Guillermo Rauch: Vercel CEO on how v0 hit 3,200 PRs merged per day (and lets anyone ship)

I'll say one thing about Vibe Coding. It's really easy to go from 0 to 1. I think we've all seen the demos of I prompt something and it's cool. I think what's harder is to iterate on a project at scale and to deploy changes safely. Every marketer I ever sell wants to change this page at some point. And the old way was one of two ways. One is what I call you had to petition to the government. You had to go to engineers and say, engineers, can you please add a logo over here or whatnot? Or pray that the CMS was perfectly wired up for any ambitious mission or dream or idea you had. So now they can just open this page in v0 and prompt anything that they want.

It reduces the friction of getting something live really, really low. The humiliation ritual of prioritization goes away, and you can actually focus your time on defending the merits of an idea, on the actual idea, as opposed to the hypothesis of the idea that then has to be implemented. And so I think it changes the speed of companies in a really significant way. So this this is truly a first-time VibePodcast that we're doing together. And I wanted to introduce myself. I'm Claire Vo, I'm a product leader and obsessed with AI. And I have a podcast, How I AI, where I teach people how to build better with all these new tools, including ones that we're gonna see today. And I'm really excited to have you here, Jeet. First, we're just gonna get to the, the thing that everybody's wondering about. What is your most favorite feature that you released this week on v0?

Well, I'll tell you, the hottest thing in AI today is skills. Everyone is excited about the fact that we can now augment agents and AI applications and agentic engineering with skills, like skills that the model doesn't yet have. And so we launched skills.sh, and the beautiful thing about what we'll show you today is that v0 can seriously go from prototype all the way to production. So we're able to conceive changes to things like skills.sh. I'm going to show you really quickly. Skills.sh is a new— you can think of it as like npm. It's a hub, an open ecosystem of skills. And it's pretty dramatic what's happening to this site. So you can see that we now have 34,000 skills submitted by the community. And this website has gone viral all over the internet. It's hosted on Vercel, but the most exciting part to me is that it was conceived in B0.

I have a quick question for the audience. How many of you have installed a skill in the last week?

Oh, wow.

Okay. A lot of people. Skill build. Um, how many of you have the top 3 installed? Actually, top 5.

It's very heavy at the top, right? It's like, these are ripping.

Oh no, I have top 7. Okay. Yeah, I have the top 7 installed right now. This is a really great resource. So for folks that are maybe watching this later or haven't been familiar with SKILLS, SKILLS is now this standard that a lot of these agentic frameworks are using to help you repurpose and reuse best practices, step-by-step flows. And so, for example, I use this Remotion best practices one to let me import components and regularly create videos really, really quickly. And I would not have been able to do this without the expertise that's been packaged in these best practices that were installed with one line, um, using skills.sh.

I think it's also worth noting maybe to peel the covers of how Vercel builds products. Yep. Skills.sh was a thing that was just conceived at the moment of inspiration. We started prompting, hey, wouldn't it be cool if this thing took shape? Would we discuss, for example, what it should look like? We, we've been calling this a style. Terminal Core, cuz it looks a little bit like— this is my contribution to the project. I was like, hey, wouldn't it be cool if we make the top of the website look like a terminal? And so the, the, the process itself of building this was very much prompt-driven, I'll say, like chatting in Slack and saying, hey, wouldn't it be nice if we had a hub for this? Just very iterative, very collaborative between the team members at Vercel. And what's really cool about this again is that it's really fast. So it takes advantage of all of the Vercel infrastructure primitives, even though it has 35,000 of these skills. Like if I start like hardcore scrolling this and then pick a random one, all right, Swift, Taylor Swift, you're going to see like the page transitions are— it's probably SwiftUI, okay. But all the page transitions are instant, production-grade. You needed to scale, there were going to be a lot of eyes on this thing.

Okay, so I think we want to get to our first workflow for How AI, and you just want to show us how you either developed this or how you and the team improve this over time using the tool?

I'll say one thing about vibe coding. It's really easy to go from 0 to 1. Like, I think we've all seen the demos of I prompt something and it's cool. And I think what's harder is to iterate on a project at scale and to deploy changes safely. In the case of how we work on Vercel products, we always work on branches and we take advantage of branch previews and then We code review and then we merge them. So we're basically going to be showing you today is how we brought those ideas of hardcore, heavy-duty, production-grade engineering to v0 itself. So I'm here, I'm finding that same project that you just saw, skills.sh, which is piped into— it's basically backed by Git. The engineer who built this pushed some code. 3 hours ago, and you have this new button within v0, which is a new branch. So what this is showing you is that v0 is making the Git flow of creating branches a first-class citizen of the product. So I'm gonna create a branch and basically this is gonna give me the same sort of like chat experience you're used to, but notice here at the top I get this beautiful new convention of project/branch, right? So I have the v0/rouge-g branch, and here within the preview, you're gonna notice that just like if you had cloned the project to your local device, we both have a full, full-scale, uh, VS Code editor. As well as the real project running within v0.

One thing I wanna pause and notice, because I just have a laser eye for product, is I love that you use the convention that all of us with engineering teams use on our Git branches, which is who's the contributor slash what's the feature. And so what I think this is really interesting, you know, we're gonna talk about how you actually use these tools to build, but I also think there's a flip side of how you design great AI products and agentic products, and I still like the small design tweaks that make something like a v0 feel like a collaborative teammate on your pool. That's right. So, I, for all the engineers out there, I noticed, I noticed that little convention.

So, I, I, and what you're gonna see in the design philosophy of the product is that we really wanted to embed those little details of what makes a real engineering workflow come to life., but in a really easy way, right? Like at the end of the day, I didn't have to go to a terminal or boot up GitHub Desktop and branch manually. Like it's the Stone Age. Just press a button and now I have a branch running. So the main idea here is that within this preview, I have the full skills.sh project running. It downloaded dependencies. It installed the exact versions of Next.js and every dependency., within the project. I have it all running here. I have it obviously within a staging or dev sort of environment. And now, you know, I, I could navigate it like, uh, I could navigate the production website. I could explore it. I could, you know, use all of the capabilities that v0, uh, brings to the table. But I figured let's actually build a feature that we could ship to prod.

Yeah. And one thing I wanna pause on what you, I think, glossed over a little bit, which is the fact that you have this VS Code instance, the fact that you have all your dependencies installed, the fact that this is running both with code and a preview for anybody who's less technical out there, and maybe a lot of your users that are using v0.app are less technical. This even like downloading VS Code, getting your local environment set up. Like I spent this morning with my designer installing Homebrew, like, like it just wasn't on her laptop. And so it's nightmare. And so if you're trying to step from this like vibe coding prototype in web experience into feeling more like a software engineer without having to have Claire handhold you through like brew install. This gets you like halfway there. And so I think there's also this learning aspect of it I wanna make sure people don't miss, but let's get into building something obvious.

Wait, so one, another part of our product development process is really listening to community and listening to customers. So people have been asking for a lot of different tools so that we could guide them towards knowing if a skill is high quality, vetted, verified, because there's so many skills. Last we checked, there were 500 skills being added every hour. And so one of the ideas that we came up with is like, could we add, um, a rating system? So let's add a 5-star based Rating system, board skills, uh, put it on the sidebar. Um, be mindful. So I'll also give you like a little bit of my real-time consciousness on if I were talking to an engineer and say like, what could go wrong if we accept ratings from the internet? One of the things that can go wrong if you accept ratings from the internet is abuse. So let's say, let's tell v0, be mindful that we should rate limit or prevent abuse on the scores that we receive. And again, for me, it's all about thinking from a production readiness point of view when I think about the new v0 and make it make sense within the style of this skills website.

What, what I love about what you're showing us is you have this very, very high sophistication prompt here, which is make it make sense. So we have 3 incomplete sentences on a production app serving thousands, millions of people.

That's right.

So let's fire it off. And you're going to fire it off. And while you do this, one of the things I want to just call out that I think, you know, why this feature is maybe important right now is I don't know if you've heard, there's like this crustacean crawling all over everybody's MacBook Minis. And skills can be a prompt injection vector for things. And so as you're trying to make sure that this becomes the centralized hub for discovering skills, which I think it's starting to be, it is upon you to trying to make sure that the quality's there, at least you have the right thing, um, right things in place so people can make good decisions.

To follow with your analogy, this is a little bit like we're vibe coding on top GitHub.com or npmjs.com. It's like a really, really big deal. All right. So I was gonna walk you through what v0 is doing, which of course, if you've used v0 before, everyone does the whole like talk over the thinking trace because agents are not the fastest. But I do wanna point out a few things that are really important. So, um, v0 is all about leveraging the integration and marketplace capabilities of Vercel. So in this case, it knows what the data source is of this project, right? We're storing data in Redis by Appstash. Obviously it's gonna go through the whole file system. It's gonna try to interpret my requirements. This is already like really nice to see that it's not inventing a new way to store the data. Like it's actually paying attention to the data that I use. Um, and so we'll take a look at, uh, again here, like it actually gave me something that kind of meets my requirements, right? Like it fits within the design style. I can submit a rating. It stores the rating. So I have now my 5-star, 1 rating. I guess I'm gonna—

it's Terminal Core. It's Terminal Core for me. You got your Monospace font.

Um, let's refresh the page to see that persistence actually works. Beautiful. There's a tiny bit of layout shift that triggers my neurosis. So we'll tell it, hey, when we don't have data, make sure there's no layout shift. By the way, for those of you that are like less neurotic, I guess. So it bothered me that when we refresh the page, when we didn't have data, like it jittered the sidebar a little bit. So we're just going to have Guillermo Rauch cook on that.

So while we're, uh, jibber jabbering while the its thinking, which I have to get very good at as a podcaster. So, I will call out that I have observed a Vercel internal hackathon, and I have seen this man screenshot like rounded corners that are not right and just put them in the chat with like a question mark. And so, I, it's, it's, yeah, it speaks to my, um, my very attention to detail heart that you saw, you saw that. Did it work?

Yeah, no, I'm fairly satisfied. Like the, the skeleton was stable.

Yeah.

Zero layer shift. So let's continue with the— we talked about this hardcore engineering workflow. Like if we were making a change like this on skills.sh, again, receiving hundreds of skills per hour with lots of visitors, we first want to make sure that things work, right? And right now you can think of this as a very capable dev environment. We're booting up the Next.js dev server in a virtual machine. It's, it's, it's basically very true to the actual end results. In fact, thank you to the Next.js engineers who sweated all of the details of mirroring to the best of their ability, the dev environment and the production environment. But there is another layer of assurance that we can get, right? Which is, so if you're more familiar with the, like the GitHub world, the GitHub side of things, you know that when you push a new PR, to GitHub, the, this beautiful Vercel bot comes to sort of save the day, right? You know that it builds what you, what you're changing and then it previews it. Not only that, but notice that v0, really cool, 'cause v0's making me look so good here because I haven't written a PR description in like 25 years, 84 years. So v0, uh, produced a PR, described it., and then the magic of Vercel is coming in, right? So it's giving me that, uh, preview. So I'm gonna, I'm gonna open it here. I'm gonna say visit preview.

I, I'm just, again, I'm gonna be a software engineer for a second. Can we appreciate how quickly that preview branch deployed?

Well, don't trigger me cuz it can be 10 times quicker, but yeah, uh, I'm proud of it.

Uh, question mark, explain.

Now I have a production-like environment. So when you see this URL ending on .vercel.sh. .versel.sh is our enterprise Vercel environment. That's why I had the 17 steps of logging in. But this is basically running on the production-grade CDN, on the production-grade rendering infrastructure, hosting infrastructure, et cetera. So now when I'm seeing that, you know, rating there, I have pretty, pretty good confidence. I was like, yeah, this is shippable.

Okay, so I have to ask a couple questions about the, you know, inside the house view of this. Is this how you all are shipping code or is this a big chunk of how you're shipping code to this? Is it 100%? How are you actually using this for production?

So it's really interesting. We, when we cook on a, on a project or a product internally, we hold ourselves to the same sort of high bar as if you had launched a product externally, cuz you wanna make sure that people are actually adopting it, right? And so before, before we started chatting, and I'm gonna give you a glimpse of, again, the behind the scenes of Vercel, uh, We've talked a lot publicly about our data analyst agent D0. Yes, we're very creative with names. We take the first initial and we add a zero. So this is our data AI-powered assistant. And I was actually asking it, I said, D0, this is me by the way, tell me about PR's merge with vZero in the recent weeks. Tell me about its growth. So again, PRs merged with DZero is a totally novel thing, and DZero cooked. Thank you, DZero. It said PRs merged via DZero have seen explosive growth in the last week. Wow. Explosive growth.

I have to appreciate whoever prompt managed this one because it did not put the explosion emoji in there, which I think would also trigger G. All right.

We should, uh, yeah. Uh, trigger warning. Starting from near zero in early January, the feature hit 3,200 PRs merged per day by January 28th, 29th, which is basically today, an extraordinary 100x. The bots, the AIs do like to like, yeah, like sweet talk us, but it's pretty amazing. So this is in very, very, very early preview, right? Like we're just letting people in, but I mean, this is just such a beautiful workflow. I mean, imagine, triggering a task like this from your phone, from Slack, from v0.app. Another convenience that we're adding is that you can take a, a, um, GitHub repo like this and I can go to the homepage and then I can paste it. And so now I could import something that I already have and create a chat from it. So anytime I have an idea for a real-world project, and product that Vercel has in production, I can now prompt. So I estimate that this is going to change fundamentally how we work, right? It's also very visual in nature, which is really cool. Obviously, there's a lot of ways of like getting preview, getting changes made by agents today out there in the world. Everyone's very excited about what AI can do, but this is actually showing me the actual results and like things that are gonna happen. So I, I grade this really high for the kind of products that we build at Vercel, and, uh, I expect this to continue to have a lot of traction.

So, I have to ask you, sort of operationally, how do you imagine companies do this? And one of the things I'm thinking is I was chatting with Caroline, who interrupted you all and say we're, said we're gonna start the podcast, and she said last night I was prepping for this demo and I V0 coded something and somebody saw it and was like, well, that's a good idea. You should just merge it and ship it. Like, do you imagine, or inside, inside the company, Who's shipping code? How are you enabling that as a CEO? How does the culture support it? How does it not?

Until now, everyone could cook, right? Everyone could create a prototype, a new design, a suggestion. In fact, uh, moment of vulnerability, 'cause I haven't really even opened this in a while, but like, uh, let's see if I have, I've probably created a bunch of things that I've been suggesting to the teams that we could look at, right? Um, Ignore this one for a second, but, um, so anytime I have an idea on how to improve the product, I nowadays create a V0. Now the difference is that until I had this mechanism to hand it off as a pull request to the engineering team, then I was kind of like playing in la-la land. I was like out there in this like prototype world. And now we have a common foundation and a common substrate so that if you have an idea, whether you work in marketing, marketers always want to change the website. Like, imagine, like, go to vercel.com. I'll show you a page that is actually quite fun at Vercel. So our enterprise page, every marketer at Vercel wants to change this page at some point, right? And the old way was one of two ways. One is what I called, you had to petition to the government. You had to go to engineers and say, engineers, please, Hey, can you, can you please, uh, add a logo over here or whatnot? Or pray that the CMS was perfectly wired up for any ambition or dream or idea you had. So now they can just open this page in v0 and prompt anything that they want, but it would be somewhat irresponsible to just ship it, right? So with a Git workflow and opening up PR, and being able to preview it, we can all build confidence that it's gonna be a good change, roll it back if needed. And, and again, this is a website that's, that's pretty, pretty large.

What I think is fun about this from an org perspection— perspective is it reduces the friction of getting something live really, really low, right? And like the humiliation ritual of prioritization goes away and you can actually focus your time on defending the merits of an idea on the actual idea as opposed to the hypothesis of the idea that then has to be implemented. And so I think it changes the speed of companies in, in a really significant way.

And you worked, I mean, to say the least, at LaunchDarkly.

Yeah.

And you know that a true production-grade release process involves things like feature flags and experiments and things need to be measured. And there is events that are critical to report from these product surfaces. And so this is also where I see the skills that we can add to v0 and that you can contribute yourself to play a very important role because sometimes, you know, like we're all operating on this like websites and pixels and whatnot. And we say like, ah, it seems easy. How risky could it be to move this button 20 pixels to the right? And so I think we can make vibe coding scale to that kind of rigor that exists within enterprises and companies at scale.

Well, and if we're being honest on that enterprise, it's not gonna be moving a button 2 pixels, it's going to be switching the emphasis. I know this because I spent my life in enterprise, switching the emphasis between contact sales and view the product. There's going to be a perpetual debate which one's the primary call to action, which is the secondary. Okay. I, we, so we've shown, uh, Newbie Zero import GitHub, which I think is really great. Do a pull request, copy and paste your GitHub URL in to import it, actually push to production, make friends with your engineers. 3-sentence prompting, no more than that. I wanna go to a quick, cuz we wanna keep this tight. I wanna go to a quick lightning round with you and ask you a couple different AI questions. So, what is your favorite non— I mean, everything's a coding use case, I'm sure with you, but what is your favorite non-coding use case of AI?

Well, I'm conflicted. My mind immediately went to image generation because I use, uh, so we built up Banger Playgrounds. I don't wanna share screen again.

I wanna take it from you, but this is, oh no, I was just gonna say image generation's how I got this pretty, pretty background.

Um, b0-nanobanana-pro.vercel.app. So what's really cool that's happening at Vercel today is that we're building so many of our own internal tools and agents. And so we're building our own design tools, like for example, to create new images, we created a playground for Nanobanana and I use that a lot. So I use it to make memes, guilty as charged, but I also use it when I wanna present information in really cool ways. When I tweet, when I sometimes have to make, present my vision in a way that's more like on the image side of things, I combine it a lot with v0 because Nanobananana is really good at like, again, letting me fire off 20 generations in parallel and then pick the one that I actually like and then I toss it into v0., and then I actually get more fidelity of what I want to implement. So, image generation is a big one, but also I'm very excited about video generation. We're going to be dropping something. I don't want to spoil it too much. We're going to be dropping something on the video side as well. And, but yeah, all AI is, I also kick off a lot of research tasks, like long horizon research tasks.

Yeah. So, one of the things I want to call out from a Nano Banana perspective, our podcast, is I've used Nano Banana to turn every conversation. I think we're now at 180. 18 workflows. Every AI workflow that we talk about on the podcast gets its own pretty consistent nano banana infographic. And so I just think there's just such undervalued use cases in both, uh, image and video gen. And I'm about to take this and turn them all into little mini videos.

We also created a really awesome— I don't know if we've written about it yet, we're gonna publish on the blog post. We have an, uh, OG image, like, uh, uh, Open Graph card, Twitter card generator that we use internally. That combines more traditional rendering techniques, but also image generation. So a bottleneck in our team was sometimes literally like, can we get that social card to get the announcement through the door? And so now we've also sort of automated and agentified that. Every day we're basically asking ourselves, how can we build an agent that takes over a task that we were previously giving to a person? And typically the person that was working on that task is now the one creating the agent. Yeah. So, uh, something we said at the beginning is we want v0 to be really awesome for y'all to create agents, not just traditional web applications. So that's, that's basically the, the, the runup of the product.

So I have to say, uh, my favorite, so that's your favorite use case. My favorite use case of your use of v0, which I don't know if you're, you're prepared for me to have this much knowledge about what you vibe code, is your agent-to-agent chess game. My kids are obsessed with this. They think it's because they got the AI chessboard and now they're like playing chess at various levels.

By the way, I learned a lot. So, so it was really cool. Like during the holidays, I, uh, I'm a, I'm kind of a visual person. So the chess AI thing has been done before, but I imagined this sort of, um, imagine like ESPN is, is, is broadcasting a final of a chess match and they're going over the shoulder of each player and showing the, the chessboard obviously in 3D. And so I figured, could v0 generate 3D code, right? Could it render with Three.js and things like that, a live chess match? And could I have two AIs battle it out? And so, you could open it, so v0-chess-match.russell.app. And the other thing, The other thing, I wonder if I am SEO'd or not.

You are, you are, no, you are number 1.

Good job.

Oh, yes. Oh, a little Terminal Core over here.

So Terminal Core, of course. But, um, so what I did is I, I, I started streaming the thinking tokens of the models. Uh, apologies to Google in advance. Uh, we're using a very old model of theirs that is really cheap, but they're losing 2,205. At least they got 5 in. So you can see the thinking tokens of the models. This is combining all of the Vercel AI infrastructure. It's using a workflow so the game could run forever. The game could literally run forever or until I run out of tokens.

AI Gateway.

AI Gateway, of course, because we can change models. Like we could see like Grok versus, you know, whatever, what dropped this week, Gwen 3 Max thinking. Uh, but also I learned chess incidentally because not from these guys, these guys are kind of dumb, but, uh, you can see how it thinks through what piece to move and then it says, oh no, because if I move it there, I'm gonna get effed. So I need to do this. So this is a fun video created during the holidays.

Yeah. So I need a kidcore version of this. Um, I also wanna see how much these models are spending to beat and lose.

Well, that's the beautiful thing about the AI Gateway is that we kind of report, you know, for this prompt, how much did it cost you with this model, et cetera. Um, but see, so this is GPT open source, which is actually pretty decent and, uh, pretty cheap.

So. Great. And then, so on this talk, so that's my kids' favorite, your v0. I don't know if they have a favorite, my v0.

It's popular with kids. I also got, uh, another parent reach out to me and say like, oh, this really inspired us and we're gonna use v0 together to create more things.

My kids, uh, uh, D.A.U. on v0. Don't worry about it. So this is my second question, which is, What is the last thing you built with your kids that was really fun?

So the other day I brought them all to the office and what we started vibe coding is, so now we wanna do things that are more like physical AI, like bring AI to the real world. I think it's the next frontier. And so at the office we have a, a thing called Vestaboard, which is a board where you can basically render things. Yeah. In the real world. And I think I really broke their, their brains that day. Not all of them, cuz I, I brought 4 of them and one was on his iPad, not paying attention. It was almost like bringing them to Sunday school for what it's worth. Uh, but 2 of them were like, holy crap, you can type in code and then you can change the real world. So I kind of taught them the concept of an API. Uh, and yeah, all vibecoded. And, um, I, I took one of the nannies too, and she was mind blown.

Well, that is—

I'm kind of teaching everyone how to vibecode.

That is, uh, you and I, we're like twin stars here because while you have the Vesta board, which is pretty big. I have this like tiny, like 32 by 32 pixel fake little mini computer that my kids and I are like vibe coding little screens on and little games on. So I completely agree. Take it off the screen for many reasons. If you have kids, put it in the real world and you can do some fun stuff and blow their mind.

Sending a packet and responding to it. And the beautiful thing about v0 and things like this is like you're literally speaking English to the computer. So I think. If you can teach them that they can express their thoughts and desires, then they can make anything happen.

Okay. Important question. Are you teaching your kids to type?

That's a tough one because I was, when I grew up in Argentina, my dad got me this soccer game. He tricked me. I thought, oh, he's getting me FIFA or something cool. No, he got me a soccer typing game. So to score, I had to type really fast and that's how I learned to type really fast. But nowadays, like they're kind of getting really into the speech-to-text thing. I need to find that hack where like, oh, I got you Roblox. And it's not, it's just typing.

It's just typing. So I, uh, I, I'm tricking my youngest right now. We have a Switch and I really wanna play Ocarina of Time for people that were born when that game came out a million years ago. And the only reason I let him play it is it's like 99% just reading.

Nice.

And so I'm like, you can play this very slow-paced game. It's a game. It's really just going NPC to NPC and reading two sentences, phrases.

I really like that. So. We played a game like this with my kids during the holiday break. Um, it was basically like a puzzle math game that looked like a game. And that inspired— so, okay, good and bad. The good was the game was like really educational. The bad was like, it was like ad-ridden brain rot slop. And so I was like, okay, huge opportunity for someone to create a game platform. You can combine all of these models, you can do image generation for the assets. We're about to drop something insanely cool, text to SVG. There's models that now produce really, really high quality. We're gonna share Recraft models through the Vercel AI Gateway so you can create assets that are beautiful, game ready, scalable in high DPI screens, whatever, iPad. And so I really, I see the future of really high quality content. At our fingertips and getting rid of all this slop.

Yeah, well, I, you know, my, my feature roadmap releases are not as confidential as yours. And so I am in the back of my mind working on like Mama Claire's dojo for Crack Little Hackers. Nice. So maybe someday, and deployed on obviously Vercel. Okay, last question. Do you, when you're frustrated, prompt AI the same way I have seen you prompt in Zoom chat, which is explain. How do you, how do you, how do you, yeah, question mark. Like, what do you do when it's not giving you what you want?

So I do think that what we were dropping now is going to help you so much for the moments where, I mean, let's be real, you can get stuck with AI, right? But now that you can essentially have this full, like, let's call it escape hatch, right? Like you can, if you want, you can clone the repo and keep cooking on your local machine, or if you need some, someone else to help you, this is fundamentally a collaborative medium. That's the thing that GitHub. Unlock for the world, collaboration between engineers, designers, marketers. And so, um, I, I foresee that a lot less people are gonna get stuck, frankly. The models also keep getting smarter. Skills is gonna help you a lot as well. So I'll give you an example. We are always adding new frameworks and new capabilities, and Next.js is getting more powerful, and the AI SDK. Now that we have skills for those that we're gonna preload into v0, the model itself is not going to get stuck as easily because now it has more resources to figure out how to solve that problem. Another thing that I've done, it actually used it for this project. So there's a lot of subtleties about this 3D thing that I didn't know anything about 3D. So whenever I would get stuck, I would ask other models. I think it was something about the way that— and kudos to the awesome soul, the the gentleman that open-sourced the 3D. So I got it from Sketchfab, and he didn't design this for creating a game or anything like that. So all of the pieces were stuck together. It was almost like you had 3D printed it and the pieces were stuck together with the board. And so I, obviously I want that sweet animation that when the model decides, it moves the piece, right? And so I had to ask a lot of questions to other models. About, hey, teach me what's going on with this 3D thing. How do I, how do I reason about it? And then I would copy what another model tells me and I would toss it into v0. So, uh, another thing that you can do when you get stuck is you, you, she, she can't see it here because it's only for me, but there's a debug button. And what the debug button says, I asked v0, hey, give me debugging tools so that I can visualize the mesh of the 3D model. I can visualize, I can turn the textures on and off. And so the, the AI itself can help itself and can give you tools to debug problems, which is kind of meta, but try it, try it, and it's gonna work well for you.

Okay. So you, you ask an expert, which is another model. That's right. And you find out the right question to ask. Well, this has been very fun. Thank you for showing us a little bit behind the curtain of how you use v0, all the new stuff. Some of the ways you use it in your personal life and fun projects. We love to see it. So, you have this new v0, you have this room full of people. What do you want from them? What, what can they do for you?

I mean, get busy shipping. So, try it out, uh, give us feedback. Uh, we will fix it very, very fast because we're gonna be dog eating our own dog food. Uh, and yeah, share the things that you built, uh, with v0.

Oh, we have a question back here. Oh, I love it.

So I had a question. Thank you for sharing like the feedback process, how you loop. In terms of like from ideation to like production, like during that process, do you do any like product market fit? How do you validate that what you're trying to build is actually like useful or impactful for like your ecosystem?

Yeah, super hot off the presses, new sort of mental model that we've been using internally. There's a customer zero and a customer one. Customer zero, we like to be ourselves. We, uh, it's like the Rick Rubin, the confidence in our ability to know what's good, whatever. Like, we like our taste. We've been around the block for a while. We want to, we have ideas of products that we would like to see out there in the universe. Uh, but customer one is also really important. Like, a close design partner, Claire Vo. Claire V0. And our, uh, Claire and our CPO are constantly texting.

We're on a text chain, yeah.

She texts bug reports, she texts things she needs. So having that group of design partners, enterprise companies, individuals, community members, people that slide into my XDMs. So it's, you always wanted that pressure test of the world. And for skills, it was that, you know, people telling us, hey, like, why does Opus 4.5 kind of know the latest Next.js, but not really? And how can we embed your best practices? So that was kind of like in our backlog, like we were thinking about that problem for a while, but then it also became really concrete. We were like, okay, how would how do we go about distributing and discovering these skills? And so, sort of the idea became very concrete, and that's where a tool like v0 really helps you because when you wanna extract out of your mind, you can just v0 it. And so, pretty much what you see today, it started with like 4 or 5 v0 prompts in a conversation with our VP of design who then took it way further and made it actually good, and our CTO and our product leader. So you mentioned about VMs. Is there a possibility of having React Native in V0? So what you saw today is if you peel the covers of how V0 is built, it's built on a bunch of Vercel infrastructure that you also all have access to. The virtual machine that we use to run that Next.js preview is called Sandbox. So the Vercel Sandbox. It is a very powerful computer. In fact, what's actually making agents really capable is not that they're perfect, is that they have a computer at their disposal in order to solve every— any problem. So the reason we showed you a glimpse of how we work internally with D0, the reason D0 is so good at data analysis is that it has a computer where it can do research, it can write Python code, it can run it, it can make a lookup to Snowflake, it can search the web, it can come back, it can fix it. It's like a 4 or 5 minute process, right? And so these computers are very powerful general purpose computers. You could imagine them running React Native. You could imagine them writing other programming languages. They can obviously already write Python and run it. And so the sky's the limit from that perspective.

I have a question on that topic. Is there going to be a point at which v0 is going to help you build those agents?

Oh, absolutely. So the main, the main idea of this chess thing, it's cute. But I mentioned the word workflow. So it's actually tricky for me to say this program is going to run forever in the presence of network or compute failures. So in fact, like, uh, you know, we're doing this live and just randomly came up with the idea. The reason I had confidence that demo was going to work is that if an LLM provider is down or a function call dies or times out or whatever, The workflow engine of Vercel will say, we're doing live, we're gonna try it again, and we're gonna try it again, and we're gonna try it again. What we're gonna help you do is create those kind of workflows from within v0. A lot of agents need that kind of reliability. For example, you send the message from Slack and you say, @playergpt, go and spec out this PRD for me.

Yeah, it's called @chatprd. It is running on workflows and Yes.

We didn't rehearse this.

Yeah. So we, we do have, we do have, uh, I, I feel the pain on, on agents where, um, workflows are really helpful because they give you durable overtime execution. You can retry things, things can fail, you can do them sequentially, you can do all sorts of things. And so I do think the ability for you to build that on the backend is very helpful for even for applications that don't have the same kind of UI that you would have it.

Totally. Because some are, some are very visual. And some are gonna be more headless, like some are gonna be background agents that are doing work for you. In fact, the world is excited about the artist formerly known as Claude Bot. Yeah. Mold Bot. I don't know what Mold Bot means, by the way.

Well, uh, crustaceans, this is how you know you have kids in like elementary schools. Crustaceans like lose their shell, so they molt.

Molt.

All right. All right. It's like a snake. Yeah.

Got it.

And then emerge to steal your bank account.

But so that's a beautiful example of the background thing, right? Because you text it and it does a bunch of stuff for you and then it responds. So that, that's the kind of thing that we want you to be able to build with v0 where it's gonna use workflows, it might use sandbox, and it might be more— I think MoldBot is like super general. It can do anything and even hack your computer. But, uh, I, I expect a lot of really cool agents to, to work that way. Like you just WhatsApp them. Yeah. And we'll, we'll help you build those with v0.

And the last thing I actually deployed on Vercel was my live Multibot conversation. So if you want to see what it is, terminal code has a Multibot. I, well, I, I shut the computer and walked away. I did not enjoy my, uh, it was an interesting experience. You can smash that subscribe button if you want to watch it, but I did, I did deploy to Vercel a terminal view of my conversation with Multibot. So you can go, it's like, it's clawed.

Multibot confession.

Yeah. Oh, that's exactly what it is. Yeah. It's real scary. Great. Maybe one more question.

Oh, wow. What an honor. First of all, thank you so much for hosting the event. Thank you so much for the new v0. I ship a lot with v0. My co-founder is using Cursor, and now we have like a mutual compatibility.

We're so back.

Love it. There you go. And so I wanted to, again, I'm also a v0 ambassador. There we go. Let's go. I wanted to ask two questions. Number one, a little bit different, but when do you think— so v0 has really democratized the ability to go from prompt to UI, UI that works and UI that satisfies like the prompt that the user created. When do you think that paradigm is going to happen within consumer in Gen UI? I see that your team has been playing with a bunch of different libraries, different ways to create that. When do you think that switch is happening? And then the second question, When can I write a prompt and then deploy to the App Store?

Yeah, so on the, on the first question, sometimes you're going to want like a flash v0. So we're playing with ideas around generative UI where an agent might decide to render something and think of it as like spontaneously creating the UI code and then rendering it right away. You're not actually creating code and an application and deploying it. You just want to make it happen. So we're doing a lot of research on like, what would that look like? So it's another tool that we can give agents. On the App Store question, you've seen some of the stuff that we've been doing with v0 iOS app, React Native Skills. A long-held dream of ours has been to really democratize pushing to the App Store like you would push to the web. And everything here is like sort of using that same, those same ingredients, React-based, best deployment platform, just deploy with one press of a button and whatnot. So, uh, yeah, don't wanna promise any timelines, but, uh, definitely something we wanna do.

Thank you. Thanks everybody. Thanks so much for watching. If you enjoyed this show, please like and subscribe here on YouTube, or even better, leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review, which will help others find the show. You can see all our episodes and learn more about the show at howiaipod.com. See you next time.