AI Vibe Coders Are NGMI in 2025 Without This Set of Skills

What's up? I'm Parker Rex. I led tech for a startup that sold for 23 million bucks. It was great. After that, I tried Airbnb for music. Then I went all in on AI. My goal is to hit 100K in profit a month solo. And then beyond that, build a hold around it. There's a whole thing, but that's not the point of the hook. I've done stuff. Here's a bunch of photos. I did the product. I did the tech. I did some of the designy stuff for that startup. I've done a bunch of different things. I tried to make like an e-commerce brand while I was there. I've gotten rejected by YC six times. flew out there once, pitched the founder of Gmail, didn't work. You learn a lot of these things as you go, right? Everything from like Android apps and managing deals with Verizon so you can get thousands of tablets and like having a bunch of different apps. So yeah, on this channel I give you the daily strategy on what I'm doing and then that is proof that I know what I'm doing. Cool. Let's go into the comment belt. I'm gonna talk about that in a second because I love the comments we're getting and people who have the best ones deserve a little recognition. So, I'm gonna modify their face and put a belt on it with this belt. So, that could be you. What was I doing? I was reading some comments. And so, I have two channels and I don't post on one of them, but starting next week I'm going to add three videos a week on the main channel and then we'll have these ones. So, these are 30-minute videos and we do a little bit of news, QA, strategy builds. So, here's Andrew trying to sell me on YouTube SEO. So, like Andrew, you got to work this. This doesn't work. Like, I I would never read this, no offense, and say like, "Yeah, I want to talk to Andrew Russell official04." So, yeah, I don't I don't really think I need YouTube SEO optimization. I'm not focused on maximizing the viewership per video. I'm focused on maximizing getting good at making videos. It's that 100 reps thing. Watch a Mr. Beast video. He'll explain it to you. Okay. And by the way, I don't mean any, you know, I'm just this is constructive criticism. You guys, if you want to win the belt, you should yap at me and tell me what I suck at so I get better. That would actually help me. That's that's that's uh good for me. So, Indian guy who's nontechnical wants to start an AI service-based business which helps businesses to one save money, save time, or get more money. Yeah, there you go. Cool. Fundamental stuff. What should I do or where should I start to the first consistent 1K per month? Okay. Well, you probably suck at stuff right now, right? Like that's the reality of how service-based businesses work. That's the reality of how algorithms work. Like people blame algorithms for not getting views, but it just means that you're not good. And that's tough reality. But the way that you're going to get good is just like anything. Literally sets and reps across your whole life. like that idea of sets and reps supplies across your whole life where if you are bad at the thing, do the thing a lot until you're decent and then it's, you know, people want to pay for it. But what I would do is I would just make free stuff for people ideally in a space that you enjoy doing because you'll get better at it. There's a Naval Raakon podcast on this recently, but yeah, it basically is Iggy Kai, the Japanese philosophy around like can you get good at something that is fulfilling that provides value for other people. If you're on the center of that ven diagram, then you're dialed in and you're happy. So then it doesn't feel like work. Yeah, I would just go and binge stuff. I'd get technical. I would try to figure out like something that you personally want to build. That's like the trick. And then ideally the thing that you want to build solves other people's problems. Like if you want to get good at the thing, make a thing that makes you good at the thing. Make a little chatbot that teaches you how to learn the thing. People that want to also learn that could then buy it from you or you can open source it because you give a value away for free and then people want to be a friend. So yeah, two two steps basically. Find a problem in your own life, go solve it with tech, and then start yapping about it on the internet. And then as you've done that process a couple times and the different problems that you have, then you'll be able to take on other people's problems. I think learn basic nad may make start experimenting, build some basic level and start outreach. Don't stuck in tutorial hell. Don't watch a agency gurus. Yeah, I agree with all that stuff. and listen to Daily Parker. That's awesome, dude. Tech Tahash 662. You gotta You're a big candidate for this belt. I'm gonna put a belt on your head. This is good stuff here. Yeah, this is all right. Hari says, "Get technical." Yes. And it's ironic because Hari, this is like a big part of this video is like I felt so stupid because I tried to vibe code stuff because I didn't do the prep because I was tired and it literally just ruined the quality of my own project that Hari is actually coincidentally helping with. But yeah, get technical, figure out the basics. You have a 247 tutor just sitting there waiting for you. You have a bunch of PhDs sitting in your computer. And if you don't have money, find aedu account. You get a month of free chatgpt right now. So, can we get more videos with the suit? Maybe. Kind of like the outfit. Yeah. Hey man, thanks for your input. I have found some pro the problem and I have come up with an idea, but the thing requires web scraping off sites like Facebook and they don't let you do that. Sorry if my questions are vague. Yeah, you can use ampify. Shout out tech to hash is for sure getting the belt by the way. There's a couple ways you could do this. So if I went to like Facebook and get ready to dox myself stupid all the stuff on Facebook is but there's a couple ways you could do it where there's this little Pokemon ball up here. You see this Pokéball? I click this. It's called instant data scraper. So if you're doing like low frequency, low quantity stuff, you can see that it already drew stuff around this. And this might be weird because it's like a mix of different types of posts, but it's like grabbing a single post. If I wanted to try a different table and I can hit that. Oh, I hit table now. Boom. It's already hitting all these things. This is what your export would be. And then if you want to locate the next button, like if you're on something that's pageionated, this is an infinite scroll. I basically just locate the next button. I don't think there is one. So maybe you wouldn't be able to do this because there isn't a next button. Mark next button or link. I don't know how you do it with this because I don't know if there is one. Well, let's see. Start crawling. Oh, I should have hit infinite scroll. Yeah, so basically use this thing. It's a Chrome extension. And then this can do like you know I've pulled thousands of rows with this. So use that or use Apify. Go watch one of Nick Surve. He's one of the few like make people that isn't an annoying shyer. But Apify basically lets you spin up a VM or like a virtual machine and then you can type in the stuff you want. It'll go scrape and scale. So that's what I do. Cool. Next. There was one other question on a different video. It was more of a comment which was just, you know, I loved it. So, this guy might get a belt, too. There's only one belt. Sorry. Sorry, Funel. There's only one belt. But you are building in public like nobody I've ever seen on major on YouTube. Major props. Thanks. Yeah, honestly, this is an accountability thing, right? It's like, you know how you tell your friends you're going to go to the gym and then you don't and then you're like a bad person? I'm putting social accountability on my income and my reputation for the whole world to see. So, you got to have cojones, but who cares? like yellow, right? God, that's cringe. But it is true. It's like why would you not? So, let's get into skelly draw. I have a like ultra ultra music festival playlist going right now. I think I finally have to pause it. Yeah, I just did. So, now I can actually kind of think. So, yeah, we're going to give you the comment belt. I'm going to do that at the end. You'll see. You'll get a free avatar. It's super super awesome news. You can get Vert.Ex AI through Google Cloud provider platform. I went into this in my community yesterday like walking you through all the steps, but just get client. Get client, open it up. The problem is with Vertex, like sorry, this is for Gemini 2.5 Pro. You can get unlimited for free uncapped. Ridiculous. And there's a lot of setup stuff. just go get client and then read the instructions and then yeah, you'll get unlimited uncapped stuff. It's a little bit confusing, but hopefully you can go figure it out. I'm not going to go into in this video. Now, there's one other piece of news that I want to talk about, which is just that even people at linear now are talking about how dope this is with how they're implementing their features using cursor. So what he does is he copies out the original issue context and starts generating unit tests for the desired markdown format. Right? So it's like here's the unit test ahead of what we're going to do. And this is so great to hear from someone way smarter than me because Kenneth Schkovix can't even say his name. He's definitely a better engineer than me because I've only been doing this for like three years and then on and off for like 10. So but that's like front end. It's not even programming. but he's definitely better than me. And to see how he's doing it, it sounds good because this is what I plan on doing. So like current flow with vibe coding and what you want to do is you want to do the planning, right? And then you want to add the testing. So this brings us to the next topic. It's all going to tie together where if you don't plan ahead of doing your vibe coding, you just jump in there and you're feeling lazy and you think you're going to get lucky. Well, you might get lucky a couple times, but then you're going to get screwed on the third one or the fourth one. It's just not worth it because you won't actually get a download of what's going on in a codebase and competency is an all-time premium. You get more leverage on every minute that you spend in an IDE when you actually know what's going on. And I did this literally two days ago where one of the viewers, Hari, wanted to he reached out and he's like, "Yo, like this sounds like a cool project and I want to jump in." And I was super excited and like but I also had my plate full for the day. So, it's like 900 PM and I'm like, "Okay, I'm gonna go rip this real quick. I'll just like jump in there. I know exactly what to do." But, I didn't take the time to actually remember the context of the codebase because it had been two months since I pushed. And in the last push that I had made like a fool because I didn't have a CI/CD set up properly and I probably vibe coded that gosh. But, I put it in a broken state where some stuff wasn't right and I had deleted. I had some config in a database that I had deleted. So like I left it in this state that was bad. So then instead of doing the planning of being like where's this project at right now? Let me run it. Let me go back a couple commits and see what's going on. I just jumped in and I'm like vibe like go. So it's like that flamethrower. The stakes sitting there and I just blasted the hell out of this thing. And then I'm like oh yeah it's good to go. And then we went and ping ponged back and forth for like four hours solving something that I've solved a million times that's not even really a problem. It's like off and it's not like like I've rolled my own off and I'm sitting there and I'm just so frustrated because I had prioritized speed over a little bit slower but quality. And it's just such a a good reminder when you have something like this happen because probably have now done probably dumped four hours into basically wasting time when I should have just done like okay do the normal thing. What if AI didn't exist? How would I approach this and then bring in AI? So that's that's the big concept. And so when I look at this post, I'm like, okay, well, how would this be a part of my flow? So I have the multistep and I do this manually. These are just prompts, right? But eventually as I get these more and more refined, there's like six of them now. There's the three around feature creation. There's three around debugging and tracing and PRD. And now I'm adding in testing. So, in any typical project, and this goes for a general framework when you're making agents, agents are there's like a couple definitions, but I always go with like, you know, you have like multiple steps, you're chaining them together, they can run in a loop, they can run independently, and they have access to tools. So, that's what an agent is in my head because you have multiple steps, multiple tools within each step, dependencies, you can have stop gaps, you can have self loop, self iterator. In the case of this, when I think of an agent, I'm like, "Okay, cool." Like, do the thing manually, right? And so, do the thing manually a bunch of times until it's good. Do the thing manually, have the outputs graded. So when I run this and I'm like okay like this is step one of many I'll put the output in a G sheet and stare at it be like okay would I write this like this and a lot of the time be like whoa like 80% of this is better than what I would have written as a PM who was technical did this and scaled a company to $73 million a year like and this thing is way better at this part then this other part sucks and so you tweak you refine that's why getting good at prompting thing like this is literally all it is is it's like when you're writing a prompt it's like okay you need the role you need the task and then context context can be docs it can be examples it can be websites it can be scraped websites it can be audio files it can be images it can be video it's crazy if it's multiple so in this flow and how this plays into it is I have right now I have one which is a researcher I have R2 which is another researcher I have F1 which is feature step one. I have f two feature step three. So it's one, two, three. And so then you also have like this would be for going from one to 1 n and this would be for oops make more sense. This would be that flow. So you're still doing research to be making that new thing, right? It's a very similar thing if you're doing a zero to one, like you don't have a thing. But then the third one would be bug. So you have like a thing that you're fixing and then up here would be zero to one. Zero to one would really just require more around the research and then tweaking F2. So F1 is technical PM, F2 is solutions architect, F3 is incident or sorry executional engineer, the person that's like doing the atomic steps that are all the output. So each one of these has an output, [Music] right? And each one of these is just a prompt. So that's why you want to get close to the prompt. This is why I hate when people who don't they're over opinionated and undereducated. And that's like most of the people in AI that are at least yapping on the internet, but they'll be like, "Oh my gosh." And also newbies, they'll be like, "Wow, cursor is horrible. What happened to cursor?" It's like, "No, dude. It's your skill issue." like 4.5 just came out. A new model just dropped and cursor can't even figure out how to use that. So use the one that's good. Same thing with Gemini. Gemini, they have not figured out how to get it to run in Cursor. So just raw dog it in client. It's way better. And so anyways, on the bug side, you have B1, B2, B3. And the way that you fly with these because you still want leverage. When I say manually, it's not like I'm typing this on a freaking typewriter. Like I'm doing a bunch of these bad boys. I'm just like boom boom boom. And yeah, I know I have a magic keyboard. Why? Because I had the Moonlander and that was just too much. And then I have another split which I need to actually go back to. But anyways, you're doing keyboard expanders with that and then all that. Yes, it's in my school. There are things that I gate. Sorry. Takes a lot of work. So in the case of this testing thing, we're going to be adding that into here. So I'm going to be placing tests likely in between the solutions architect. So we'll have T1 in here. And this will probably be unit tests. And then there was one that my buddy who I did the Airbnb for music startup with, he's a really good engineer when it comes to tests. He called it like the golden test. And I'm pretty sure that's just like end to end. Someone let me know if I'm dumb. But my idea with this is then you'd have your unit test here and then you'd have end to end tests at the end and then these would go back and forth until it's done, right? And these would not be in the same. So I guess it go F3 to let's just call it and then test on that set of new file changes checking the discs and then it go into a QA. So it' be Q1 and then from Q1 we're going to use MCP to then do the get thing because I set that up in client yesterday and it just bangs. It's ridiculous. So I need to use the G1 agent which has GitHub MCP. Another one that I'm thinking of changing in here for F2 which is the solutions architect is the sequential thinking. Yeah, it's called sequential thinker. And what this does is it's absolutely ridiculous, but it requires the model to step through up to six methods. Methods is not the right word. You could use that if you weren't a nerd, but I don't want to use that because it is a different word in nerd world. But up to six approaches, rather. And that means that it's going to go and it's going to come up with all these things that it could do in order to accomplish the new feature you want and then stack rank them. So after it does the fourth one, it still has context of all three and then it's like, whoa, four is actually baller. Let me do four. And so it's built in iteration on top of chain of thought because chain of thought is part of hybrid if you're using Gemini. So this is what I'm building now. And gosh, does that look confusing, but I promise it's not. Like each one of these is literally a keyboard shortcut and you go bang bang bang bang bang. And then each one creates a dock and then fires it off to the next thing to then create the output. And then finally, what's going to happen at the very end? This is literally going to be a team like a bunch of different engineers and people on a product team. This is my product team. After this, then you'll have your GitHub CI CD. And in here, this is more like I got to figure this part out. It's unknown, so let's make it square. But there's got to be some sort of thing beyond the normal like tests that you can run in there that would be higher leverage, but I'm just not sure. So, normal test would work, but then it goes live, right? And then let's assume that it's live. And when the changes are live, well, you got to let people know, right? So then it's text Parker preview link. And those aren't a Versel specific thing. You can set those up wherever, right? So I get the preview. Actually, no. If it didn't go live, it do sorry preview and then it go there. And then if I approve then live this assumes it's only front end and like if you're doing back end as a front end which isn't the only stuff I do we have dedicated back end and map that I'm be doing but that would then bring up like if you need DB stuff then I need to mark that. So this is for me just to kind of figure out but I always use superbase because you can self-host and gosh that looks bad superbase and you'd load this up with l psql.txt. So that's like you take everything in the postgress superbase related knowledge base and then you make the llm.txt. They don't have one for postgress specific which is strange. They have them for their like SDKs and I do that and then I do the superbase MCP and these are just like this is a piece of context and then this is a tool right set of tools because in here it can do reads and then what else would it need access to in this agent I would probably give it a tool around generating types. So as this got better, I would actually make it a proper agent, not just a prompt. It's like, hey, you can go generate types. And that would go and run a bash command that is like this. It' be like bun db types. And if you're familiar, that's just like shortand that's a script in your package. JSON that does something like superbase gen types. And then you like pass it off to database types and that gives you front end to backend to DB types. It also gives you syntax and like an LSP that we associated with it for both writing migrations. And then for the types there that's for like when you're writing code and you want to see like auto. So yeah, I'm pretty excited about this and I just had seen that and it was like oh my gosh, more validation. Cool. This is what I talked about a little bit, but like f vibe coding and it's just if you go in with only that like do the planning. Don't do what I did and waste a whole lot of time on something that's total waste of time. Other stuff that I'm doing strategically is I just there's going to be a mobile app for map. I when I launched the GPTs for devs which was 50 GPTs on the chat GPT website. All I did was I used a scraper. I scraped this is before firecall existed. So I made a scraper that I forked from someone else's project and just modified it. But it would go and it would go to every single Apple doc that was something that I needed to learn. And this would be an example for the guy that asked the question about like how to learn how to learn is you build something so you can go faster. So what I did was I went and I made Apple developer docs and I scraped all of them that I needed and then I made little GPTs by just literally uploading the outputs and then splitting them. So doing like at the time it was lane chain morf but I use that to split the JSON outputs and then I'd take those and I'd load them into GPT and then I launched it got 100,000 plus I think got like crazy amount of hits and people were using I still get them like emails today should reby the domain but point is is when I put that out an Apple developer that worked at Apple hit me up and he's like yo this is sick like can I help you with this? And I was like, "Sure." And then he ended up building us a satellite app, which is on iOS. When I say satellite, it's it means like companion. And that's another thing that you need to know if you're doing like multiplatform, is if you do an iOS app, I'm sure you know, but you get a 30% rake from Apple on anything over a million dollars, which like is a lot of money. Like making a million bucks off your app and then 300 G's just like falls out of thin air. like you're going to be upset. So, the way around that, and you'll notice this if you use Kindle, when you try to buy a book, it's like, "No, go to the website." That's because it's a satellite app. It's a companion app. They don't allow you to do commerce there. And so, we did that with map from the guy who found my project through Reddit. He was like, "Yo, so we're going to spin that back up." And in order for it to work, there's this project called Midday. And I was talking to Pontis, the guy who makes this. He also made, if you're familiar with Cursor Directory. It's got a bunch of stuff on here. Pontis made this site and it's sick. It's like he's crushing it. I'm sure it's it's not like his it's going to have like some crazy enterprise value because it's just an info directory, but he jumped on the trend real fast, spun it up, and it's beautiful. He's getting, you know, 4,000 hits a month. This is similar to web, by the way. So, with him, he has midday, and I love the patterns he's doing. He's doing crossplatform stuff. What he is using that I'm not is I would go native iOS. I wouldn't go expo, but I was talking to him this morning on Twitter and I was like, yo, when are you doing next 15? And he's like, boom. And he sends me this freaking branch and I'm just like, holy smokes. Because if you're trying to learn again, learn how to learn, go read this codebase. Like the midday one's epic. There's a lot of controversy around like if you follow a guy named Dax on Twitter, there's like a lot of Next.js haters, but I think it gets to a point where you're like, "Okay, which one do I like?" And you don't, you know, you just pick a thing and you go with it. So, I pick next. And I'm open to trying client side stuff because I definitely like will do that on future client builds where it needs to be like if I'm not in a rush. But I'm looking at this and he's got TRPC11 in here. And when I asked him, he's like, "Yeah, it's because it's going to make it a lot easier to work with the API across different platforms." Don't totally understand, but I see that even just the kind of way he set this up instead of having like a use user context or the state stuff and all this like it just looks a lot cleaner. And I haven't gone through all of them, but every one of these actions is a little bit different because obviously next 15 has different things. This is another reason why five coding without prep will screw you and ruin your day and potentially your whole project because it won't know like oh you got to awake cookies because that's a new thing now. So that's what I'm looking at now is would this be required? Would it not? Also like standardizing doing an ETL on the data because we have different providers. I need to be thinking about that. And he has something in here called an engine. when you connect your bank accounts, the way that Stripe, the way that Plaid, the way that go-kardless, the way that all these different providers bring back their information, their data, like one of them might use underscores, one of them might not have capital letters, one of them might just have a different name. So, taking that and then extracting it is the E and ETL and then I think it's transforming it. So, transforming it to look like the way that you want it and then loading it somewhere else. So he has an engine that does that and he did it in hono. I would not do it in hono but I have built I literally rebuilt it in hono and now I'm not do that in hono. Cool. So other strategy stuff. So I did the 5K and 60 minute challenge yesterday. I failed because I'm stupid and I didn't have Gemini 2.5 Pro set up. So I just sat there and waited for a long time. I could have done it in 60 minutes had I done that. And then other things just like loading in a bunch of images and like clicking around in folders. Who clicks around in folders anymore? You just write a script. And by you I mean Gemini 2.5 Pro does it and you're like, "Yeah, do a safe transfer. There's a script that you can write. You just say, "Hey, I want to do a safe move for this or safe copy." And it will safely copy everything. Save me some time. Community is growing nicely inside of AI, which is cool. We got some smart people in there. I kicked everybody, all the freebies, all the freeloaders. Sorry about that. But yeah, no such thing as a free lunch. And it's awesome now because people actually care. Like I I that's what it bothers me about all these like free communities is like it's glorified Facebook group. It's like that's not what I'm about. I'm about bringing in people and ultimately like it holds me accountable to make stuff and then stay on the edge of what's possible and then also you get use cases out of it. It's kind of cool. And that brings me into I'm just trying to think of everything as a flywheel. So everything that I must everything that I do must play into the next thing. So in the case of the daily upload, it keeps me on the on the front of what's going on. So at night, I go through and I try to figure out like what's worth actually talking about. But then that brings in people that are interested in like what I like to talk about, which is AI. So then I get friends, which is cool. And then some of them might go to the paid community which is vibe with AI. Some of them might not but I don't really care. It's more so like getting the compound effect of getting all the skills around this and then other random things happen where like I'm going to be open sourcing map which means that you're going to want to have people know about it. So then you can talk about the things and then you'll get more devs and more people that are interested in it that can help you. And it's not just for like oh I'm gonna make content so I get devs. It's more like people that come in that are interested in this. It already is a filter. Like you're going to get a certain type of person. They made it, you know, 32 minutes into a video of someone yapping about very specific field. Like that is the filter, right? So you're not going to get like a bunch of lowquality people and you're going to get also like because of that filter you're going to get diverse sets of opinions which sounds fluffy but it is not like I can only think of so many things through my perspective and then someone can come in and that's feedback that's oxygen like feedback is literally oxygen to the stuff that you're making you can't breathe without it and so having people come in and give you that is awesome like it's so so it. So, yeah, I just think that like this ultimately like then gives you more input to then do the next thing which then will spin out and then it's like we're going to make a SAS out of the use cases that we see in the community and we're going to build it live and then we'll put that content in the community and then we'll have more stuff to talk about in the videos and more reasons why people should join because they want to make a SAS and then people will eventually like I could just see this going nuts and that's the plan. Like I have a Mount Rushmore of guests that I want to have in what I call Vibe TV and it's like how do we get Elon? How do we get Satia? How do we get Sam Alman to like come in and yap? And if that's on, you know, the three people of however many there are on Mount Rushmore, it's like well you have to do a lot of cool stuff in order to get there. We're in early days. This isn't like a flash in the pan thing, but it's important to like kind of think of that and then the trick to it to continue doing things for a long time is to make it fun. So yeah, then on the builds for Vibe TV, I'm gonna open source this too, but Pool Suite is the inspo for it. So it's like how do you make a site look cool and my sound's muted, but it just started playing music. It looks cool. Looks like a desktop. So something like this with a window manager in it and then like the free content that I'm doing and then ideally people that also want to make content within the Vive TV network. It's gonna feel like MTV like vintage 90s. Um, but building I built a window manager like mini OS by following someone else's project. That's how I learned. I like I rolled this. This is before even vibe coding the thing before agents were a thing. But yeah, there's a guy on YouTube that made it in public and it's like a really popular project. But that's what's going on there. And then on the map thing, yeah, this is what I yapped about before, but I hadn't touched the repo for two months and then I made a stupid commit and then now I'm changing everything and making it awesome again. So yeah, that's the video, guys. I was supposed to do this in 30 minutes, but I ran the motor a little too long. Um, comment the word motor if you want me to send you a free motor. No, that's the video. Like and subscribe. See it.