Vercel's CEO Shares 5 AI Startup Ideas So Good You’ll Quit Your Job

I sat down with Guillermo Rauch, the founder of Vercel, which is a $9 billion company. In this conversation, he dropped 3 powerful startup ideas that really got me to stop and think. We also went through v0, his vibe coding product, to tell us how to make money from it, how to build software from it, and how to use it to the best of its capabilities. I always find when you sit down with the founder of these vibe coding tools, they really, really give you the sauce around how to fine-tune it so you get the most out of it. I'm not sponsored by Vercel. I have no relationship with them. Just thought it would be cool to give you some ideas, see how their platform works. I hope you find it interesting. At the end of the episode, I bought a domain that I'm giving away to one of you. I think it's an incredible domain that all you got to do if you want it, like and comment and I'm gonna pick one person, one person randomly to give it away. I can't wait to see what you build. Enjoy the episode.

The Startup Ideas Podcast. It's sipping time, baby.

We got Guillermo on the pod from Vercel. Guillermo, by the end of this episode, what are people gonna learn?

Well, hopefully they're gonna learn a little bit about my system, how I think, how I use our own tools, how I prototype, how I come up with ideas. I have some ideas that I think are good. Maybe most are bad, but some might be gems for some of the listeners. Yeah, I like to work in public a lot. At Vercel, we do a ton with open source. Maybe I'll share some of the things that we've been open sourcing that can be great starting points for people that are entrepreneurial. And yeah.

Cool. And I just asked for one little commitment from you. You know, on this podcast, we talk about sauce, you know, giving the sauce to the people so that can, you know, can you commit to giving this all sauce?

I'm all sauce.

All sauce, no breaks. Yeah. Okay. Let's do it. Let's go. Let's, let's, let's rip.

All right. So Greg, how familiar are you with v0?

I'm pretty familiar.

Okay. Awesome. Yeah. Yeah. I think, I think I've seen some of your tweets. Uh, so what you're looking at here in case you're not familiar is, uh, the Vercel workspace for v0. So I'm literally showing you like all this awesome, right? Like our, our own workspace, how we use v0. But you're looking at the favorites, which are mine, obviously. I created a couple things that I wanted to show you all. So, and by the way, funny little behind the scenes thing. This is how I pitch product. We have a lot of huge enterprises using v0. Yesterday, I met with one of the world's largest companies. They came to the office, and I basically did a version of this. So it's also a little bit of advice for for people watching, like, how do you sell? Like, do you do decks? Like, I try to show product as much as I can. So I'll walk you through a few things that I've created. So one use case of v0 that's really interesting is freeform data visualization. So it's really hard sometimes to explain complex technical concepts that have to do with different things of the Vercel infrastructure, etc., to customers. So I use v0 a lot to basically give some prompt ideas and then come up with unique visualizations. So what you're looking at here is that something that I think could have taken me quite a long time to create with slides or whatever, and the AI walked me through how to represent this, how our fluid compute system works and create something that I can hand off to other teams, I can hand off to customers, I can hand off, you know, I can further solidify my understanding of technical concepts. I actually use v0 for learning a lot. So that's a fun one. I'm actually really proud of this one that you're gonna look at now. So if you're super familiar with Vercel, this might look like our blog. And so what I did is I started from the point of view of the Vercel blog and what it looks like. And by the way, if you're, if you're curious about how v0 works, it's like, it's all prompting. And so really what you're looking at here is that when I engage with my Vercel team, I could take two routes. Is that one, I can make a suggestion based on like 10 Slack messages, or I can just talk to v0, create something that I kind of like, and then share with the team. So what you're looking at here specifically is it's a pretty novel component. I think I've seen it this only a handful of times on the internet. So when we launched the Vercel MCP, uh, pay attention to this little component here and the evolution of the video. I suggested to the team that in order to highlight our partners of this launch, which were Anthropic, uh, Cursor, I believe VS Code as well, I wanted to highlight them front and center. Uh, MCPs are a highly, highly, highly technical concept. So, um, I wanted to show you, okay, how does Vercel MCP work? And so I actually cooked on the entire component. Like you can, uh, kind of like jump to the different markers. And so the, then the team can take it, implement it. They can actually share and sort of copy paste a lot of the code. And now this became kind of a foundation of, of our blog system. Whenever we have a changelog or we have a new blog post, now we have this really cool resource. Little bit of a spoiler alert. We have the ambition that whatever you vibe code in v0 can actually become what we call remote components. Imagine prompting on a slice of a website and enabling anyone in your team or organization to do this. So this is a kind of like a little example of like what that would look like. Like I'm basically like prompting the blog. I'm also pretty proud of this one. I collaborated with our head of design on this. He sent me kind of like a rough sketch of— and I'll explain why I got so involved with this. We're launching a new product, or we launched, which is already kind of growing sort of exponentially. It's called the AI Gateway. Greg, I don't know if you're familiar with things like OpenRouter or AI Gateways.

Yeah, OpenRouter we've covered, but yeah, you know, explain it a little bit.

Yeah. So the main idea is that when you access models, you, one, might not have made up your mind about which one you need. Two, because of the insane demand for AI tokens, a lot of model providers are unreliable or out of capacity. And three, you always want the best price and performance. And I will actually add a fourth one. You don't want to lock yourself into a model, even if everything that I just mentioned is perfect. You have the right model, you have the right provider, whatever. Like, it's not might not be the smartest thing to lock yourself in. So we created the AI SDK, which is the universal developer experience for AI. I'm very proud of it, particularly because I didn't invent it. My team did. I invented Next.js. I have that going for me. But this is something that emerged organically at Vercel when we were building v0. Maybe a metaphor for people listening, it's almost like the React of AI. It's a very, very, very cool developer experience. So AI Gateway is very exciting to Vercel as a company because it's a new entry point into the Vercel Cinematic Universe, meaning you can adopt this AI Gateway regardless of whether you're using any other Vercel product. And so what I wanted to basically communicate to the team is a couple of things. It's like one, and frankly, this is inspired by Apple. I have to come out and say it. I wanted to create this sub-product navigation. So that I can go to vercel.com/ai-gateway and you believe that that's an entire company. And second, I wanted to show this really cool— this actually was my idea— this really cool animation that represents how easy it is to switch between models. It's like that easy. Also, we wanted to represent that we support not just our own AI SDK, like I said, that Reactive AI, but there's other options, et cetera. If you go to the page today, does it look exactly like this? No, but that's the point. I was able to have this super high fidelity conversation with the team. And I did this on a Waymo ride back home. It was actually pretty fast for me to do. Let me jump to something that I think is interesting for also for people to understand how we build v0. Like imagine that you have the ambition of building something like v0. So I have a surprise for you. We use v0 to build v0. And so, um, and I kind of nerded out hardcore on this, by the way, because of many, many moons ago when I was 16 years old, I created a component like what I'm about to show. Um, so I'll explain the growth mindset here, like from a growth hacking perspective. Whenever you share a Vercel deployment, you're gonna see them, uh, throughout the web as .vercel.app. The way to think about this is like, this is like the Vercel marketing machine. And what I mean by that is that we're not actually spending any marketing budget. You asked me for the sauce. This is the sauce, right? Like we created such a great product, hopefully, that people build, ship, and then say— it's almost like saying it was built with Vercel, right? And it can be toys, it can be big ideas, it can be your next startup, et cetera. And obviously we support custom domains, but with v0, I also wanted to preserve that identity. It's almost like a communication is like a significant bit. You want to communicate that the thing was built with v0. And so by default today, you might have already noticed this, we default to v0 as a prefix of the things that you ship when you hit that button publish. But I was noticing that when people were customizing, they were nuking the v0 prefix. And so I wanted to both give them the freedom to nuke it, but also if they don't mind it, keep it just like they're keeping .vercel.app. So Notice that if I select text and I say Guillermo talks to Greg, I'm not nuking it. But if I go to the beginning of the input and I press delete, I'm deleting that entire token. Mm-hmm. And if I press Command+Z, it comes back. So I couldn't have imagined describing this with any other tool. I could, I would've spent more time putting words down. And you know that batch little token thing that you've seen on Facebook when you do an @mention? Well, imagine it says, I actually had tried to do that, to be completely honest, and no one understood what I was talking about. So I just went ahead and built it. And because v0 is so smart, like things like the— you can tell it, well, when you hover, do this, select the text by default. All those kinds of things are just prompts. So, and 14 prompts probably took me like, I don't know, 15 minutes. So really cool. You know, I could keep going on some of this. We do all kinds of like product specs, but I also wanted to show you something that kind of became what I would almost say, like the foundation for like a real product almost. Did you see this on X, by the way? Okay, so, and this is connected to something I think really relates to your audience, which is that I've posted all of these free ideas threads. And no one picked up on this one. So I'm actually gonna like, uh, you know, uh, point this out to the people. I think one person paid attention to this one. So I had this concept of an AI camera. Um, the idea of an AI camera came from a frustration that, um, my, my, uh, relative takes, loves to take photos of the moon and like, no, no diss, but like They're so bad. She loves the moon. Heart goes out to her. She's obsessed with this thing, but they're just so bad. And I was thinking also the— I just upgraded or downgraded, you could say, to an iPhone Air. The camera got worse. And so I was telling my team the other day, I don't mind because I love this slick form factor and also because I believe that in the future cameras will be inputs, or what you take with a camera will be an input, not an output. And why is it going to be an input? Because every photo will go through an LLM or an AI model. And so that was kind of like the idea. And so I actually, because I got this iPhone and I was telling the v0 team about it, we at v0, we don't have devrels, we have vibrels. It's someone that vibe codes to show the potential of the tool. And so he built this. I took a selfie of myself when I was in London, I believe, and then I took another one. And so let's see it here. So I opened it. So you can— the idea is that you can choose between different filters, actually, because let's see, Kill Bill, because I'm from Argentina, I'm going to go with Argentino. So Photo goes in. So it was actually quite easy to prompt the success of this. And by the way, like what I mean by success is like it went quite viral. The success of this speaks to— okay, that actually looks like me a lot. The success of this speaks to just how simple we made the interface, which I'm really proud of. You know, we obviously— it's inspired by the iPhone camera. We used really cool prompts for the different styles. So I'm gonna go, let's see, let's do another one. I'm gonna go with maybe more focused. I'm gonna go with disco now. Okay. So when Nanobanana came out, I told my team, and by the way, for people listening, I still believe this. Nanobanana is a GPT-4 moment for image models. And And you can create amazing things. I think there's gonna be a lot of great consumer products that come out of this, little great utilities. I was showing this off, by the way, going the meta work here. When I was in London, I was obviously pitching to companies, startups, et cetera, like what Vercel and V0 can do for you. And I was in a lot of meetings where like this was kind of like the memorable bit of the pitch. Taking a selfie with, with someone and like showing them what, what, what v0 can do. So I actually ended up executing on that idea. So from the, from the free ideas thread.

So that, that whole product that you showed me you made with v0?

Yeah, 100%.

And how long did that take you?

Sorry, say it again.

How long did that take you from start to finish?

So it was lunchtime here in SF. By the way, you know what's funny about the importance of— and by the way, this is not just me patting myself on the back because like, I have a lot of bad ideas. Like you should see my team, my chats with my DevRel team. But I do think it's in the age of AI. And I think obviously this is going to resonate with you preaching to the choir. The value of ideas is pretty freaking significant. Why? Because I pitched this idea to a lot of people and like they were like, I don't know what you're talking about. Like, oh, that's cool. You're a CEO. I'm not going to annoy you. But like, I'm not gonna be mean, but like a lot of people kind of ignored it. But the day that we're at that launch and I saw that the lens of the iPhone had just gotten worse and whatever, I pitched it hard. I was like, guys, we have to actually do this. And so, you know, most of the process was actually downloading it into the prompt. And so there's a lot of debate about like, is prompt engineering a thing and whatever. I'll say it again. If you have a vision in your head and if you're able to explain it in English, amazing things can happen. So I think it was mostly— so I explained it to my colleague and I told him, and in the beginning he first didn't get me. So it goes back again to like, sometimes you might get frustrated with AI, but remember, like, us humans are also not flawless in that department. Like, explaining yourself really, really well, using the right metaphors, et cetera, is really important. One thing, by the way, that if anyone wants to continue to riff on that idea, which I encourage, is, and by the way, you can actually open it in v0. So like when you click this, you're gonna go to the v0, when you click Build with v0 here, you're gonna go to the template. So I had originally pitched it as Instagram in the sense that like I wanted this to be a preview of what the different filters are gonna look like. We implemented it as like the different types of like photo portrait, whatever, which makes it super relatable. I think that has a strength for sure. But there's another version. And this is what's exciting about AI because like the models are going to keep getting better. So the first pushback that I got was, oh no, that's going to cost us a lot of money because you're going to have to do 20 nanobanana runs and whatever. So we started litigating like, well, but you could do you. Maybe you could prompt Nanobanana to give you a grid of all the permutations, and then you tell V0 to slice it. And because like the vision was just like Instagram invented filters, we can invent AI filters. So you could have, you could have seen Guillermo Rauch's Solarpunk almost like a little preview, and then they click it and then we do a higher fidelity AI pass. So I think there's a lot more to do with this particular idea. But yeah, 100% V0. There's a nano by nano template. So if you go to the v0 community, there are a lot of really amazing starting points, right? So we have all kinds of templates that you can use to, to use as your starting points. There is game if you want to create games, landing pages. I mentioned data viz is really big for us, especially for teams that throw unstructured data, CSVs, whatever. There's really cool components that can— when Liquid Glass came out, like, everyone lost their minds and started trying to create different versions of Liquid Glass. So this is a great place to get inspired, frankly. I have my own contribution here. I guess I cooked, huh? So I love particle effects. And so speak of the value of vibe coding, we were hosting a happy hour with AWS at re:Invent about v0, and we didn't have a promotional artifact that I felt truly proud of. A lot of people, I think, that listen to us are always curious about like, how do you get big on X? How do you get a lot of impressions? And I'll tell you, you know, the, like, just because you have a lot of followers, it doesn't mean that you are putting things out into the world that people are interested in. So like, I actually use v0. A lot to cook on visualizations of data and things like that. That, um, yeah, that, so for example, when, when Next.js came out and we wanted to communicate that it, that in the past 12 months it had 500, over 500 million downloads. Mm-hmm. And then over previous 12 months it had a ton of downloads, uh, basically 500 million in, in 9 years and then 500 million in 12 years. I used v0. To create my banger tweet. So there's a lot, a lot of cool stuff here to learn from. So, yeah.

One thing I'm learning about you is, I know you say you don't have, you know, you have bad ideas and good ideas, but you do have a lot of good ideas and you have a lot of good taste. Like you're able to, you know, think of, okay, Instagram works like this, therefore I'm gonna use that as a reference to put it into v0. So, you know, what advice do you have for people who are trying to like build their inner Guillermo, basically?

Yeah. So you mentioned something really cool, which is that what comes out of me is good ideas. What started out was probably a ton of ideas, bad, good, medium, whatever. And then there's a filter that happens for sure. Yeah, I try to be a less is more guy. So like what I put out, I have some degree of conviction that it might hit. You also talked about the idea of, well, Instagram exists, therefore you can bring this over to this other product and whatever. I do think that that's extremely, extremely, extremely important to basically like accumulate a bunch of exposure hours to products in general. So even just from what I've been doing here, you might have gotten ideas about how v0, how I think about v0, right? Like you see the, the way that we have the community with templates, projects, search, like little nuggets of like how products work that I think are really interesting. I'll show you actually something also super cool. Notice that when I've been opening these v0s, they're in full screen, so like they're ready to be shared. So I care a lot about the first impression that a product has on people. And so I can reveal the chat, but the chat is not interesting to the presentation that I've been doing. So I wanted to also communicate that to the team. And I basically created this like mini V0 inside of V0 to like basically show like, Imagine a world— and by the way, the design ended up being quite similar to what ended up landing at the end. Imagine a world where it's all content, everything is computer, and we hide the chat. So there's one version of Vizzier that was like so much chat in your face. And the, and the nugget of this idea, or like the origination of this idea, came from watching people use the product. A colleague of mine that works on Next.js sent me a, uh, a photo of his girlfriend working on a V0 for her education class. And she had a big desktop monitor and I saw that half the screen was chat and the other half was the content that she was working on. And immediately I had the idea of like, well, that doesn't seem right. And so the TL;DR there is that you have to always be rethinking your assumptions, trying to observe how people use your products in the real world. And you know, I, whenever something doesn't work, when I never think, well, it's the fault of the user, is the fault of the people, I don't have followers, whatever, it's, it's the product, is the design, is the idea.

When you, when you were creating, let's say, the Nano banana photo selfie thing. Did you prompt v0 with the reference image of Instagram, or were you plainly using English and just explaining what you wanted?

I think it was all English in this case. Which kind of speaks to like, there is one defining characteristic that I do want to point out is that you have to be able to visualize things in your head. And if not in your head, like, you have to have some close reference so that you can guide this system. Right. And so all of this that you're looking at, it's just maybe paying more attention to the details, right? Like if you open the, if you open the camera app, there is subtleties to the border, there's subtleties to the background, there is subtleties to the gradient. Another big thing that it might, might seem small, but it's actually freaking huge. Is that a few of the initial passes were overly complicated. So there was a process of trimming stuff down and deleting. If I could actually give one lasting piece of advice to everybody, including my own team, is always— and myself— is always delete, delete, delete, delete. There should be as few buttons, links, et cetera, as possible on any given product. And I think we kind of nailed that here because like How much, uh, how much attention can people give you, right? Like 5 seconds, 10 seconds. Another not minor thing is this is actually a very good, um, mobile product. So by having less UI, you are almost forcing yourself to have a good mobile product. What's very interesting about this little app, if you go to it, v0bananacam.vercel.app, is that it's, it's almost impossible to believe that this is a web app. It feels like you just downloaded a native app, which also speaks to the insane potential, I think, of these Vibe Coding platforms because we're soon going to launch the v0 mobile app. You're going to be able to create apps like this on the go. And we're actually putting a lot of work into the model, the context, etc. To make them really good on mobile devices out of the box. So TL;DR, fewer pixels, always better. And remember that most people don't have time and are on their mobile phones. So you're little— you have such a little window to get, to get this stuff right.

By the way, this just for people listening, like this idea that you had, like the V0 banana cam, the 20 different filters that you had, if you wanted to, you know, create an app that, you know, just did a little bit of cash flow and was a fun project to do, you can unbundle this and just do like Wes Anderson cam as an app, right?

Or— I mean, there's just so much to do with this. And you can, by the way, again, you can click Open in v0 and you can start prompting. There's so much to do with this. Yeah. Because I mentioned also that the original idea, By the way, I also will give credit to my colleague who suggested to focus it more on selfies. Because let's analyze the ideas thread for a second, right? AI camera was the first idea. And I was thinking a lot about— another inspiration for this, by the way, was like, we took a business photo with a group at a large bank. And it was so awkward that we had to take like 5 or 6. And I joked because obviously the conversation that we had had was about AI. I joked to them saying like, oh, don't worry, AI will fix this. But in reality it didn't. Like we took a lot of bad photos and we took 6 of them. It was awkward. The lighting was bad. I was like, guys, like, so to give you context, this was in a high-rise New York City. The background was all sun and the skyline of New York.. And I remember thinking to myself, AI has seen that freaking, uh, I mean, it's beautiful, but it's seen this skyline a gazillion times. Yeah. Uh, we are very clear on the photo. The lighting is not good, but our bodies are very clear. All it should take is, by the way, it shouldn't even take a prompt. And, and this we didn't explore with this idea, but you could ask AI first to critique the photo. So, um, that process that I did of saying, oh, the photo's garbage, I'm sure GPT-5 would do as well. Uh, and so once you get the critique from GPT-5, you give it to Nanobanana. And so this is why I also believe that people sometimes fear that like the one model is gonna take all the jobs. In reality, most cool things happen when you combine models, combine ideas, market it well, focus it well, create easy interfaces, etc. So I think since you're— since we're riffing on other things that people could do, the automatic perfect photo app where, uh, or the lighting fixer photo app and whatever, uh, especially if people can build the muscle memory to always go to you when they're taking that kind of photo, um, So by the way, it's fun to compare it now because like, I believe that AI camera was probably the best of the bunch, but it wasn't necessarily the one that people got the most excited about.

So what other ideas are in this list?

This created a lot of excitement that I'm still bullish. So it might— I might cook on it at some point, but obviously encourage people to try it too. So I believe that forms need to be disrupted with AI, period. All forms, contact sales forms. Uh, we're gonna be doing an experiment on Vercel.com for this. I think other, I heard rumors that another like big tech company's gonna try to do like a conversational sales form. I think it's gonna be really interesting cuz like the, the classic thing is, so if you, if we go to Vercel.com/contact, and then go to contact sales. We do have personalization here. Like we're trying to like eat our own dog food, right? So like it knows that I'm from the US, it actually knows that I'm coming from an enterprise, so it's asking me a few more questions, et cetera. So we're constantly doing experiments on this, but like what if you didn't have to fill out a form? And, and some people, by the way, might think like, why don't you just ask nothing but like the email? Well, Interestingly, we've done the experiment and like we get such an insane amount of input that to triage and get more signal out of it, we actually do need more information sometimes to actually be able to prioritize who we talk to. And so we do need more information, but a chatbot could ask you that for the follow-up. And so that's an experiment worth doing. The other one is— and by the way, Typeform, I think, cooked. Like, I wasn't trying to like tell Typeform, like, Sorry guys, like it's over. Quite the opposite. I think Typeform is really inspiring here because they created a— I'll say this in, in earnest, a revolutionary UI for forms because they broke it. They— I think they empathize with like the problem of like, I don't want to see the daunting list of 100 questions. I want to see only one question at a time and going back to like fewer pixels are always better. I think a Typeform maximizes the chance of success and minimizes the chance of error by just having the user do one thing at a time. But conversational AIs also ask for one thing at a time. And so imagine if you had basically an interface where you define the data that you want and then you let the AI create the intake interface. It might have elements of a form, by the way, but it might just be purely conversational. So, I mean, two questions.

One is, how big of an opportunity do you think this is? And two is, if you were going to prompt v0 to, to do an MVP of this, how would you do it?

Yeah, so how big can it be? Well, Forms, I think, are the most underrated, like, fundamental particle of the internet, right? Like, nothing happens unless— like, the most interesting thing of the internet is capturing input from somebody.

It's a primitive. Forms is a primitive of the internet.

Yeah. So the TAM— and when I sound like wannabe entrepreneur here, like, the TAM is the entire internet, the TAM is the entire world. But I do think that, like, I think for this idea to succeed, though, maybe it needs to be narrowed down to like, it's not just forms with AI, but like a specific use case. One company that I think has done phenomenally in this idea of like, we're making forms more whimsical, more domain-specific, etc., is Partyful. I see Partyful as like a really cool specialized form of sorts. And so the idea is big. The problem is that I cannot yet tell you like the exact incantation is going to get big. That's your job. But I'll tell you, people have the need to create Google Forms or Typeforms all the time and they're willing to pay for those products. And so what I would focus on here is like, can you demonstrate that by creating a more AI-native interface? And by the way, when I mean AI-native interface, I want to be really clear, it's for the, for the submission side as well as the creation side. Because what if there's like the fastest way to create a form on the internet because You just paste what you wanna know, like build me a form. You, you're not even gonna say this because this is gonna be like cool folder builder.com and whatever you're gonna say, need data on what this user thinks about my product. And because I wanna qualify them for early access VIP application, like something like that. And so AI is amazing because AI is so good at knowing what more, most, what, what the information that you're after is. And so it's gonna save you a lot of time. Most form builders have all this like drag and drop and click and select the type of the question and whatever. No, no, it should all be AI. The other thing that you can set out to prove is that you can have a more dynamic conversational UI to submit. Maybe even there is a way where you reach out to people through voice or text message and like the form submission actually happens in a pure conversational way, but in an ad hoc conversation. So how would I start prompting about this? Like I said, I always work backwards from the interface, so I would probably start describing to v0, well, I want a conversational chat app but it's for designing forms. The goal is that the user can talk in natural language, and then as they're typing and submitting, they're seeing on the right-hand side what their form is going to look like. It should use LLMs to parse the input, and I'll get something out, right? It may take me maybe like 1 or 2 more prompts to get it to look what I— like I want and to make it work well. And then when I have something, I start playing with it and I can start seeing if it feels right. The other entry point of this can be imagine the ideal— imagine if you have a vision for like the ideal submission UI. So one thing that I'm a really big fan of, and you're seeing that in action with v0, is hyperlinks. So imagine, put yourself in the shoes of somebody that receives your really cool hyperlink of your form.. So it's like supercoolform.com and the title of the form. And then imagine what interface do you want them to see? And so it could be actually something very interactive. It could be something where it's like, I just have a few questions and maybe it has a little character or maybe it's very sober and it says complete this questionnaire to get X. And maybe it looks like ChatGPT where you're just going to maybe it looks like a chat interface. And so maybe I'll start prompting Visio with like, build me a chat interface for a product that is like Google Forms, but way more simple. Probably AI already knows about Typeform. So like, you can tell it, it's somewhat inspired by Typeform. And the goal is that the AI will ask me a bunch of questions in order to meet this data structure that the creator is after. 'Cause I think what's really cool about dynamic conversation interface is that you don't want to let the user complete the chat until you got what you wanted. You got the first name, the last name, the street, whatever. Like all of those things would happen conversationally. I also would explore, you know, if I'm asking for an address, like what is the right input for asking for an address. So it's like more of a Google Maps thing. At some point, I will say maybe the screen splits into two. So imagine if like when you're asking for a location, because now we have all the space in the world, we split the screen into two. And on the right-hand side, we show the map. And the left-hand side, we show the input. So yeah, I may— this is kind of an encouragement for people to think about, okay, forms are a big deal and there's gonna be something new that uses AI and turns it upside down.

By the way, what, what, what you just said at the end is potentially a good prompt for, you know, ChatGPT. Yes. Which is like, hey, I'm trying to create the you know, a conversational AI Google Forms killer. I need help imagining what an interface which might, you know, be super clean, super fast, super beautiful to use that's different than anything that's out there. Give me 3 or 4 ideas and sometimes that could become the— Yep. The fodder for what you end up putting in v0.

Yep. I was gonna also say if you're more comfortable with engineering, What, what v0 is gonna use under the hood is this AI SDK that we created. I mentioned it, it's like the React for AI, and we have a lot of really cool templates for it so that you can get started with something that's already been built. So I'm gonna show you, I think it's chat-sdk.dev. So this interface right here, if you go to try it out, You're going to see this is basically an open source ChatGPT. I call this open source ChatGPT for enterprise. And the main idea behind this is it can basically— you can mold it into whatever you want it to be. So I'm going to ask it a question first. I'm going to say, hi there. So it's using Groq vision models. It's using the ISDK, obviously, so you can bring whatever model. Something really cool about this is we are big fans of this concept of what we call generative UI. I'm gonna ask, what is the weather in San Francisco? Notice that it's responding with that UI component. I believe the future of this AI conversational interface is gonna be this intertwining of chat and rich UI components.. But perhaps my favorite thing about this demo that is pretty wild is let's ask you to write an essay about Silicon Valley. Look at what's gonna happen when it calls, it detects and calls the tool for writing. It splits the UI into two. We also use this pattern in v0 itself, and this is all open source. So imagine this concept that I just showed you, but for that conversational interface that we're just talking about, where when I get the next question, I might actually get a much richer interface for answering that question. Maybe this is a really cool interviewing UI. Maybe this is a really cool aptitude test UI. Maybe this is the future of how you give students a really cool interface for a test. It's like, hey, I'm, I'm your digital teacher. Like, let's let's evaluate your skill in math. But you know, when you're teaching kids in education and AI, especially for children, is something that I'm really passionate about. I've been thinking a lot about recently. Imagine if like one of the steps in the test needs a like little virtual game for testing math and addition. And so this is the beautiful thing about v0 is like all of those things that I just talked about, we have examples of games, we have examples of like all kinds of, of, uh, of UIs like this one. So I, I do think the possibilities are kind of endless here.

I want, I want you to do one more idea and I just wanna talk about the template stuff. So what I think one good idea for people is to go through the templates and be like, okay, this is, you know, this is a really cool idea, but if I focused it on this niche, yes. Then I think like if this for education or this for K through 12 or this for fintech or this for whatever. And you know, I think that a lot of people are just creating their own stuff when they could just be seeing what other people have done and just kind of duplicating it and focusing on a niche. Yeah.

There's also, uh, OSS vibe coding platform. I'm extremely proud of this one. The team really cooked. Again, this is a little bit more advanced. Yeah, but we basically open sourced v0 if you think about it. So we have a bunch of models here and you can say build me a little to-do list app. Okay, I have to sign in over Vercel. You can, you can do it in at your home. But what you're going to see is that when you start creating this, we're going to write an application here on the right-hand side of the screen. We're gonna create a sandbox for every creation. This is basically like open source Replit, if you will. Another really cool starting point from, from the Vercel Template Marketplace. And again, these two are a little bit more advanced. If you are not that comfortable with coding yet, I would start with the v0 community templates. But you mentioned going back to some ideas, so maybe I'll show some of the ones from the most recent ideas thread. Yep. Okay. So Oh wow, speaking of vibe coding, this one, the inspiration for it. So it's a Notion-style document tool where every block can be vibe coded. For instance, if you need a chart, it's just a matter of adding a block backed by a prompt. These blocks could be live, reactive, and revalidated. So this one is taking this element further and realizing that you probably, instead of wanting to write everything, you want this to be like promptable blocks. Yeah. Could be interesting for like data viz tools. You mentioned like in terms of like niches, it could be a really cool way of creating like very interactive essays. The idea there was why are we writing everything from scratch, you know, and why are we limited to the 5 or 6 types of blocks that the document tools have invented and decided for us if we now have AI? I might decide to like, here I want, you know, a really like, right here I want like a mini game or I want a very, very cool chart or, you know, you know, another inspiration for database I think is going to be big is, have you ever seen those videos that showed the growth, growth of a thing when another thing was, was big? It's like the top one is like, it's like a wide rectangle, and then it walks you through the years, 2020, 2021, and then you see all of the players reshuffle and advance and whatever. The best data visualizations on the planet are not rigid. They're not part of like the line chart, the pie chart, et cetera. They might have elements of navigation, et cetera. So this is a big idea. Like I was trying to think like, what is the future of documents with AI? This one again, like, I feel like it's underrated. Like, this is like AI camera vibes because the— I think I mentioned the— what? Sorry, can you read?

Can you read the idea?

Yeah, the LLM vibes rater. Periodically ask AIs for opinions and rankings. What's the best burger in SF? Who's the best candidate? Data visualization could look Google Trends-y. Uh, use ISR, I mentioned like a nerd thing of like, you don't want to like query the AI on demand. You want like a really good caching system. And so there is a technology that Vercel has called ISR that's really good for this, especially for high traffic websites. This tool can help the world become more aware of biases in AIs, be the wire cutter for everything, entertain, inform businesses on how they're falling in or out of favor. The idea is that this is kind of like a content virality idea. And it's a good way to like zing the AI model vendors. Like everyone loves to expose biases in AI. And like, uh, there's a thing that had gone really viral during the last US election, uh, when you would ask it about Trump and different models would create different things and. So there is almost like a polymarket, but of AI responses thing. It's almost like a free content generation engine that I think is going to interest people. What does ChatGPT think is the best burger in SF? A lot of this, I think in order to nail this idea, you have to find how do you, like, how do you visualize those questions and the question selection. So that it sparks people's interest and then ride news cycles so that when something interesting happens in the world, you're able to like go to this site and find out like what the AIs think or whatever. I also think that the timeline is also super cool. You can almost be like an archive.org, but for what LLMs have been steered to thinking. If you can discover interesting patterns over time, that can be huge. Like, you ask Claude or Grok or whatever, like, who is the smartest person in the world? And you see it change. Like, in 2020 was this, in 2021 was this, in 2022 was this. Or I think— I don't know exactly what— again, like, there's a lot here, but I think there is something that could have the potential for virality.

Give me one more idea.

All right, this one I actually think I might build at some point because I need it. Like, there is asking an LLM. Okay, the prompt or idea is deep-est research, emphasis on est. I noticed that when I really need to study a topic in depth, I don't want to bank on the viewpoint of a single LLM. Many times I fire up ChatGPT, Grok, and Perplexity and Claude. In a bunch of tabs. I would love a tool that uses all the available intelligence on the internet to produce the best possible report. Key, tell me how the experts, aka AIs, differed, especially if they have contradicting facts, figures, or conclusions. So have you heard of this being done before?

No, but I'm obsessed with the name Deepest Research, by the way.

Yeah. Take it. Yeah.

Um, also the idea— I'm gonna— I was just seeing if deepestresearch.ai is available. It's $30,000, which honestly isn't crazy, but it could also be Deepest.ai. I'm— would you say Deepest.ai? I was thinking that.

Yeah. Cuz I also think people underestimate how important it is that you build something that is so simple and effective that you return to it by typing the URL. 'Cause like maybe you had, have an instance where you got exposure through it on X, you went, you clicked the link, maybe you didn't have quite the use case for it at that time, but then you're gonna be in a conversation or something and you're gonna say like, oh shit, I need a, I need to do the deepest research here. And you need to be able to recall that URL and then type it into their mobile device and This is why I was insisting on like the AI camera being successful because it was mobile friendly. And, and so you need to be able to like get back to the URL quickly. So I think short here might pay dividends.

I'm going to buy the domain deepestresearch.net and which is kind of like old school '90s vibe because that'll be the— I do like the .net. The .net is under— I actually think is criminally underrated. And I'm gonna give it away, you know, if you want to compete with Guillermo, because he's going to do this in the future.

He's— well, keep in mind that my— the way that I execute on this is like I open source things, so like you might actually take benefit from it.

But you should race me to it for sure. Race Guillermo. I'm gonna, uh, for people, I'll just pick a random comment and, and I'll give it—

I'll gift it to someone, um, and let's see who builds it. Okay. Yeah. And to give you some more context on this one, I think the people that select into this kind of thing are going to be highly, highly, highly qualified buyers, I think. Yeah, you don't do deepest research of like, you know, what kind of banana like grows in the Mediterranean, whatever, like none. But it's going to be for people that are doing financial stuff, that are doing competitive research, that are like, maybe they're like really stuck on a bug and like they need like every angle explored and whatever. You might want to have some prompt presets like select the level of the kind of analysis that you want to do. And the prompts would be like kind of like your secret sauce, like maybe do like analyze critically, find edge cases, tell me about public opinion. So there might be— by putting this tool in front of users, you might get into some really cool, like, what kind of research do you want to do? And again, like, you know what's really interesting about the world? You can argue that ChatGPT will continue to get super sophisticated and will eat every startup and whatever. But what I find is that some products are so important, or some capabilities are so important that they always deserve their unique entry point, their unique URL, their unique interface dedicated to that problem. And you might actually outcompete the giants in things like this, much like what I think ChatGPT is doing to Google. Because what ChatGPT is doing to Google is that I don't think to go to Google to, ask this long, like long-form questions with prompts and I want to engage in a chat. So it's so hard for Google to compete with ChatGPT because the best they can do is do that AI overview thing, right? But pay attention to what they called their ChatGPT AI mode. You have to go into a mode. You have to go to a quote unquote, a different product. In order to compete with ChatGPT. And so my argument is you can do that to ChatGPT itself, because when I go to ChatGPT in order to do deep research, I have to get into a mode and they're trying their best to route and whatever. But when I get into that mode and I want all of this new capabilities, I maybe there's buttons, maybe there is levels of depth, depth of research. It's like a pro tool for deep research.

Damn, I want to do this idea. I'm like, I don't want to give away this domain to someone, but we'll give it away. We'll give it away. All right. I think we're out of time. I, I, I'm sorry.

Sadly, I want to keep going, but, uh, uh, uh, yeah, it's, uh, there's a lot to do with this.

Sadly, we're out of time. One thing I, I've, my biggest takeaway from, from just talking to you is It really is like the limit is your ideas, you know, coming up with good ideas. The tools are out there. Uh, this fires me up. Ji, thanks for coming on. I'll include, uh, some of the links, um, in, you know, like the templates and, and all that stuff in the show notes. I'll include where to find Ji on X. Anything else you wanna leave people with?

No, uh, if you do build any of this, just ping me on X. I'd love to take a look and, uh, maybe give you feedback.

I appreciate you, man. I'll see you next time.