to pick up against Vibe coding. Now, of course, it's much more niche, but it's definitely a thing. So, I just wanted to point that out. And there's a lot of videos going around about how important it is to have the fundamentals of programming down before you start vibing because you won't get that much leverage. And I've been saying that since the beginning on this channel, but there's a bunch of examples going around where in this case, this guy is showing you how he made a graph database that represents all of the code that he's writing. So he has a way for him to graphically show his codebase. So I started recording this part and I didn't realize that I'm actually talking over him. So this is going to be a quick voice over of it, but as you can see here, this is a bunch of nodes on a screen. And it's similar if you've ever used Obsidian, but it's a visualizer for all the different parts of your codebase. Different idees or code editors like Zed and Windsurf and Cursor all have different ways of indexing the database or your set of code and then it traverses them in different ways. So what he's done is he's taken all of the code and throwed it in graph database and then it will graphically show you the areas that you want to work on and then algorithmically set step through all the different areas you should check out. That guy, he's going to release a beta version of it, but it's really nice and it shows you how he was able to throw that together really quickly because he's already competent. And then the second one is a guy who's talking about how he was able to take a very old codebase. They use different patterns written in different syntax, different flavor, even different languages as a use case and then use clawed code with tons and tons and tons of context. So this would require you to do a spec prompt. This would require you to already know what you're doing in order to get that one shot. So there's different tools for different people. I always call them the training wheels tools. Like when you're learning how to ride a bike, you have the training wheels on. lovable, bold, those types. Those are good for those types of folks because they don't really have an opinion. They don't even know how to code most of the time. So, how are they gonna have an opinion? Whereas this guy, he comes in and he has exactly what he wants to get done. He is able to then put that into words and put that into snippets and put that into very defined steps with the interfaces with the data structures that you want and then you get that magical oneshot with cloud code. He also talks about how Grock 3 is like a mathematician in a box, which I think is a really good way to put it. How you can then take a thousand lines of code, put it in there, have it worked through the algorithm over and over and over again. And next I talk about how Superbase is releasing declarative schemas and OMS are object relationship managers. It's kind of like a a schema language that allows you to more easily work and connect with your database. also lets you declare the schema that you want to write in Typescript. So you do a constant with the name table or whatever operator you're trying to do. And I think it's a good step in the right direction because drizzle was very popular. It's very newbie friendly. And by doing this with Superbase, they're definitely going to catch a lot more users. Export const. So you're making a variable of table or constant of table and then you're declaring the thing that you want which is a postgress table and then you can define what type of row and then what are the things around the row is it going to be an int is it going to be allow for nulls or non-nulls are there foreign keys involved so it's kind of this other querying language that you'd end up learning a lot of people don't like it I think it's nice that superbase is releasing this because at the end of the day, this is just another thing that people want for the newbies. So, I also think that there's probably an advantage to having it be more readable um just for like version control purposes because migrations can become kind of a a nightmare if you have a bunch of them. So, this is something new that they added which is nice. I will likely be adopting this in a future project. Last is that VS Code dropped their C-pilot agent mode. So, I haven't even run it, but it is kind of cool that they're doing that. And then they're supporting the MCP protocol. So, let's see if we can't open up a co-pilot. Um, co-pilot. Command shift I. I don't even know if I'm on the newest version. Let's see. Check for updates. Yeah, I don't think I am. And as you can see, I'm on the road today. So, probably worth checking out, [Music] though. Next up is a new model dropped called Quar Alpha. And the hype behind it is that it's a foundation model that is not being named. So that's kind of wild that they're doing that where it could be open AI, it could be Grock, it could be whoever, but it's somebody that does not want to be known and they want to gather more training data. So good for open router for getting access to this. I was testing it. It is number three on a lot of the coding benchmarks. So why would I use number three? Well, because it's really, really fast. So I guess that's an advantage. But at the same time, I don't think people prioritize speed. I think prioritizing speed on a model when you're coding, it's it's not that great because it actually kind of tricks you into letting it just take off and you won't take the time to read it. Something there like you you even notice like the easier that it becomes to vibe code, the lazier you're going to get, the more likely that you're not going to take the time to think through it. You'll get a couple quick wins when you're doing like stuff that doesn't matter, like small like front end things, but anything that has like logic, anything with meaning, you need to take the time to think it through. And it's there's just things that are worth learning. And it is a trap because it's so easy. Like I had this happen with MAP in the video I explained yesterday where it's like, oh, I should have taken an hour to redownload this map database to my head or a codebase instead. Wasted a bunch of time. So, okay, that's on the new side. Let's look at some of the questions that came in. This guy says, "Please avoid yapping and get to the point." This was an older video and I said, "Yeah, I love running the motor, but I'll improve my ship to yap ratio. Thanks, Mr. Lunar Clock." And that was on my old channel or my main channel with a video I posted a few months ago. But that's good feedback. Let's see what else. How's this video doing? Marlin says, "Great stuff." Thank you, Marlin. Thanks, dude. Let's see those videos [Music] fly. And then what else we had? Tech Tahas says, "Gemini 25 is insane and it's almost also also it's almost free. My broke problem solved." Yeah, if you host it on Vertex, it is free. So, go do that. And I explained that in yesterday's video, but you basically enable billing. You set up a cloud profile with Google Cloud Platform and you'll get a ton. What else? Uh, this guy asked, "What did you sell for 72 mil?" I didn't sell anything for 72 mil. I was not the founder. I said that at the beginning of the video as a hook where I said I led the technology for a startup that sold for $23 million. At our peak, we did 7 uh $2 million in sales. So doing the sales each year doesn't equate to the exit sale. And also there's just general notes in that where it's like a marketplace doesn't have high profit margins. So it's really tough business. If you have razor thin margins, then your multiple on your business when you sell isn't very high because it's not a desirable business model. There's a lot of NAN experts on YouTube, but not many AI and NAND experts on YouTube applying it to their craft in deep and technical ways. the former is uninteresting. Yeah, totally agree. I think that that's also one of the challenges with media and code is you'll have people that are really good at code but horrible at media. And though that's what I'm striving to do is be on both sides of the aisle, so to speak. And then CT hunnel 95 says, "Working on building a small marketing agency since that's what my background is. Thinking of ideas on how to build my own SDR. SDR is a sales development representative." So somebody that would go and work on behalf of closing more sales. Toss this on ideas for future vid. Most businesses can benefit from SDR or BDR in one way or another. Yeah, you immediately you win the belt for the day. You win the comment belt for today. Best comment by far. BDRs are definitely needed. The challenge with this is that you can't mimic a personal like relationship. So whenever I see these BDRs that are closing sales from first touch to close like wired, I just wonder what average ticket, what niche that's in. So it could definitely work. I'd have to think through what niche would make the most sense for this. It' probably be high. It wouldn't be I don't think it'd be high ticket unless it's like insurance, but insurance isn't really high ticket either. But it's a really good idea. I totally agree. And I'm curious what kind of marketing agency you're going to be going after. What kind of marketing agency are you focused on building? I go into this further in my vid that goes up [Music] tomorrow. Cool. And that's all the questions for today. So, let's get into some of the strategy. On the strategy side, there's a couple things going on. So, on the map side, we're going to be end we're going to end up uh you got it. You just stay on this. I just didn't know if I should be this side or not. No too. This is just for us going to be open sourcing map because I think that's a great moat for the future. I think that it gets people interested. I think it's a way to there's a lot of different benefits. The way I look at it is you get goodwill with the community. You get developers magnetized towards it and you get better ideas. So I think that's good. I think that in the era of AI abundance, they open source makes more and more sense and there's probably some sort of profit share around it. You see this with Gumroad. They recently did this and went open source and they have bounties on their page if you were curious. You can go and try to spin up their project. They use Rails and um and Next, I believe. But yeah, you can go to their page and they have like here's $3,000 to do this feature and you know, you can do that. But I think we're gonna really benefit from going open source on it. There's some work that I need to do over the weekend after this wedding to get it to a place where I can have other developers working on it. We tried to do that prematurely. Someone reached out, his name's Hari, and we pingpongked back and forth because I had left the project in a bad state and then tried to vive code my way out of it. Horrible idea, but I've rolled back and now we are making it or I'm making it in a place where it's going to be a lot easier to work on. So, that's exciting. And it'll be a turbo repo. Some of the stuff that I'm inspired by is this Project Midday. So, it's got a lot of good patterns in it. It's really well done in my opinion. I don't think that he nailed it necessarily on what to build because I don't think he's making money on it, but I think that uh his his code's good. And then that brings us to this idea of a turbo AI starter kit. So what I'm doing is with the help of Hari and whoever else wants to get involved is I haven't found a starter kit that I like and we're not going to sell it but it's just something that will be the source of truth for bootstrapping any new project and everything like saying it's an AI project I think is moot. I think that that's just this tweener word that we're using right now, but that that'll be a part of every project. And so having the best practices in there as a playground and then having other developers who want to do a PR of it makes a lot of sense. So you can kind of pick and choose. It'll be like a buffet of the best options. And that will be a turbo repo. You don't have to use Next. We're going to probably have a React router app in there. We're going to have the packages set up. And so far he had just basically forked the superb basease with next starter which is nice. And then I put in a bunch of the docs. So some of the docs in there are around like having a prompt library, right? So instead of it being somewhere else and then you having to bring it into a project, you would just always reference your starter kit. And in there there's the multi-agentic workflow from going from zero to one on new features, you just go straight. And also the 1.0 to 1.10. So any sort of future improvements as well as a workflow for debugging as well as a workflow for testing. So you can see all of these and right now I preach this a lot but before you try to automate anything it's just like Elon's five-step algorithm you have to go and do it manually and in this case it's not totally manual you're still using prompts but we'll have analysis and planning we'll have for new feature work it'll be a folder and it's just research so there's research one research two you should do two researches bas based on you just go straight you you just do one research if it's something really basic, but a double research is really nice. I can show you an example of that where I was messing around with this idea for map where it's like, well, how can you stand out? How can you make productivity fun on a one pager? And I think if you could take a lot of the patterns from majorly or massively multiplayer online games or like MMOs like World of Warcraft or Guild Wars or like kind of the OG ones that that'd be interesting. And so this is one example of the research phase where I'm going through and I get a few different documents and I'm kind of like using it as a thought partner to tease out what that initial technical PRD would be. And then as we go you get into the feature development. So deep research for context and PRD. Then PRD plus architecture doc plus atomic steps. Atomic steps being the one thing that you can fit into one line. So it's a task can't be build a website. It needs to be make the header like this D. It's like broken down. Then we have a debugging flow that includes tracing is step one. Step two is root cause analysis. Step three is atomic steps to fix and then actually fixing them. And then finally is on the testing side. So I'm not sure how I want to bake this in, but right now it's just towards the end, but I think I want to be writing new functions and methods using tests and then write the functions and the methods. So this might move up. This why I think that nodebased planning and architecture or workflows makes a lot of sense. like the way the NAN works where you can kind of glue things together and see it visually I want for this. So it's kind of like maybe langwise or something, but basically just take like a a flowchart, apply these prompts to it, see what the outputs are below it, and then that way you can say, "Okay, actually the thing that I want works better if I take the unit tests and bake that into the feature flow versus doing it after the fact." And then that would make it a lot easier than just like this as like file tree looking thing. Um then end to end test, golden rule test, QA in the browser using browser hand. And lastly is the deploy step. So the deploy step is going to be inclusive of MCPS just because it's convenient. um where it's like I am trying to realize that some of the technologies that we have today won't be necessarily there quite yet but it's worth exploring how it could work cuz it will get better with the adoption of MCP some people will hate on it and say it's oh it's middleware oh it's uh you know insecure whatever but I think that this is here to stay because Microsoft's supporting it open AAI is supporting it it's a new proto protocol that the biggest companies are adopting. And so it's worth seeing, okay, well, when we get to CI/CD, can we just literally talk to our codebase and then have that be the pickup of the tool? You have the GitHub MCP that can literally listen to that, run the diffs, you can run tests after the diffs, you can do all sorts of stuff that then has conditionals set up. So you can sequentially have it go from be like, "Hey, like we've gotten to the step where we want to deploy and it would have a a tree of decisions that it needs to make decide whether or not it's going to go out um and then yeah, you'd have it pushed to your preview branch or preview URL and then message the person back and then you can say, "Yep, looks good." Because at the end of the day, all of this is just so you can do one thing. So that in two years from now, you can text your phone and say the idea that you have or you can talk to your phone and say, "This is the feature idea that I have. Go make a branch and do the whole thing." And you only get there by methodically doing it manually over and over and over and over again. And there'll be a scale where you'll get to the level two out of 10 difficulty features quickly, but I think it's it really like unlocks what you're capable of doing cuz it's going to also change over time. Like a level 10 difficulty thing that you thought would never work will become the seven and the six and the five as these models get better, which means that we as builders can push the limits of what's possible. So, very excited about that. That all comes in this Turbo AI starter kit. If you want any access to this stuff, then drop the word turbo below and I will get in touch with you and we'll figure out how to get you guys access to this repo cuz it's fire. So, next up is we talked about that. Oh, we have some stuff going up on the school. So, Vibe with AI, it has 50 pieces of content that are all video kind of like this, but like in detail and very very tactical about like this is exactly how you do it. We cover general productivity with the AI. So, the 23 tools that I use every single day to kind of fly around and the keyboard shortcuts so that when you're doing stuff in your computer, you getting more leverage. I basically teach you how to talk to your computer, which is what all the smartest engineers are literally boiling it down to. Like, there's a reason Andre Karpathy is better uh better at vibe coding than you. It's because he just knows how to talk to the computer. And so that's a basis for a lot of the content that is in Vibe with AI. And so what I'm going to be doing is I basically just like take the best content and I put it behind that payw wall just to be totally frank because I also want to have a community of people that actually engage. If you do a free community, there's no engagement. So we pushed up a bunch of new stuff and I think it's going really well. We got a few new members. We're going to hit 10 soon. And then here, it's basically daily AI tools to master. So, I update this. I'm on a schedule to put three new pieces out a week. And we just added a new call. It's like a scheduled Yap sash where we can just like run the motor and talk to each other about stuff. And I added in what did I add? There's four rule framework is how I'm kind of thinking through. And I'll give you the high level, but you should definitely join where I'm just teaching you how to learn to talk to the computer because there are dozens of PhDs trapped in your computer. They're stuck in there. You got to get them out. And you need to learn how to know when to tool hop because a lot of the time it can be a total distraction. The rule of thumb for that for me is basically just like during hours when I'm trying to build something that's high leverage and I'm trying to build something that will make money, I'm not tool hopping. And then other things that I added, this is actually free, you don't have to join, is this episode road map. So people will be able to go in and add what stuff that they want in here. And I think by open sourcing this, we'll get better ideas and then it'll be more interactive and people will actually get it. And then I also made this extensive learning thing that I'll be adding things to, which is I'm not your guru. I know a lot more than a lot of people about this. I get that. But I also know that there's other people out there and there's other content that I'm consuming. So I want to give you access to that. So when you're on your Tik Tok or you're on your social media apps that you're passively kind of getting that information, there's podcasts you can be listening to, all those sorts of things. So the new piece that will be going out is me coding up a contract that I'm delivering. And I had originally done it in web flow because that's what they wanted. But then it was about 9,000 times quicker to just hand code it because that's just the truth. And there's also some customization stuff in there. So that'll be a really high fidelity nice build that will be in there. And lastly, just to kind of close out is I've just would love your guys' opinion on this kind of idea of like taking popular patterns from the best psychological triggers in video games and then using that to mirror your own life. So I it's funny because in my head goes towards this idea where oh maybe maybe they built Second Life too early. Maybe they built the Sims too early because they didn't have the technology and they didn't have the same tools that we have access to. But I really do think it's interesting that you could take like if you ask yourself what makes a video game addicting is this false sense of achievement. But real life achievements so much better. Everybody knows that and you don't get any sort of guilt and you can dump hours into a video game and you get this fake achievement. I love The Sims. Yeah, The Sims is fun. But what makes me interested is this idea of I'm already like the first version of MAP allowed you to pull in all of your live data. So, we're pulling in data off of what you can pull in medical records. You can pull in all these things like around your biometrics. You can pull in your goals. We could pull in literally anything on on the internet. So you could go so far as having like your live bank accounts in there. You could have everything about your life in one spot and then instead of quests, which are video game terminology for like projects that you need to achieve to get XP in the video game, it's like how do you gify life in such a way where you just basically are chasing that best version of yourself every day and it doesn't feel like work. So, I've tried different ways with this. So, I was exploring last night where I'll just show you and uh this will be how we kind of wrap, but I was exploring different ways how you could kind of achieve this and I was using Gemini and at first it's like okay I kind of prompted it around this idea of we're making a piece of software like you're responsible for building software for children because children don't understand like all these productivity systems right they haven't like filled their brain with garbage around like I don't know productivity porn. So it's like if we go the other direction with it, we're like hey you're responsible like if we think in that way where you don't even speak English. Let's say that you don't speak English but you need to know what time of day it is and when something important is coming up through symbols through sliders through interactive 3D graphics then you end up with some interesting stuff. So, this is just one concept where it's like, okay, there's the sun and the sun moves and it turns into the moon at night and it has the live weather and you have the actions that you need to take in the form of icons and then these would be dynamic to the thing that you need. So, if you're trying to lose weight and you have a goal around that, then just by looking at this fork, you can see either maybe a meal plan or you can track it if you wanted to, you could take a photo to get it. Like I view all these little utilities that are making like thousands of dollars online as like little features within here. Um, another exploration that I was doing with this was literally having it be a full-blown like video game that looks like an MMO and it didn't get that far. Let me see if I can't find a previous version, but it was using websockets and it allowed me to like see you can see at the bottom there's like a heads-up display. So, it had the information about where I was. I would imagine you'd actually see like maybe an avatar of yourself, but it's like hyperrealistic to what you actually look like and then one that looks like you in the future. So, if you had a fitness goal, it's like maybe if you're fat now, then you look skinny then. And it's like, hey, go do this thing so you can look more like this version of you. And it's through the eyes of your future you. That's like what's kind of meta about this whole idea is it's not like a random AI talking to you. The AI is just a reflection of what you want. So, it doesn't come off as offensive or anything. And I went so far as to finding a repo that has someone online was trying to rebuild World of Warcraft. You can see the last post was four years ago, but I'm going to get this running and I'm going to use Claude code and gets into the specifics of basically bringing this into a modern stack. but he uses websockets and he has all these different worlds that are built in. So there's just a basic server with your world client, your world entity. It's like all kind of old school stuff. But it is cool cuz you can see like you know you have health, you have all these different models and then you have the different like shared stuff. So terrain constants, noises, very very cool stuff. And it looks like a game. So there's definitely a balance. I'd love to get your guys' opinion on that because if it feels too much like a game then at the end of the day like it's directing you to go what like do the work. But it is kind of interesting where it's like if you could just frame each one of those hooks that are in video games that people get serotonin dopamine release from as things that move you forward, that'd be cool. Um like an NPC, you walk up at you to it and you talk to it. It's like, "Hey, here's what you need to get done today." and then it links out to the different apps or the different things that you need to do. So, this is definitely geared towards people that are knowledge workers. I know that the mobile version would be a little more challenging, but yeah, drop your guys' questions below. Check out the the school and like and comment and you can guess where I'm going if you're good at that uh that game. There's like those world mapping people that can like take a screenshot of like a sign from back here and reverse engineer it. Anyways, all right. Thank you for watching and I'll see you in the next one.