How AI Is Replacing Entire Dev Teams in 2025 | Vibe Coding EXPLAINED

Hey, welcome to the Next Wave podcast. I'm Matt Wolf. I'm here with Nathan Lands, and today we're talking vibe coding. In this episode, I'm going to break down exactly how I've been rebuilding the Future Tools website completely with AI and how you can build your own website. We're going to break down a whole new business model that's emerging because of these AI tools and show you a business that's doing $1.4 million a year from one single creator. I'm going to break down what MCPs are, what APIs are, what all of the acronyms are, and we're going to deep dive into how you can build your own businesses and products and different tools using this AI vibe coding concept. So let's go ahead and jump right in. Today we're going to talk about vibe coding, one of the hottest buzzwords in the AI world, and it's all about vibing and coding, I guess.

How would. How would you define vibe coding, Nathan?

Yeah, you know, when I think of vibe code, I think of actually, you know, Riley Brown, because he was the first person I saw, like, talking about vibe coding, really? Like, he was sharing all this stuff online. We're like, hey, I'm just, like, chatting with Whisper Flow, and then it's transcribing what I say into text, and I feed that into cursor, and then it makes something. And he. And he's kind of like, describing that whole process, how different that is versus, like, you know, you sit down and you got to code everything by hand and you got to really think it through and. Yeah, it's a different process, I think. So when I think of vibe coding, I think of, like, you know, it's. It's a more, you know, in theory, it's a more relaxed process. Right, where you're like, literally could be. You could even, in theory, be doing something else, you know, like in the evenings now, I find that, you know, when my. My wife is, like, taking her shower and getting other stuff ready and other things, I'll be, like, hanging out on the couch, maybe watching something on YouTube. Educational and vibe coding. Right?

This is the same thing.

Yeah. So has, like, become a common thing for me where, like, now, like, in time periods where I probably wouldn't have been productive in the past, I'm actually productive during those time periods.

Yeah, I. I mean, I'm. I usually do it in my office. I got a sort of double monitor set up here. Right. So typically, what I'm doing is on my main monitor, right? In front of me, I've got, you know, my coding app. Open my ide. I'm usually either using Cursor or Windsurf. I sort of switch back and forth. We'll talk about that in a few minutes here. But I've got my coding going here. And I've got YouTube typically opened up on my right screen. And I'll give it a prompt of I want it to go code for me, and then hit enter, and then I'll press play on my YouTube video. Sort of watch for a minute. Notice out of the corner of my eye that the coding has sort of finished. Pause my YouTube video. All right, let's check out what you did. All right. It's not quite working. Right. Go fix this. Enter, press play on my YouTube video again, go back to watching it. And then. And. And so it's kind of like almost like you can code passively. Right. Or, you know, I'll do the same thing. I might hang out in my living room, grab my laptop, and just be sitting there on my laptop doing some coding while. While the TV's going. But, yeah, I mean, essentially, I think vibe coding is using natural language to write code. So I actually looked this up on perplexity to see how perplexity would define it. And it says here, coined by Andrej Karpathy in February 2025, this method involves describing software ideas in natural language, allowing AI to write the code, and iteratively refining it through conversation. So basically just talking to a coding app, saying, here's what I want you to create. Go do it. Letting it go and write the code, you double check it, and either saying, cool, that worked, or, no, that didn't work. Go fix it, and just going back and forth using natural language until eventually you get the app that you wanted to get out of it.

Right. And there's been a whole conversation on X around this too. Like. Like, engineers are really skeptical of vibe coding and, like, turning it into meme, like, in a negative way. You know, like, they're talking about, like, vibe coding, you know, the systems for, like, how planes are flown. Like, you probably don't want that.

I agree, I agree. I think there's definitely critical, like, infrastructure areas where you don't want to hear that it was vibe coded. Yeah, I don't want the software that the air traffic controllers use to be vibe coded.

Right, right, right. And a lot of times right now, the AI, when it does make things, there are typically, like, small bugs, and there's probably things that are not the way that you would have hand coded if you were like an expert. I kind of believe over time those things are going to just get fixed. Like the models are going to get better, so they'll be able to go back and like refactor your code for you. But, you know, it's definitely a big debate on X right now. It's like engineers are kind of taking the other side.

That's called coping. I'm just going to put that out there.

That's.

That's called coping.

Yeah, I think so. Yeah.

Yeah, I, I mean, I. There's very little that I've wanted to get coded up that I haven't been able to. I mean, sometimes it might take me like hours of going back and forth and having to like, you know, talk to perplexity to find answers or going to grok and ask questions to get answers, or, you know, I'll use other resources to try to help me overcome roadblocks. But there's very little that I haven't been able to do that I've wanted to get coded up. Which sort of brings me to the topic of today's episode or the sort of thing I want to feature in today's episode, and that's that I've actually been rebuilding the Future Tools website completely through vibe coding. So I wanted to sort of break down my process a little bit, talk about the whole process that I've been going through. We also wanted to talk a little bit about what Peter Lovels has been doing with his airplane game, because that's been pretty fascinating to watch. Let me, let me start with like a little bit of context. So I've. I built Future tools back in 2022. So it's been around for, you know, three ish three plus years now. And about a year and a half ago, I started the process of overhauling it and making it better and I started looking for developers. At first I sort of cheaped out. This, this was back before I was really making any money. I sort of cheaped out, went to Fiverr and found somebody that was really good at webflow and was trying to get them to sort of rebuild my site for me just by going back and forth and trying to get them. But that, that guy flaked on me and then he fixed a few things but then eventually just disappeared. I never heard from him again. Right. I probably over the whole, like maybe three or four weeks I was working with him, paid him $800. Like, I don't even think, like I even got to $1,000 with him just on, like, doing little tasks and trying to fix things with me. But that process was frustrating because he just sort of disappeared on me at one point and then literally never heard from him again to this day. Hopefully he's still alive. I don't know. But then. So then I started the process of looking for a more, I guess, legit developer. Not the people on Fiverr aren't legit, but I wanted somebody with more of a track record. So I started looking for recommendations. I found somebody that came highly recommended, that had built some really cool websites and software for other people, brought him on board to start working on it. He mapped out this whole roadmap of what he was going to create. He actually made some new designs, gave me a whole, like, branding sheet with color schemes and fonts, and, you know, had a whole workup for me. And I went, cool, that looks good, let's do it. Gave him a 50% deposit on the project. You know, the final 50% would be done, would be given to him right around the time it was being completed before he handed over all the assets. So I gave him 50%. He started getting going with it, and then he sort of disappeared for, like, months and months and months and was stopped responding to me. And every once in a while he would send me an update and be like, hey, I've made a little bit of progress. Here's what we've got. And it was like, barely any movement at all, like six months in. And then eventually he asked for final payment and was like, yeah, I'm really, really close. Can I get final payment? And then we'll, you know, we'll finish this thing off. And I was like, all right, finally. We're going to finally get this thing done. I gave him the final payment. And then, like, it was still barely done at all. Like, he hadn't made any progress. I mean, he made some progress, but not enough progress over the time. And then I started saying, hey, here's some of the things that I wanted in the site. Like, originally the site had a light mode and dark mode. Can you add dark mode? And he was like, oh, that's out of scope. That's going to cost you this much. And I was like, but that was in the original plan. Like, if you look at the original site, that was already there, that should have been in the scope, because that's what we had. I don't want to lose features that I had. And he was like, okay, I'll figure that out. I'll just do it for you, for you. And Won't charge you for that. And I was like, okay, cool. And then I mentioned another feature that was in the original site that he didn't code into the new site. And he's like, that's out of scope. This is how much is going to. And this went on and on and on.

Typically, how it is working with engineers, like, look at, like, you know, I did several startups in Silicon Valley, and, like, even with, like, more professional, you know, engineers, you run into the same thing. And I think this is why Vibe coding is going to get more and more popular too, is because not only do you. Did you already have these issues before, those issues are probably going to be worse in the future because the people are just going to be. They're going to be Vibe coding. You're going to be paying them to do the Vibe coding.

Might as well do it yourself.

Yeah, exactly. And I mean, that's. That's what ended up happening. I eventually just, you know, cut my losses and went, I'm not getting anywhere with this. I feel like if I keep going with this person, it's going to be years. So I cut them loose. I actually went back to X and made a post. Hey, is there anybody out there that is really good at webflow? I'm just going to keep it on webflow to keep things simple, but I need somebody to, like, redesign it and rebuild it for me. Got a bunch of people to reach out, and they all gave me, like, these insane bids, and I went, you know what? I'm just going to go try to code this thing up myself. We talked to Riley Brown. Go check that episode out if you haven't already.

And we Vibe coded some stuff with him, and I'm like, I bet I can rebuild this myself. And I basically managed to clone all of the existing functionality of the Future Tools website in, like, two days. And now I'm actually building new functionality on the site and actually improving the design and stuff, which is why I haven't pushed out the new site yet, but I managed to clone what I've already got, like, super, super quick. And, yeah, I mean, that's sort of the context of why I decided to do this. Like, you know, people. People are out there going, oh, AI is taking jobs. But the thing is, I really, really, really struggled to actually pay people to do the job for me. Like, it was really, really, really difficult for me to give people money and have them do work that I was expecting. And so I was trying. I was trying to not be the guy that goes, all right, I'm just going to do everything with AI. I was trying to redistribute the wealth and be like, hey, I'll pay people to go do this for me. But it was a nightmare. So I basically decided to go code it myself out of necessity because it was such a pain in the ass working with other developers. So, anyway, that's the whole context.

And a slight rant. Hey, if you take a look at my web presence online, it's safe to say that I'm a bit AI obsessed. I even have a podcast all about AI that you're watching right now. I've gone down multiple rabbit holes with AI and done countless hours of research on the newest AI tools. And every single week, well, I've done it again. And I just dropped my list of my favorite AI tools. I've done all the research on what's been working for me, my favorite use cases and more. So if you want to steal my favorite tools and use them for yourself now, you can. You can get it at the link in the description below. Now back to the show. Yeah, what I wanted to do is sort of break down the process of how I did it, because I didn't just, like, go into Claude and say, hey, make this for me. I've actually been jumping around and using a lot of different tools. So that's. That's kind of what I wanted to break down.

Have you been mainly using Cursor? Like, how have you been building this? Like, how did you get started?

Yeah, so I've been doing Cursor and then Windsurf. I've been sort of switching back and forth to see which one I like better, you know, cursor. I was running into problems, so I went to Windsurf and then I ran into problems with Wind Surf, so I jumped back to Cursor and I'm paying for both of them right now and jumping back and forth. But the initial starting point, here's how I did it. I actually went to Vercel's v0.dev and this site is actually really, really good at cloning existing sites and actually improving the design for you. So I think I just fed it an image. I don't even think I gave it a prompt. I fed it an image which was like a screenshot of my original Future Tools website. I know it's tiny on your screen right now, but I literally took a screenshot of what Future Tools looks like, fed it into Vercel. It looks like I didn't even give it any text prompt at all with it, it was literally just a screenshot. And then Vercel went to. Or V0 went on to say, future tools, like AI directory. I'll recreate this AI tools directory website with a clean modern design similar to the one in the image. Let's build a responsive landing page with all the key components. And then you can see it went through and, you know, did its thing. And what you're seeing on the right was the design that it spit back out at me, which is a cleaner looking design than what I have now on Future Tools. And I went, cool. That's. That's the good bones for it.

That is so magical. Yeah, we take all this for granted though, right? Like if you looked at this like if you told somebody like three years ago that like you could just do that, it would just make the website for you, it'd copy your existing one, give you a new one, looked better and you just get the code for it. They're like, what are you talking about?

Yeah, no, it was like this was, this was first try, right? Like this was me throwing in a screenshot. And then it went, all right, here's a sort of nicer looking version of what you already have. So just for comparison, let me just pull up, here's the Future Tools website. Right now this is dark mode. I'll flip to light just so it's a better comparison. So this is the current site. This was the one that it designed. It's just cleaner, it just looks more modern and clean. But similar, like bones, right? Similar structure, similar bones. And all I did was screenshot it, throw it to Vercel. And it gave me this. So it gave me the code up here and then I'm able to download a zip file. So I downloaded the zip file here and it had, you know, these handful of files that it made for me. And I took the files that it made for me and I threw them into Cursor. You know, basically you create a folder and then you open that folder inside of Cursor. And so that's what was the sort of first step was. All right, I've got like the newer cleaner design that I like. I used V0 for that. My next step, and this is something that you and I have chatted a lot about, is when you're building with like AI coding tools, having like context in a roadmap is, you know, a really sort of key way to do this. So what I did was I went to GPT 4.0. I actually did this in, in, in GPT Pro mode originally. So this was actually written by Pro Mode, but once it was done, it switched it over to 4.0 for some reason. But I literally gave it my existing URL. I said, hey, I want to do a deep dive on everything the website Futures Tools IO does. I need a detailed breakdown of the entire site and all the current existing functionality. So I literally gave it the URL and said, break down all the functionality that I've already got on this site. I want it. I want to know everything. And you know, you obviously use 01 Pro. 01 Pro likes to follow up with like a few extra questions, right? So it says I can definitely do a deep dive. It asked me a few questions, I answered the questions, and then you can see future Tools IO website structure and functionality. And it just wrote up this entire breakdown. So we've got a navigation and layout breakdown, user interaction breakdown, the search and filtering breakdown. And then I asked it to suggest some unique features. Oh, these are the unique features that it has. Personal curation by Matt Wolf Exclusive deals, news aggregator Educational resources Similar tools, Anti abuse and quality control, Automated content generation. And then this kind of stuff, it started to kind of, you know, pick its own sort of features. But you can see it just like broke down everything for me in this, like super, super detailed report here. And so then what I did was I just, I copied all of this stuff and I post, I. I pasted it into a text file. Is like, this is my roadmap, right? And then when I went into Cursor for the first time, what I did was I pulled that whole text document into a chat inside of Cursor. And I literally asked Cursor to give me a step by step roadmap. Like break this down into steps that we can complete. And I use Claude 3.7. I can barely see it down in the corner, so I know you can't see it at all, but I'm using Claude 3.7. And then we can see the roadmap that it created here. So Future Tools IO rebuild roadmap. And then here's, here's the full on roadmap. It broke it down into nine phases. I think it's funny how it says one week, one week, three to five days, three to four weeks, two weeks. I'm already through like the first like six steps and I did it in two days.

Now, are you saving all that as like a text file or like a markdown file or something?

Yeah, this is, this is a file called Roadmap md. Right. So now whenever I ask questions. So you can see the roadmap here. It's really, really detailed. You can see I still have a lot of steps that I want to complete on it. But this is all new functionality. All of the existing functionality of future tools has already been built. This is all brand new here. But now what I can do is whenever I give it a prompt, I could come down here inside of cursor, and there's a little button that says add context. I can click on that and I can just type roadmap.

And you can see it gives me the option to attach my roadmap MD file, my markdown file, with all of this entire roadmap here. So now every time I prompt, it has the context of my roadmap with every prompt. So it knows where we are in the process. And I'm actually checking off stuff as it gets complete. So it knows what stuff has been done and what stuff it hasn't gotten done. And it's using that context every single time I give it a prompt.

Yeah, I think most people haven't realized, like, how powerful that is to have like a, a markdown file or text file with like your plans in it while you're working with a code base. Because then like, you know, when you use like the agent mode or whatever, like the LLM automatically can look at that document to give it better context about how to help you implement something. Right.

Yeah. So here's, here's the site. I didn't load up every single tool in here right now. I only put 20 tools in here because the actual future tool site has about 3,500 tools. And I'm like, that's overkill for like the testing phase. Right. So I only put 20 tools in here to test. But as you can see, it's got, all of. It's got these tools here, they're tagged, they've got like a sort of liking system that you can put on them. It's pulled in all of the, you know, the tool details. All of this filtering works. So let's see, you got this tool called in magic here. So if I type like I N M A G I C, you can see it automatically filtered it down to that one. I could select any of these filters here. Like, since I only have 20 tools, it's not every, every tag has tools in it. But you can see I could filter down to productivity instantly. And then I can further filter it by price. So I think all these. Are you.

Are you still using the same back end or did you have it create some kind of new backend.

And, you know, so my old backend was all webflow, right. So I built the whole backend, the whole system on webflow before this new one is all built on React. And then it's using Supabase for the database.

Yeah.

And then I'm using Wasabi for the CDN to host all of the images and, you know, media files. So this one's, this one's built from the ground up. It's not using webflow. It's not using any of that kind of stuff. It's all, you know, React, TypeScript, Supabase and Wasabi. So that's like, it's got all the functionality, it's got a news page. I'm going to, you know, I'm going to do some improved designs. My first goal was clone what I have, so. So that's kind of what it did. Right now there's going to be a new section for gadgets which is, you know, the same as like the software tools, but it's going to have like, you know, the rabbit R1 and the meta ray bans and, you know, physical AI tools.

But everything is, is functioning and working exactly like it did on the other site right now. And I built this in two days. It's got light mode, it's got dark mode. It's like I can click into one of these tools. I still need to design out the individual tool pages a little bit better, but it's got, you know, the image, the ability to favorite the, you know, the details about the tool. You can click on it and go to the tool itself. You know, it's, it, it's got everything in it. I'm just still sort of putting finishing touches on it. And yeah, I built it.

Now you're in control of it right now. You're in control of it. It's actually yours. You can modify it however you want to. You also get the extra benefit of every single time the LLMs get better. You probably can go in and it'll get better. It'll probably improve your site automatically. Like a new version comes out. You like, hey, analyze my website. Look at my roadmap. Yeah, yeah, yeah. How could you improve it? I've noticed that every time they get better, like, you just, you just show it the same code and it always discovers new things. It can improve.

Yeah, exactly. And you know, the other thing is like, you know, I was having my struggles with a developer and one, one thing that, you know, I post on X, like, hey, I'm Rebuilding future tools. Do you guys have any ideas of, like, cool features to add? I'm sort of like, fish for additional functionality that might make the site even cooler. And somebody said, like, you know, is it really that much easier than working with a developer? Don't you run into, like, issues and bugs with the, like, sort of Vibe coding process? And my response to him was, yeah, you do run into issues, but you can usually fix the issues within like, an hour or two. Where if you're actually working with a developer that doesn't respond very quickly, it might be two weeks before you get, like, a small bug fixed. Right. But now if I run into a bug and I'm like, oh, that's not good, I can typically fix it in like, 10 minutes. Sometimes they're a little more complex and it takes me like two hours and Claude won't do it inside of Cursor, so I'll go over to Grok and ask Grok, or I'll go to, like, Perplexity and do a search on Perplexity to see if I can find this solution. So sometimes I go to some other tools to help me find the solution, but I usually get a solution within, you know, a couple hours at the very, very longest, versus going back and forth with a developer trying to explain the problem to them. Maybe they misunderstand what I'm trying to fix and they fix the wrong thing. And it's just the struggles I have with Vibe coding are way smaller than the struggles that I have with actually hiring developers.

Yeah, like, communicating with people, like, what's actually in your head is very hard. Like, like, very, very difficult. Almost impossible. I think we talked about this on another episode about in game Design. You know, Jonathan Blows talked about that, like, the discon. A designer and a business person and an engineer working on a game or any product, there's always a disconnect there. And it feels like with Vibe coding, you can actually get closer to building what was actually in your head that was so hard to explain to other people. But, you know, you almost have this connection with the AI where you're going back and forth and you understand how it works and understand what you're wanting and that's going to get better over time. And all the issues you were talking about, you've got bugs and things like that, and it's kind of annoying. I've noticed every time a new model comes out, there's less of that.

Yeah, well, another thing that that other people have have said is, like, well, you're able to do this because you know about the platforms, you know to ask it to use React, you know about Supabase for the database, you know about, you know, places like Amazon S3 or Wasabi for, like, hosting the media files. You know, you have that knowledge, so you know how to prompt it to get that stuff. And that's BS because what I did was I went to ChatGPT and said, give me a roadmap. What? Like, I don't know anything about this stuff. Tell me what to build it with. ChatGPT is what recommended using React. It actually recommended using Firebase for the, for the database portion. But when I was trying to set up firebase, I was kind of running into some issues where the API wasn't connecting very well.

And so it's more expensive too.

And it's more expensive. And so I literally went to Perplexity and went, I'm having issues with Firebase. What are some alternatives? And it went, try Supabase. And I went, okay, I'll try that. So, like, I did not know to tell it to use React. I did not know to tell it to use Supabase.

I did not get stuff. I think you're, I think you're making a great point, though. This is like what in Silicon Valley, they've started calling like, people who are high agency, right? Like, people who will actually go do things versus just like, oh, I can't do it. Like, I think people have gotten used to being like that, like having lower agency because things have been so hard and it was outside of their, their wheelhouse or their area of expertise. They're like, well, I can't do that. So, but with AI, that's all changing. So, like, people really should be like, re, you know, restructuring their, their thinking to be like, higher agency of like, oh, I can do things because, like, the AI, whatever, I don't know, it just fills that gap in now.

Yeah. And, and then there's this, this, like, sort of newer kind of concept that came out. This was released by Anthropic called MCP or Model Context Protocol. And before we hit record on this episode, you, you were saying, oh, it'll be super nice when the AI can just like, look at your website for you and go, oh, this isn't working very well. Let me fix that for you. Or I noticed there's some broken images on your site. Let me go and fix that for you. MCP fixes that for you. So what MCP is. Let me share my screen again. So a lot of people go, well, isn't mcp just like an API, right? So an API, is it Application Programming Interface, I think, is that right?

Yes, yes, I believe so.

So basically an API is like there's software tools all over the Internet, right, that you can tie into with your own software, right? So let's say you Want to use ChatGPT and build ChatGPT into your own product. Well, you can go and use OpenAI's API and make your tool talk to OpenAI's tools. Or let's say you want to integrate with Google Calendar, you can actually make your tool connect to Google Calendar through an API, Right? It's API is how programmers make their tool talk to other tools, right? Well, the problem with APIs is every API is different, right? Every API has its own documentation, they're formatted in different ways. You call upon the APIs in different ways. Everybody sort of builds their APIs in different ways. It's not super, super standardized. Right. The problem with that is when you ask a large language model to go and integrate with an API, it basically has to go learn how to use that API for every tool it's trying to integrate with, right? It needs to. You need to either feed it documentation or through trial and error, it will figure out how to connect to that API. Right? So here's like a sort of like really simplification of it. So with Cursor, so you've got like, let's say you've got like a GitHub API, you're trying to connect cursor with GitHub, and you have a Slack API and you have a local file system API. Right? They're all different and Cursors trying to communicate with each one. And it needs to do it differently every single time. Well, what MCP did is it added a different layer, it added another layer to sort of standardize it. So now all GitHub has to do is create an MCP layer which communicates with this API. All Slack has to do is create an MCP that communicates with its API. And now Cursor only has to understand how mcps work. It doesn't have to understand how every single API works. These companies can basically tie their API to the mcp. And now all of these tools, or now all of these LLMs can communicate easier because there's a sort of standardization of the, of the API connection.

Yeah.

So like how that looks in effect. Let's see. I'm going to switch over to Cursor here. All right, so in Cursor, if I go up here to my Settings. And you can see that there's an MCP setting here. And you can see that I've got an MCP for Supabase, I've got an MCP for Browser Tools, and I've got an MCP for GitHub. This basically means that when I'm communicating over here in my chat window, in my Agent window, inside of Cursor, I can tell it to do something like break down the schema of my current database.

Right? And I can give it that prompt. I'm going to switch it to Chat mode so it doesn't go and try to write code on me. And what it will do is it will go and use this MCP to connect directly to my Supabase account and look at the databases for me. So in the past, it would basically create SQL files and tell you to copy this SQL file over into Supabase. And then if something wasn't working, it would look at the SQL file and try to figure it out. But it wasn't actually looking at Database for you. Well, now this is actually looking straight into the database for me. And it can break down the whole, like, schema for me. It can break down all of the various tables and all of the, you know, various columns that are in each table, and it can break it all down by literally looking directly in my database for me. Same with the GitHub MCP. I gave it access to be able to look at my GitHub account. So instead of having to write Git commands, it could literally just push and commit stuff straight directly into GitHub without having to do, like, terminal commands. And it can go and read my various, like, files inside of GitHub and things like that. And then there's this Browser Tools mcp, which is actually a local network that you connect. And let me actually show this here. So this is the GitHub for browser tools MCP. And what this allows it to do is it allows. It's actually a Chrome extension. If I look at my Chrome extensions here, I have a Chrome extension called Browser Tools MCP that's turned on. So because that Chrome extension is turned on and I have the MCP installed over inside of Cursor, what this will do is it gives it the ability to look at all of your console logs, look at all of your network logs. So, you know, like, if you go to Inspect and you have this, like, little sidebar here, you've got, like, console and sources and, you know, memory and all of this kind of stuff that you can view. Well, this McP gives your LLM access to all of this, so it can go read console errors, it can go read network errors, all of that kind of stuff. But it's also capable of capturing screenshots. So I can sit there and so I can give it a command. And I don't want to do this right now because I actually have to fire up the local server and stuff. And it's visually, it's not going to really add anything, but I can go in there and I can say, make this change for me. And after you make the change, take a screenshot of my web page to make sure the change worked. And so it'll go and do all the coding, make all the changes, all the changes, and then connect through the mcp, take a screenshot of my website, and then if it looks right, it'll say, good, it looks like the changes were implemented correctly. If it doesn't look right, it might say, oops, it looks like we've got some broken images and some console errors. Let me fix that for you. And then it will just keep on processing and then it'll process again and then write more code and it'll take another screenshot and look at the console logs and go, oh, it looks like we fixed it. Now it's actually working. So with these model context protocols, it can actually connect directly into my browser now look at the console logs, take screenshots for me, look at my, my Supabase database for me, look at my GitHub account for me, and actually pull information from all of these places to sort of better inform the next decision that it makes. And it is, it. It just like, makes everything so much easier because it spots its own errors. Now when it spots its own errors, it then goes, oops, not working. Let's go ahead and fix it. And more and more Companies are creating MCPs to be able to connect with large language models. So Firebase has an MCP as well. Now, obviously Slack has one. We were just seeing that as one of the examples. The GitHub MCP is really cool because you can actually, I mean, you could have it go and look at a whole, like, GitHub repository and have it break down what's going on in that repository for you now?

It's just, I did notice it like, like MCP's got really big on social media in the last week, but people were saying, like, the concept's been around for like a year now or something, right? Was it just like the models were not good enough to actually be doing this or There was not enough like MCPs out there that were actually useful.

No. Well, the big change, I think. So MCPs have been around here. Let me actually, I'll tell you when they actually came out, because it was Anthropic who created this. So they came out in November. So November, December, January, February, March. So they've been around for four and a half, five months. It came out at the end of November.

So here's Anthropics website introducing the model Context protocol. It came out November 25, 2024. So, you know, four and a half months or so is when this technology came out. The big difference, though, was the IDEs didn't have model concept protocol like direct integrations yet, and you can actually write some JavaScript code to make it connect, but you kind of needed to know how to do that. Right now they've just implemented them into the IDEs like cursor and Windsurf and tools like that. They have like an MCP setting section now to make it a lot easier to connect. So there's that element, and the other element is that a lot of these tools had to create MCPs that connected to their APIs. Right. Like Supabase's MCP just came out like two weeks ago. So we've only been able to connect directly to Supabase using the MCP a couple weeks ago.

Yeah. So, I mean, it seems like this is a big moment because, like, now almost any kind of software you can use, if they develop an mcp, you'll be able to use it and interact with it through the LLMs now, which is awesome.

That's why everybody's going, holy crap. This is like, this is pretty big. It's because, Yeah, I mean, like you said, you can connect it straight into Unity and it could probably look code into Unity for you.

Well, I mean, so actually the LLMs already can push code into Unity, but the problem they have is, like, setting up the scenes and different things like that in a game is something you do in the editor, but apparently with the MCP now, you can actually create the scenes and different objects in the editor through the mcp. So I'm super excited to try that out.

Yeah, no, I mean, like the, like, like we've been talking about this, this episode and last week's episode. Like, this stuff's just getting easier and easier. So it's, It's. It's crazy how easy it's gotten to be able to do this stuff. And the more, like, the, the more the models get smarter, like like you mentioned, the less errors it creates in the code and the more tools we can directly connect to the LLM, the better it's able to tell if it's doing a good job or not, essentially. Right?

Yeah. And then combine all that with the models getting smarter and smarter. And then I guess you got AGI, or at least what I would probably call AGI.

Yeah. Yeah. I mean, it's, it's really wild, but. Yeah. So that's, that's how I've been building future tools. That's, that's been the process. I think, like, the really sort of valuable part that people should, should know is like, you don't really have to start from scratch. Right. If you have a concept for a website and there's a design out there that you like, go start as like, with a template. Go and find a site that you like the design of that sort of maybe has the functionality you want, pull that into v0, let it clone it, and use that as your jumping off point. Like, don't make, you know, blatant copies of other people's sites, but use it as like a nice beginning template to start from and then use something like, you know, Google's deep research. Or you can. I mean, I even think GPT 4.5 will work well to write up a roadmap for you. I think it would actually do a pretty good job. But you can use one of these tools to explain what you want to create. Let it write up a roadmap for you. Let it tell you what tools to use like React and Supabase and all that kind of stuff, and you don't have to figure this stuff out. The AI already knows the best tools to accomplish what you're trying to accomplish. So if you have your starting design of what you want and you have a roadmap that AI helped you generate with the tech stack already figured out for you. Like, the rest is just putting it into cursor and telling Cursor to follow the roadmap, right? Yeah.

And especially now that there's mcps. I mean, it's like you're gonna be able to do beyond just coding, Right. It's me vibe everything, right? Yeah, vibe anything you want. Because it's, it's not going to be limited to just coding anymore. You're going to like, like I said, even like game design or almost any kind of business. You're going to have the same thing where I will be able to help you do anything.

Yeah.

And it's going to be more and more Important to be high agency, you know, and actually, you know, you don't have excuse anymore to not say, oh, I don't know how to do that. The AI will help you figure out how to do it. Matt didn't know how to do database stuff. He's now a database expert through Vibe coding.

Yeah, it's wild. There is one other topic I want to talk about is sort of a shift from future tools. And we don't have to stick around on this for too long because I think this episode's already going longer than we wanted. But so the other topic that I think would be cool to talk about is Peter Levels. Right. He just built this, like, airplane game, and it's honestly not that impressive looking. What's impressive about it is that he's just like, one dude that Vibe coded this whole thing by himself, and he just passed a million dollars in monthly annual revenue from this one person coding up this app by themselves.

Yeah. Should we show it or should we just talk about it first?

Let me see. Yeah, you want to pull it up on your screen. Yeah.

It's a very basic. And so if you're. If you're. If you're listening to this instead of watching on YouTube, it's a very basic game he's built. It's like, in 3js, which is a framework that helps you build 3D stuff in the browser. And it's a. It's a simple, you know, flight simulator where you fly around. But he. It's the. The interesting thing about this, I think a lot of people are going to look at, like, oh, this is just a stupid thing. It's a fad. The. The game itself, I would not doubt if it dies off. Like he's saying. He's going. He's got like a million mmr, you know? No.

A million ARR.

Yeah, yeah, yeah, yeah, yeah. Million ARR. Yeah. ARR. But I mean, that's. Yeah, obviously he's like, assuming that, like, everyone will keep playing it. I wouldn't doubt that this dies off, like, in a month or two. What I think is really interesting is that this is like the ultimate example of Vibe coding. And he built it like, he shared a concept and he built, like, a really basic demo of it, and he just shared it online and he started talking with his community and tweaking it based on stuff they were saying, which when you were talking about building your future tools website, it kind of reminded me of that. I was like, matt, you kind of were Vibe building your.

New future tools with your community. Right, yeah. Where they were giving you feedback and you're building your roadmap based on what they said and then you just went off and built it. He did the same kind of thing with this game where he's just chatting with people online, his fans, about what should he add next. And now that it's so easy that you can actually just do that yourself versus going off and hiring an entire team and it takes all this time. No, you just chat with AI and you just change stuff and you just push it live and then you just share it again and get more and more attention like that. You know, it's almost like how people used to do stuff on Twitch, right?

Yeah.

Live streaming. I'm kind of convinced this is going to be like a huge trend over the next year or two where you're going to have all these influencers realizing like, oh, wait a minute, I could use my audience and I could just like live build something. And it doesn't be a game, it could be software too. I'm going to live build something and my, my audience is going to interact with me as I build it and I'm going to make it for them.

Yeah, yeah. It's crazy. So I just pulled up this tweet from, from Peter here. It says, yeah, what you see 200K MRR. So he's at $200,000 a month in MRR. So that's what, 1, 1.4 million a year? If you're extrapolating it out, what you don't see, it's only stripe fees. That's like the only cost he has. Apparently no Apple fees because it's all web based. He owns 100% of it. He spent $0 on ads. Be your own influencer. No co founders and happy customers. Right. Like that's, that's wild.

And game designers are so pissed off. They're all like tweeting about it.

Yeah, yeah. And I mean, it's not a great looking game. I mean, it looks like a game that came out in the 90s, the early 90s. But like the way he's making money is, see all the ads, they're like everywhere. You literally put an ad on planets and he put an ad on a blimp and he's got ads on like the buildings on the screen and he charges five grand for one of these ads that like a fake plane, like flies past really quick. I'm like.

And he tweets out their names too. That's another element of it. Even tweeting out all of them.

That's probably the bigger benefit because, I mean, the fact that they're like, ad is in this game. It's not even direct response. Like, people aren't like clicking on these ads to get to the website. It's just pure branding. So, like, I even have a hard. Like, I actually looked at it. Cause I'm like, oh, maybe I'll throw an ad up there for future tools just for fun. Because who's using it right now? Vibe coders and people interested in AI. Future tools would be great. And then I saw five grand a month to put my name on a blimp that planes are just flying by. And I was like, I don't think I can justify that.

Right. Yeah, yeah. And the people probably stop playing it like a month or two, you know, I think. But. But still. And just conceptually, the fact that you can build something and then interact with your community, because now it's so easy to actually build something, you know, in near real time like that. I think that's like, that's a game changer that most people are not really fully thinking through.

Yeah.

Yeah, yeah. This is wild. So. So you actually get into the game and you have missions. Missions. Shoot down all the balloons, shoot down other planes, fly to space.

It did not. It did not have any of this. Like, when he started, he literally every feature has been people in the community just saying stuff like, why doesn't it have that? He's like, cool.

Yeah.

And then he's. And then he. And then he'll like tweet out a, you know, a link saying, like, it's up now. Like a little video clip.

So wild.

So that's been. The whole process is literally him just sharing little short videos on X and then people saying, cool. Why doesn't it have this?

Yeah, we need more UFOs. We need a castle.

Now it has it. You want to. Oh, someone should, you know, next he's going to be charging for planes. Cool. Here's a. Here's a stripe where you can buy planes. Buy the planes. It should have advertisements. Cool. Here's ads. And the entire process has been that of, like, him Vibe coding it and then just pushing it out to the community and doing it interactively.

Yeah, it's wild. I mean, we're entering this crazy world where literally anybody can code anything they want. They can build anything they want. If you have an idea, if you've got a vision. Yeah, go and use AI. Right. Again, you can use something like V0 to help you with the original design. Use something like ChatGPT or Claude or Grok3 to go and help you map out the product.

But seriously, beyond just games, because the game part itself, I don't think that's the main thing. I really think it's idea of if you're in a business right now, you have a new idea for a product and you have a big community, why don't you just, instead of making it super polished and like waiting years to build out the thing. Why, why, why doesn't the one business leader who has the idea literally just push it out to social media and share a prototype and see if people have any interest and they say hey, why doesn't it do this cool now it does that, you know, and like have an iterative process with, with your community to build software. I think that's a huge opportunity that people are not realizing.

I think that's a good place to wrap this one up. I think, you know this whole vibe coding concept, I've been nerding out about it. I've actually been prod YouTube videos on my own YouTube channel lately because I've been so obsessed with just coding stuff. Future tools is just one thing that I've been building. Like I've been building other little apps for my own like personal helpful use. We've talked about this on a past episode. I love finding little bottlenecks in my business and going whole bunch of little tools like that now like a, a tool that easily converts one image file to like or any image file to a single JPEG file. Doesn't matter what type of file I drop in, converts it instantly. I created a tool with Whisper where I can take any of my short form videos, drop them in and it instantly transcribes it and then I can take that transcription and use it as my description over on Instagram. So all of my Instagram sort of descriptions on everything is all written by a high. Now I showed you on a past video when we had Matthew Berman on, I showed you this tool that I'm calling the video producer where I can throw a whole bunch of videos in and it helps me with ideas on new videos based on the videos that are in my database. And so I'm just like finding bottlenecks or things that I want to work through in my business and then using AI to build tools that solve those problems for me. And I'm doing it as a one man show just by myself. Like I'm just, I have this idea, let's build it. I'm not making as many videos but Like I'm obsessing over this right now.

Yeah, I mean, I mean it does create a new problem of like, what do you focus on? Yeah, it's like there's like there going to be more and more infinite possibilities, right, with AI that it'll be able to go off and do things for you. And it's like, what do I focus on? You know, I became obsessed with my game. I'm like, well the game may not even make money. I'm like, you know, with AI and everything else, is it eventually even going to matter? Like as long as I enjoy it, you know, I enjoy the process of making it.

I see a lot of this is like upfront work now to save me time in the future. Right? So like maybe, yeah, I. I'm producing one less video this week because I made a tool that improves my workflow. But next week I have one less task on my plate because I improved this workflow. So it's like I'm building all these little things that ideally will free me up in the future. The problem is I keep on getting more ideas for things that will free me up in the future. So I never actually end up freeing any time because I end up building the new ideas that I have that might free.

Have you ever considered just like open sourcing all that stuff that you're doing that you're building?

Yeah, like I've open sourced a lot of them. So yeah, a lot of the little tools that I've made, like the Any image to JPEG converter, I put that up on GitHub, my AI.

Have you told people on X and YouTube or no?

Yeah, I've mentioned it. I haven't made a big deal about it, but I have mentioned it in the past. I'll talk about a tool that I made and people will be like, oh, that's really cool, can I get access to it? And I'll be like, Yep, it's on GitHub now, go grab it. So like I pointed it out, but I haven't really went out and megaphone announced it.

Right.

But yeah, as I'm building it, I'm putting everything on GitHub while it's still in the process of being built. I keep it as private on GitHub, but once I feel like it's ready for other people to go use, then I make it public on GitHub and I set the license to an open source license so anybody could go and use it. I don't think I'll open source future tools. There's already too many clones of my website. I don't know if I want to enable more of that, but pretty much everything else I build, I open source.

What's the key takeaway for people? Be high agency and you don't have excuses now and go build stuff.

All of that. Yeah. No excuses. No excuses. You know, you get an idea, you have bottlenecks in your business, go build workflows to fix them. You want to go build a new website, you don't need that developer anymore. You need a new business model. Maybe you can monetize your fans by collaboratively building with them and working with their ideas and helping convert them into your marketing force because you're building with the community. Right? There's more takeaways in this episode than a lot of episodes we've done in the past.

I think.

I think a lot of people have a lot they can take from this one. But can I make a little app, like to build a workflow to get past that bottleneck? And I've made a whole.