Build a Website Using Vibe Coding in 45 Min (GPT-4 & V0)

Hey, welcome to the Next Wave podcast. I'm Matt Wolf and I'm here with Nathan Lands. And today we're diving into vibe coding again. In fact, we're bringing back Riley Brown, one of the guys who sort of led the charge on vibe coding. We're going to talk about what's changed in the world of AI coding since the last time we chatted. All sorts of cool, amazing new tools that make it even easier to code than it was before. So we're going to dive into all of that, make sure to stick around, because we're going to build an app live on this episode, a super fun app that I think you're going to be pretty blown away by what it can do and how quickly we build it. So, super, super fun episode. So let's dive right in with Riley Brown.

Thanks again for joining us. Riley, how you doing today?

I'm doing great, I guess.

Let's start here. Like, out of all of the developments that have happened in the last three months, like, which ones have impacted your coding the most?

Ooh, I would just say how well Cursor and Windsurf are my. The two tools that I use the most, how well they understand the code base. And whenever a model changes. Like, recently it switched from Claude 3.5 sonnet to Claude 3.7 sonnet when Cloud came out with their new model. And there's this weird period when these models change because the cursor was basically built for Cloud 3.5. And so it got really, really good at 3.5. And once it switched to Cloud 3.7, it acted a little bit differently, and it stopped following all of my requests perfectly. And it happened for a lot of people. And then I think we saw a lot of people who really love Cursor start switching to Windsurf. And so I think that both of those tools, like, I think those are the two best tools to use for AI coding. If you want to, like, dig deeper, you can use a lot of the simpler tools, lovable, replit, bolt, etc. But they just understand the code base really well. And you're able to do longer tasks.

Between Cursor and Windsurf. You know, I have both. I'm literally paying monthly for both of them. And I switch back and forth. Like, sometimes a cursor will get hung up and get stuck in a loop, and I can't get it to solve a bug. And then I'll open up Wind Surf, open up the same folder in Wind Surf and Wind Surf, one shot fixes the problem. And then I run into the same thing with Wind Surf and I'll jump back to Cursor and Cursor will fix it. To me, they're pretty much the same. But sometimes one figures it out while the other doesn't. But like, what do you see as like the biggest differences between the two?

I think Windsurf has a nicer user interface, but I notice it does get, it does get stuck on some weird things that Cursor doesn't. But Cursor also gets stuck stuck on things that Windsurf doesn't. So I think I had a tweet two days ago about how like the cynical AI crowd will tell me, they're like, oh yeah, good luck when you get a bug. And then I wrote a list of like these like seven things I do whenever I get a bug. And it's like almost guaranteed to beat it. Because you can switch AI models. If all of the AI models don't work, you can switch to a whole nother software that's also really good. Windsurf. And then you can use the Internet. And now with mcps, I think probably the second biggest thing is the fact that I have like four or five MCPs built directly into Cursor that surfs the Internet, which is actually just built directly into Cursor. But you can actually set it up to do Perplexity Deep research directly in Cursor, because Perplexity released an API that you can use via mcp. So you can say, I want to build this app that requires to use, let's say you want to use the Deepgram API for speech to text. You can have Cursor or Windsurf search the Internet, come back with that information, and then look at your code base based on that information and then it can generate the code with the information. So it's like a research agent. And that has been really impactful and really cool.

So basically in mcp, for anybody listening, it stands for Model Context Protocol. It was created by Anthropic. We actually did an episode where we talked a little bit about it. But it's sort of a layer that lives between like the API of various tools that you might want to connect with and your large language model so that it sort of makes it more of a uniform way that the large language models can communicate with tool APIs. So now basically every single API is like a little bit different. Every single like company's API has these full on docs of how to use their API and nobody really quite does it the same way. So the large language models communicate with the MCP and then the MCP goes and communicates with the APIs on the behalf of the large language model. That's probably the simplest explanation I can give. But you know, in even simpler terms, the MCPs make it easier for the large language models to use tools.

Good explanation of it.

Yeah. So when it comes to the mcps, the ones. So I started using Supabase or Supabase. I don't know why I keep calling it Supabase. I think it's Supabase. I know you're not a fan of Supabase. You, I think you're more on the Firebase train. But Supabase has an MCP where I can build my database and then Cursor can actually go and look at the database and double check that things are working properly. On the database side of things, there's also a browser tools, one that I've been using that can actually look at your console on your browser and take screenshots of your browser, so it can sort of double check its work. But those are really the only two MCPs that I've even used at all. And I'm actually just now learning that Perplexity MCP works well as well. So I'm gonna have to add that into my mix.

So this is another thing about MCPS right now. I think it's incredibly early and I think if you're not technical, unless you're creating content on it, like I am, or you are, like, it might be perfectly okay to wait a little bit. Cause they're not perfect. There's a ton of potential and then there's like so many. I think there's like four different YC companies in this batch that are literally trying to make it easier. As simple as just one click, add an mcp. Because right now you have to like copy and paste code, you have to get different API keys. And so like, if you don't understand mcps, if you want to, like, get ahead, I think there's a lot of business opportunities in mcps. But like, don't feel like you're falling behind anything big because it's still in its infancy stages. And so one that was really useful for me that I did try was the there's a Fire Crawl MCP that can literally like crawl a website and like figure out how it's designed and then build the ui, like down to the pixel perfectly.

That's sweet.

Which was one for design.

Well, and I just saw. I don't know if Sam Altman Posted it today or Yesterday, but apparently OpenAI is releasing an MCP as well. And I'm really, really hoping they release one that can communicate with their new image gen model, because that would be really cool to actually communicate with an MCP inside of Cursor and have it generate, you know, image assets and things like that directly from within Cursor without even having to leave. So that, to me, is also really exciting. It'd also be really exciting if, you know, it works with things like One Pro. Although I think that'll get really, really expensive really quick if you're tapping into the Zero1 Pro API.

It's like $600 per million output token, right? It's insane.

Something like that. Yeah.

Yeah.

Which I think is why Cursor and Windsurf haven't rolled it out yet. And if they do roll it out, I imagine it's probably going to be a bring your own API sort of situation to be able to use it.

Like, there was a short window where 01 Pro was absolutely amazing. And on some benchmarks, I still think it is number one or number two now. But now, like, the difference is so small. I don't see why anyone would, you know, pay that much to use API especially.

I've been hearing a lot of rumors about Deepseek R2 and how it is going to be as good as 01 Pro. And, you know, it's probably going to be a hundred times cheaper and it'll.

Probably be open source and you'll probably be able to run it straight through something like Groq Groq Groq to really, really crank up the speed. And we also now have Gemini 2.5, which just came out from Google, which a lot of people are saying is sort of on a similar level to 01 Pro.

Also, I have been obsessed with how Sam Altman and OpenAI, they're so good at dominating the narrative. Like, Google released arguably the best model for coding. And we're at a time where Vibe coding is at its peak in terms of popularity. Vibe coding as a niche has dominated Twitter. And then on the day they released the best tool for vive coders, no one's talking about it because Sam always releases it right before. And I go back, like, Sam has done this since the beginning, like, as a troll to Google. Every time they have something lined up, it's just like, oh, OpenAI releases something that they've been holding in their pocket. I think we should talk about the image model. I think it's really big for Vibe coding. Because I think a lot of people who vibe code love it for design. And this image model can generate perfect design. If you ask it to, like, create an iPhone app layout, it'll do it, like, perfectly to the pixel.

Somebody made a whole, like, Wikipedia page inside of the new ChatGPT as well, and it looked identical to a real Wikipedia page.

It's insane. And so a lot of the reason why people use tools like figma are to ideate and get their ideas out because, like, a lot of people are very visual. But imagine GPT4O their new image model, but imagine if it was five times faster and it's safe to assume that it's going to reach that point within by the end of this year. It's probably safe to think that that's how fast this is going. And so if it's, you know, it's three to five seconds per image generation, imagine just like, instead of having to, like, draw everything out manually, you're just like, no, edit it. Like, move this bar slightly lower. Create an image. And when you generate an image on GPT4O, pay attention to how it does it and pay attention to how midjourney does it. If you look at mid journey, they use diffusion. And so it's like, it's all being. It goes from like a jumbled mess into like, really refined. Where OpenAI. I forget even what it's called, Auto regressive or something. I don't even know, but it like goes top to bottom. So it literally generates the top row of the image and then it's like pixel perfect. It generates all the way down.

Yeah.

And so I think we're going to see a design company come out of this that is just like a multibillion dollar company come out of this that just uses like, voice to like, do mockups and stuff. I think it's. It's a big deal.

Yeah, no, it literally feels like you're sort of standing over the shoulder of a designer and saying, all right, hey, fix this for me. Hey, change that. Hey, add this text. In fact, I'm going to share my screen because I've been. I've been really sort of messing with this thing a lot. I was actually using it where I made thumbnails for my YouTube video. The thumbnail at the top here that you could see was actually one that my designer made, but I took the one that the designer made and then I started like, step by step tweaking it to get something more out of it. So I wanted it to like, change my facial expression. And put WTF in it instead of worth it. And you can see it made pretty much the same thumbnail, but just change the text and change the face a little bit, right? And then I said, you know, make me not smiling. And then it made the same image but with a frowny face on it, right? And then I basically prompted it again to get another variation. This one's sort of looking to the camera. This one's sort of looking more at the device. And I was just sitting there, like, saying, hey, change this thing, change this thing. And it was sort of going through the process and changing them all. I had this image here of me holding up all these devices. And then I said, add the words worth it. And it put it twice, and one of them was sort of messed up. But then I eventually got it to do it just one time, right? And then I asked it to close the mouth. And then it started to look less and less like me, unfortunately. But, you know, I was sitting here just prompting different, you know, different tweaks that I wanted to make to the image. And it was just adding the tweaks and making better and better thumbnails for me. I threw in some images of, like, the Padres because baseball season starting. And I told to make this image in GTA 5 style. I told it to make it in Rick and Morty style. And then I was sending these, these to some of my friends who are not Padre fans to sort of troll them a little bit.

I had it making infographics for me earlier. Here's another infographic that it made. Here's another infographic. Here's a Venn diagram that I had it make. I was having it make, like, Studio Ghibli style and Simpson style of, like, family photos.

That's a huge business, by the way. I think somebody should be out there, like, selling that to mom and pops and like, individuals right now.

We're about to see as soon 10 hours after they release their API, we're going to see 10,000 of the same rappers be released. Yeah, absolutely. I might be one of them.

For fun, I put an image of myself. South park version, Minecraft version. Let's see. Pixel art video game version, 3D voxel version. Like, I've. I've been nerding out over this. Like, I've been pretty much my whole day so far. Today has been playing with this.

So is Mid Journey screwed or like. Cause they were supposed to come out with B7, right? And B7. The big thing of B7 was it was going to be more Consistent. It was finally going to understand your images better. And it's like, well, that's funny, you know.

So I guess today was the office hours of Mid Journey.

Yeah.

And I saw people live tweeting during the office hours. And I guess David from Midjourney was on there going, oh, the ChatGPT model's not that good. I don't know why everybody's so excited about it. It's not nearly as good as what we've been building and like, just sort of this very, very negative tone towards it. And then like, all the replies on that tweet were like, this sounds like copium to me.

You know, like.

I really like David. I just love the fact that, like, Mid Journey was completely bootstrapped. Like, I don't think they've raised any money. And he, he seems. There's a tweet that he tweeted a while back where he just doesn't seem that interested in the competition, which may come back to bite him now. Because I actually do think that throughout the last six months, midjourney, their images do have a little bit more energy and soul to them compared to a lot of the other image generators. And there are other good image generators and Flux is good. But I've just respected their brand, I guess, and I think their site is amazing. And I think that, you know, if ideally you just. They would add their. The OpenAI's API into their site, but I don't know if they're going to do that. I don't know. I do have faith that they're going to build some cool 3D stuff, but I also think OpenAI is going to do that. So I don't know. Who knows?

Yeah, totally agree. I think, you know, I agree. I think Midjourney has like a specific style. You can see images and go, okay, this is Mid Journey style. I find myself using Midjourney a lot less than I used to these days. I think, you know, Ideogram has caught up. I really like the Leonardo Phoenix model, but I'm also, you know, I also have equity in Leonardo, so there is that.

They're good.

Yeah.

But, you know, there's a lot of other models I just find myself use. I still pay the subscription, but I just find myself using it less and less. And now I really like Chat GPT in this new model for the ability to edit, like to take existing photos and then have it add the text or change the style or, you know, I did make some images in the studio Ghibli style and was posting them because it's just fun. But it's, you know, we're hitting saturation point on Studio Ghibli right now, but it's still really, really, really fun.

I've been skeptical for a long time because I've always thought that, you know, like mid journey, like I said, it doesn't really understand what's going on in your image or doesn't seem to. Especially if you try to edit, you can see that really quickly, like, oh, it does not understand what's going on. It doesn't exactly know how it made the beautiful thing in the first place. So it has a hard time of editing that beautiful thing. And it seems like with OpenAI and the resources they have, their LLMs are going to be able to understand images, which is now we're seeing that they're actually understanding what's in the thing. And it feels like fundamentally different technology.

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 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.

I think, I think the biggest takeaway from this is that it's a commodity. I think AI generated images. Even Imad Mostaq, the original founder of Stable Diffusion, I don't know if he was a founder, but I really like him. He was like, AI images have been solved by this model because of its command over the pixels. It knows exactly what to do. And I think because of that, and you think of like AI generated videos or just a bunch of images string together, you know, 30 images strung together in a second, we're going to see that level of command. And, you know, I don't see a reason why we won't be able to create Studio Ghibli animations at an incredibly high level by OpenAI if it has that level of a command and it can just generate all 30 pixels for that frame. Especially the quality, like if you look at some of the 3D, like, if you create like 3D vector or like 3D characters, and you Zoom in. It just has like, it understands physics insanely well. It's like scary, honestly.

Yeah, Yeah. I mean you take some of the studio Ghibli images it generates or some of the like voxelized images that I was generating and you. Right now, if you throw them into something like Runway's dream machine, it'll actually animate them and make it look fairly close to like a studio Ghibli animation. So I imagine it's only a matter of time before you can just prompt that straight in chat GPT without having to use multiple tools.

It's so smart. I literally handed it my, my game docs like defining like the, you know, explain like the aesthetic of my game and the story and everything like that. And it generated just like shocking art that like got so many details. They're just like, I tried to get Mid Journey to do this and like it made it beautiful, but it just missed all the details.

I want to, I want to jump back real quick to like the, the Gemini 2.5 because you know, the image generation sort of overshadow it because you know that's sort of OpenAI's MO is like to try to overshadow Google whenever they can. But the Gemini 2.5 Pro, that model, it actually has a million con, a million token context window and it is insanely fast. I actually took a transcript from a 30 minute video and was having it help me with ideas for titles for the video and it took this transcript. I don't know exactly how many words it was, but it was, you know, tens of thousands of words. I plugged it in there and like within three seconds it had like 10 title ideas. It didn't even seem like it took the time to read it, it's just that fast. And then, you know, if you're talking about things like Vibe coding with a million token context window, if, if you have a, a fairly decent size app, you can actually copy and paste your entire code base in there and let Gemini 2.5 read that entire code base and sort of find bugs and find redundancies and help it and ask it to like refactor the code for you and things like that. And it'll actually do that. When you, when I try to do it with OpenAI's 01 Pro, I use something like repo prompt, but on a PC, repo prompt wasn't available. So I was using one called Repo Mix. And with Repo Mix I was copying my entire code base. But O1 Pro was saying there's too much Text here for us to read it. Like, it was already over the limit. But now with 2 point Gemini 2.5, you can actually throw the entire code base in there and actually get it to read the whole thing for you. Which to me is wild. But like you, Riley, I actually haven't spent a lot of time using 2.5 Pro with coding yet.

Where are people using it? Like, is it just in the Gemini studio or whatever? I forget what.

Yeah, it's in their AI studio. So I think it's AI studio.google.com. let me. Yeah, a studio.google.com is where you can use it.

That's why I asked.

That's why I asked Logan. Last time we had him on, I was like, why don't you guys have it like on like a. An actual property? Like you're like hiding it behind the AI Studio thing. And like, I get it, that's they're like targeting a different audience. But it's like, it feels like it could be great if they like made a great consumer product with that in it.

And is it, Is it accessible via API here?

Yeah.

Okay.

Yeah, it looks like the API is available. You can see here we've got the token count here and it's got 0 of 1,048,576 tokens on the screen. So I mean, it's that that context window is really, really what makes the difference here.

And they said 2.5 pro is going to get 2 million soon, which is nuts.

Yeah. Yeah. And so I, I mean, I would imagine by like the end of the week, probably by the time this episode is live, you'll probably see Gemini 2.5 Pro in windsurf and cursor, if I had to guess, because those guys, whenever these new models come out and the APIs available, they get them in there quickly.

I do know notice that anthropic, for whatever reason, their models seem to be better in cursor and Windsurf. Whether or not the code is technically good enough, it just kind of understands their whole tool system better. I don't know. I don't know the science of why that is, but I remember testing Gemini in cursor. And even though the. The previous version of Gemini is good, when I copy and paste code over when I use it in cursor, it's just not. Doesn't quite understand the code base as well.

I've noticed the same thing. When I was trying to use like 01, not 01 Pro, but just the standard 01. I never got results that were nearly as good as what Claude 3.5 or 3.7 would do for me.

Another update that was just released, actually, that we should Talk about is V0 released some new features.

Oh, cool.

I wasn't even aware. Should we try to build something real quick? Maybe we can. I'm kind of curious to see how some of your workflows have changed since the last time we did a video together. So I think it'd be kind of cool to try to build something simple.

So let's go to ChatGPT, right? And let's come up. Let's make. Let's start an agency. U.S. 3. We're starting an agency.

Okay.

And we are. What do we do? What do we sell? You know, we sell our AI services. We're AI experts. So what we could just. We could ask. Oh, yeah, I need to switch to 4o and reminder. That's four. Oh, and we can say, come up with a logo for my company. Where we sell our. Where we are consultants.

We're really going to do this, right? This is a good idea.

Yeah. This is a real business being formed right now on this episode.

Dan, make the image, please. Okay, so the reason I'm doing this is I want to talk about V0's latest feature where you can just paste in images directly into the chat and say, use this image in the app. And that wasn't a thing before. And to my knowledge, like, Lovable doesn't do that either. Bolt might do something like it. I think in Bolt you can actually just go into the code files and put it in the public folder. Okay, this is.

Please make it a bunny and don't include the text.

Make it in the style of.

I guess we could use the style of the rabbit, which is teenage engineering.

I mean, I do like the slogan.

Yeah, yeah, I think the slogan's good, but it's too. Too corporate for me.

Yeah. It also looks very generic. Whenever you ask an AI to generate an image of what AI looks like, it always does this sort of brain with like neural network kind of imagery.

Yeah, it's. It's. It's in the training data and you will never get it out. Okay, this is good. I like this already. Well, they have to describe. I'm not liking these.

What is.

Yeah, completely disregard the styling before. I want a cute rabbit mascot, Pixar style. Simple, please, like a logo.

Okay, so what you're saying is we can actually make a logo and then give V0 the logo and it'll sort of design the site around that logo?

Yeah, so let's say we want to. Okay, so build a landing page for my agency that helps people with AI and come up with good copy writing. And so what we could do here is we could say, and so if you download the image right, all you need to do is just drag in the image here and you can do as many images as you want. Like, you could, let's say you have a portfolio of all a bunch of projects, or you're a designer. You could upload all of your designs and you could say, build a landing page for this. My name is Riley Brown. Put it at the top. And my number is this. Call me if you need to. And come up with copywriting and convert people to fill out a form for email.

So is this your normal flow? If you're building a new app, do you sort of design it out in V0 first and then pull it into cursor?

I do that sometimes, especially with landing pages. Please use this attached image as a logo design. I don't know. We'll see. We'll see what it does. And then we can always ask it to edit it. Yeah, I think. I think when I'm doing landing pages, yes, I just use v0. And then v0 has this feature. If you press this button right here, which is download zip, that's how I do it. I just download the zip and then open it in cursor, and you just open that project after you unzip it, and then you can just immediately start editing it. And it's just a really fast way to, like, concept things. And then now what this feature does basically is when you press download the zip, this image that you just paste into the chat will actually be in the file. So you can open it up, straighten a cursor, and then cursor does a lot better, in my opinion. After prompt five, you know, once you start getting deeper into the project.

Gotcha. Yeah, that's been a similar workflow. I like sort of getting the Design out of V0 for whatever reason. V0 seems to be the best right now as sort of coming up with a design that looks nice and clean. And then you grab the zip and then what I'll typically do is throw it into cursor and then I'll say read this code and make sure you install all the dependencies to be able to run this code. Because a lot of times you'll start running into errors and it'll be like, oh, you need to install this thing, and then you need this thing installed on your computer and you Know, a lot of times, like, you know, the versions of React or whatever you're using aren't. Aren't installed yet. So I just make sure it installs all the dependencies and then we're off to the races.

Totally. I agree. It's. It's. It's a lot of fun. See? Okay, there you go. You see it. It put it in the page. Obviously. I would probably go to Canva and use their quick, you know, background remover feature before you did that. So it kind of hovers over the site.

You can do that in ChatGPT now too.

4.

0. You can throw the image in 4.0 and say, remove the background and it'll do it there too.

Now wait, will it actually be like a PNG without. It'll be a clear background or will it be a white background?

It'll be a clear background. It'll be a transparent png.

Yeah, it's good.

I did not know that.

Yeah. So you should be able to just give it a prompt like use the same image and remove the background or make it a transparent png.

That's crazy. I did not know that.

Yeah.

Wow. I'm telling you, these design tools are trouble. That is such an ideal workflow is to just be able to just say, get rid of the background, add text on it. Like that is the future of design, in my opinion.

I wonder how many people.

That's why I'm so excited about like, maybe four O's getting the API or getting the MCP inside of Cursor and then it'll just do it straight from your ide.

Yes. Yes. That is huge. Like, mm, yes. That's insane. That is crazy, actually, because now what we can do. And so notice how we already have this image in here. We can say actually replace the image that we just added with the one in this message. I don't know if I worded that right. It'll figure it out. Now it'll be like hovering over the app and you know what I mean? And this is.

Yeah, it matches the color style of the logo and all that kind of stuff. There we go.

It's already done. And it added this and it did a quick edit too. So now it's hovering over it. And like, what we can do is we can say, make the rabbit animated. Animated. Have it polls. I forget the name of the animation and make it look fun. I don't know.

Sometimes it's fun to just let AI get creative with itself and see what it comes up with.

I mean, that's Vibe coding. I mean, that's. You just got. Sometimes you just gotta let the AI do its thing because, I don't know, it's probably a better designer than me, but. Yeah, I think that's cool. And V0 just recently added a database feature so you can actually store data. I haven't tried this yet. I think you. I don't know if it's through Supabase, and so I just know that they have a database feature. So it's. V0 is basically trying to become like Bolt, where it's like you can build a full app that uses AI tools that, like, uses APIs, that has fold files that you can add and a database. And so we're definitely seeing a lot of competition in this space, for sure.

Can you actually import something you already built into V0 and iterate off of it, or is it sort of designed to start from scratch with V0?

Hmm. I've never imported anything, actually. Yeah, it feels like this app is definitely made because it's made with Chad CN or Shad cn. And so that's why it has, like, relatively similar components when you make it. And I don't think you can import anything that's not in that framework. And so I think for that reason they don't allow it. Wait. Oh, you can see it's like animating a little bit.

Can you see? Yeah, yeah. It actually added a shadow below it too, and the shadow is sort of animating with it. I mean, it's subtle, but it's there.

I like that. That's cool. And yeah, and you can just vibe code and. And they also actually made like, these little shapes that are animated. Oh, yeah. That's kind of cool. I didn't even ask for that, but, you know, I said, and make it fun. And there you go. There's the fun. Yeah. What else do you guys want to say?

I was just thinking about. I've known Guillermo, who created Vercel for a long time, like, back from the early, like, Node JS days. And back in the day, like, Next JS was like the easiest way to create like a very simple, beautiful website. Like, very simplistic, though. Minimalistic. And this is like the natural evolution of that. Now when you want to start something new, instead of just creating a basic, bland next JS website, you go in here with your V0 and you can create whatever style you want versus just having to accept their style.

Totally.

Let's like make it so if you press get started, it opens like a modal box that you can. You Know, put contact details in or something. It doesn't have to actually like submit the contact details anywhere. Let's just get the design working.

I want to test this now with doing 4.0 to actually generate the different elements of the website and then paste that in and see.

Yeah.

Like, you know, like, here's a testimonial section. Like, make that really cool. And then like, oh, let's make or have four.

Oh, just design a whole website layout for you. Like, and then just pull in the image and say, make this website from this design.

Yeah, I bet there's some still limitations there, if I had to guess. But this is. It probably could do a lot.

Yeah. Here, I think let's, let's create. So if we hit get started, let's have this character animate into a different position. I've tried this before. Let's see if this works. So I just said to ChatGPT, and this is one of the, this is why it's so powerful for designs. Like, you can do this with buttons. You can do it with basically any component on your app is you can get it to like slightly change and like when you do something on the site. So I, I, I have an idea. It, it'll make sense in a second. Hopefully.

This looks different and it's automatically giving it the transparent background too.

Amazing.

Oh, nice. I like it.

Okay, so here it's getting pissed off.

At you if you want to get started.

Ooh, no, no, no, no. Here, what we can do here. Okay, so what happened? Okay, this is what we'll do.

Please have a third button to the right of Learn more that says not interested. If the user presses not interested, the rabbit will change to the image that I just uploaded and then grow 3x the size and.

Attack the user. Do a shake animation and then have big text pop up below it saying, no, you have to press get started.

Nothing but the money from money. Python, the killer bunny.

So was that Whisper Flow that you were using for that?

Yeah, that's Whisper Flow. I've been using this since. Geez. Right when I started vibe coding. I actually know the founders.

Yeah, it's a great tool.

That's what you use too, right? NATO?

Yeah, yeah. Riley. After we talked, I injured my hand and then I ended up having to use it. Like I kept saying, oh, I'm going to use it. That's really cool what Riley's using. And then I injured my hand. I'm like, oh, this is, I can still get a lot of work done by using Whisper Flow. It's just been amazing.

Oh, yeah, it's great. I hardly type anymore. And it's spelled whisper without the H or the e, so I always got that wrong. Very easy. I love it because it's hidden. It's like a very subtle tool. It doesn't try and be more than what it's made for.

Yeah.

So you just have, like, a hotkey set up on your keyboard. You press it and you can talk.

It's the bottom left button on Mac. You just press it all. From a user interface standpoint, it's just this little thing right here. When you release, it just takes what you said very quickly, and it will, like, try and predict what you say. Like, if you say something and then you're like, wait, actually, I mean this. It'll just do what sentence it thinks you mean, which is. Oh, cool.

Yeah, it's really amazing for people who have an injury because, like, I mapped it to a. I actually have, like, a MMO mouse with, like, all the buttons on the side, and I mapped different things, including whisper, to one of the buttons on there. So I was using one hand. I still could get work done. So, like, I'm using one hand, and I just press the hotkey on the.

On the.

On the mouse button and then just start talking. And so I could still get everything done or a lot of things done with one hand.

That's awesome.

Yeah.

Yeah. Love it. All right, Wait, where were we with it? Oh, yeah. Not interested. So if we press this.

What the.

He's like, we should change to red. You know, it should be. We should. You know.

Well, all right. Yeah. I mean, we could. If we. If we had more time, we could probably change it, but here. Here we go. Not interested. That's actually kind of fire. I like that.

See, the web's getting more fun. I told you guys. It's getting more fun.

Yeah. So if you're listening, it zooms on the bunny. The bunny gets pissed off and starts, like, shaking.

It won't let me out. My computer's hacked. It's over. I mean, the bunny's got it. Okay, there we go.

Now we take all their money, and it's a great business.

That is really cool the way it, like, animates in. That looks kind of natural.

I.

There's something here. I'm gonna. That's fun.

Anyway, the rabbit looks so cute until you're not interested.

Like, on a pricing page. Right. Like, somebody's, like, looking at the lower tier and, like, the. The bunny's kind of sad now.

It's like.

When you cancel your subscription on Duolingo, the. The Green owl, like, pulls it, I think.

Yeah.

Threatens you or something. Yeah. But, like, even something as simple as this, like, this used to take a while for someone to, like, design a character and create an animation. Like, look at us. We're doing it for no. Literally no reason. Like, we're just, like, sounds kind of fun.

Often would take a try. Now a person with an idea can just do it themselves, which is wild. And if you think with four. Oh, all the things you're gonna be able to do with that, it's just nuts. I saw this tweet from Balaji earlier where he was talking about all the different things are gonna be impacted by this. You know, if you think about advertising, like, what does this do to advertising now that you literally can just take an ad and just say. Or you can make an ad from scratch for anything you want, you know, and then now you can create a landing page with V0. It's like one person could create, you know, in one day, could create like, 100 landing pages with, like, 100 different ads and then have AI help set those up probably in the near future. And that's just like. That used to take an entire team. That could be like, 10 people, 20 people.

Yeah. I mean, in the same way that I was making thumbnails earlier, if you wanted, you could go find ads. Right. Like, in the marketing world, they, you know, they have a thing called a swipe file, right. Where you keep it. Like, you save images that you find around the Internet that maybe you want to use again later. You could have a swipe file of ads that you really like, throw them into GPT4. Oh, and then say, make an ad like this, but use my logo and change the text to this, and it'll make the ad, but with your logo and the new text that you want, but in that same design. Yeah, like, it's. It's just wild.

I saw.

I saw one example where somebody took a product photo of theirs and they said, make me an ad. Like, imagine the people from Mad Men were making an advertisement for this and make the ad, and it looked beautiful. It was like a really big thing. And like, you know, the huge text, you know, the huge font and just. Just long text underneath, and it was just. It was beautiful.

Super cool.

Is there any other ground that we want to cover in this episode? I mean, we've sort of talked about all the different models that have come out since the last time we've chatted. We've talked about the updates to V0, we've talked about cursor versus windsurf, we've talked about MCPS. I feel like we covered a lot of ground, but is there anything that we haven't touched on that we probably should?

Honestly, the only thing that I think should be said at this point is just to get your hands dirty and try this stuff, because there's no. You can spend. You can only spend so much time comparing the top models and the top tools before you're, you know, you. Your ratio of consuming, you know, content on these tools and actually using them should be, you know, you should spend more time using these tools because that's actually how you learn and that's how you actually build, like a felt sense for it. And like, the longer I've used and I've been vibe coding with a purpose, I'm actually like building an app right now that has been like, really, really, really fun to use. And for me, it's. I've cared less about the tools and more about the just like, kind of just celebrating the fact that we have, like, we can vibe code. You can just concept things and you can share ideas for apps or for specific features or designs, including incredibly fast, if you just know how to use them. And so I would just work on. Just as soon as you have an idea, make it real and practice that over and over again, because eventually you'll find an idea you want to spend, you know, the next decade on. I truly believe that. So that's kind of my thoughts on it.

Totally agree. I think, you know, one of the things that I've found really helpful, and I think we might have talked about this the last time we chatted, is that it's cool to just make little quick apps that solve workflow problem. Workflow problems that you have. Right. If you find yourself doing something repeatedly, you could probably create an app that can automate that thing that you find yourself doing repeatedly. And you could probably do it in less than an hour. Right. It's just. It's crazy that you can just, you know, have AI solve these problems by creating little apps for you. I think one piece of advice that I would give is also learn about, you know, committing and pushing to GitHub as well. Because one of the things that I've run into in the past is while coding, I will have it make like a small change, but it'll break something completely somewhere else on the website or on the app I'm making, and then I can't actually get it to restore the previously Working version. Well, if you know how to use GitHub, that kind of solves it. I know Cursor actually has a feature built in where you can sort of roll it back, but I've noticed sometimes that doesn't always work. I'll try to like roll it back and the rolled back version doesn't work. So I found that like, having it restore it from like a previous commit on GitHub has been still the best solution. Just to make sure you're sort of saving as you go.

Yep. And it's as simple as creating an account on GitHub, creating a new repo, and then pasting the repo into cursor, saying, commit this and Cursor will commit it directly. All you have to do is ask. And so it's not as hard as you think if you're worried about Git. I was scared of GitHub to start because GitHub sounds scary.

I didn't know you can do that. I'm still manually doing it. I didn't know you could do this through Cursor.

It's crazy. Oh, just ask. Just ask it to create a repo. We'll do it.

Well, there's even a GitHub MCP now too, so it's even easier. It'll sort of check directly with GitHub to see what changes have been made. Instead of having to run a bunch of Git commands and pull information back, it will just look directly using the mcp. So it speeds it up slightly. But yeah, you can directly connect now.

It's amazing.

So people have no excuses now. Like, if they want to build something, they've always had that idea and they're like, oh, I could build that. But I don't, I don't, I can't hire all the engineers or the designers. You don't have that excuse anymore. Like, you can, you can do it.

Become the designer.

Yeah.

Yep, yep. I mean, I've always sucked at design too. And like some of these tools, you know, they've, they'll make designs and I'm like, that looks good. I could have never thought of that myself.

If you're a decent designer too, they make you better, which is crazy. I've always, I've. I've always been around a bunch of amazing designers and, you know, kind of through osmosis became a decent designer myself from that. But then like, these tools will make you better as well, which is crazy.

Well, cool. So how's. So what sort of apps are available? I know last time we chatted you were working on Yap thread. Is that still out there available? Is that one of the ones you're still pursuing?

Yap thread is not what I'm working. We've the company that I started where we decided we were going to build some apps. We found another.

Thing we wanted to work on and I'm very glad we did. It is tailored specifically for people interested in vive coding. We haven't announced it publicly yet.

But I'm sure you guys have seen videos of it on Twitter. That's all I can say.

Okay, cool.

Well, once it's finally launched, maybe we'll have to have you back and you can give us a tour of it.

Cool, will do.

Awesome. Well, thanks so much for joining us and demoing this stuff and sort of nerding out about Vibe coding with us. Really, really appreciate you hanging out with us today.

Cool. Yeah, this is a great time. Love.