Everyone’s an engineer now: Inside v0’s mission to create a hundred million builders | Guillermo Rauch (founder & CEO of Vercel, creators of v0 and Next.js)

27:58And I think that idea of sharing things and putting yourself out there and broadcasting to the world. So another thing that I do is I take my kids to hackathons, which is went to an awesome hackathon at University of San Francisco, USF. It was called the Bloom hackathon. It took two of my kids. And I wanted to watch how people presented their ideas. And we had a lot of fun. We also ate waffles and grilled sandwiches, which is a bonus. But so presenting and putting yourself out there. I mentioned in the beginning of the podcast when we were chatting, I learned so much from you and your guests because you put out all this awesome little post on X in these videos and this snippets of your interviews.

28:45And so the ability to present what you've built and put yourself out there. An incredibly important skill in the future. Especially in the world where the marginal cost of producing software and new things are going down. You need to build an audience. You need to know how to talk to people. You need to build your own signature brand and style. And so maybe they're a little too young for that one, but yes, taking them into hackathons. Probably like, you know, back is influencing their neural networks or pre -training data for the future. I love it. They're going to tell their friends, and my dad took me to hackathons.

29:23What's that? So are you actually, are you encouraging to learn to code? Because it's interesting. You mentioned math, eloquence, presenting, and then, okay, it's also learning to do that. Yeah, I think again, learning how to prompt, learning how to code. With V0, we show you the code when we build things. So if you can build that mapping of like, maybe not learning how to code in the necessarily in like, as an abstraction. If you do have a knack for it, kind of like, you know, I'm a big believer also that my five kids have like super diverse personalities and inclinations. And I don't want to be, you know, pushing for something that they wouldn't want to do or whatever.

30:06And so learning to code in the abstract might be good for some people. Might be good. Might not be the fun thing to do for other people. And so what I would recommend is try to understand how things work. So if you prompt V0 or any other tool, and you generate some code, try to build an understanding of what that does at a high level. It's like actually maybe an extension even of eloquence. One of the bets that I made early on with Versel that really paid off is, Versel, maybe as a metaphor, is like AWS in easy mode for a lot of people. We have a very large user base of people that would have otherwise not have been able to configure all of the in, in -senouts of the cloud.

30:52But do want the scale, flexibility, speed, et cetera. They want to create a very high quality products and services. So I like to give this a good example because one of our customers ramped at a 43X increase in traffic when they're, when they're at went life. The engineer that worked on that only needed to learn next .js. Then they pushed their code Versel and now they can reach an audience of 100 million people without a blip, 100 % uptime. That superpower comes from, we made it as easy as possible to get started. And the language that we choose is actually very relevant in this story. JavaScript is almost, in my mind, there's always been almost like the English of programming languages.

31:41It's a language that if you learn it, you reach billions of devices. So it's not a coincidence that when you ask chat GPT or an anthropic or Gemini to build you a web app, it uses this tools. It uses JavaScript. It uses React. It's become the lingua franca of building products on the web. So I would tell to make, I would say to make it look, if you do want to go deeper into programming, start learning there, you can reach huge numbers of people. If you have a passion, I would say there's going to be a fundamental engineering skill that's going to be useful for decades or centuries to come, which is creating foundational infrastructure.

32:26Think about LLAMS in terms of their like oracles that can go and write software for you, but there's a limit to how much software they can write. There's context windows. There is time and computational constraints. So it's very hard for an agent today to go and say, I'm going to write a cloud from scratch. I'm going to write all the foundational services. I'm going to write the framework from scratch. I'm going to write the compiler. No, the LLAMS orchestrating those tools and infrastructure. It's not writing the compiler from scratch, otherwise you get into the Newton thing in order to create an Apple, you have to create the entire underlying universe.

33:08No, the LLAMS are interoperating with the universe as it exists. And so the engineers that learn foundational infrastructure are probably going to be extremely empowered still for years to come. Like there's a world where you could argue, chatGPT will build the next version of chatGPT. What I'm hearing from you is that's a long ways away, if ever. Absolutely. This is what, you know, the common, the running joke is that all of these companies have, you go to their career spasions like engineers. Engineers. Engineers, right? The counterpoint of that is that at a per cell, we had, we have 150 engineers that can write code.

33:50And 600 total headcount. Now we have 600 engineers. One of the best things that I, some of the best things that I've seen created with V0 have not come from our Azure team. They've come from the marketing team. They've come from the sales team. They've come from the product management team. The product management team is fascinating, right? Because now they're actually building the product. So last night I saw how we've specced out in V0, think of it as like a live PRD. We've specced out how the new functionality for deploying a V0 to a per cell is going to work. The amount of detail that was contained in that V0, I mean, we're all just saying, well, just ship it.

34:34There's nothing else with this guys, right? Like it was animated, it was interactive. We were demonstrated the error state, the success state, the slow stream state. So it really empowers product builders, not only with technical skills, I think that does a disservice to the tool. It empowers them to explore and augment their thinking with a lot of things that perhaps it wouldn't have considered otherwise, a lot of states of the product that wouldn't have considered otherwise. The name V0 implies the product is for prototypes, for the kind of like the first attempt at stuff. And that's definitely where all these tools are finding product market fit.

35:15Prototypes, PMs showing a thing that working versus just design. Do you expect V0 and other tools to get to a place where you can build salesforce .com and scale it to billions of dollars. Like you do. We already have a customer, an enterprise customer of V0 that only works with V0. All their products, all their features, all their client communications are V0 native. Two days ago, I just heard anecdotally an ex, someone tells me my brother just sold his first website to a client completely built in V0. Yesterday, an investor conference, an investor walks up to me and says, two of my friends just got engaged on V0.

35:58Okay, okay. V0 is dating out now. So the engagement website, the proposal, the wedding, like it's all V0 native. So because we've integrated V0, the versatile infrastructure, we can do that whole story that I just told you of, like I have a website to build and it can get it in front of 100 million people. We can enable that for everybody now. And so the end -to -end, full stack, V0 native and build on this awesome fluid, serverless infrastructure that scales to billions of people, all just from prompts or screenshots or just copying and pasting your PR days into the tool. So let's help people be successful with V0 and then let's also do a demo, but before we get there, let me ask you this.

36:47So imagine you could magically sit next to someone who's about to use V0 for the first time and whisper a tip in their ear to be successful with V0. What would your, what would a couple tips be? So number one is you can be as ambitious as you want in terms of what do you ask the tool. If you can steer the tool towards some kind of inspiration that you have, you're always gonna get better results. If you don't have ideas on what to build or what to prompt, I would recommend using the V0 community so that you can find something to fork to get started. I would say in some ways, if you have technical skills, this one is interesting, but have some suspension of this belief, like what I, like it humbled me, I was saying about like accessibility.

37:40So the open mind that about whether the tool actually knows some things that you might not know. And so focus more on the product description, right? Like focus more on like what do you want the end user to experience? What do you want the product to do and try to be open minded about how well the tool can implement it? Those would be my main ones. You also have to have a sense of iteration, I guess. Think of it this way. If you were working with a design firm or an agency that you've hired, you will go back and forth and say try something else. If you were coaching an engineer that's getting a stuck in something, you would say try something else.

38:28It's amazing how many times I've gotten unstuck in this hero by just saying like, just try something else. Just saying that as the prompt. Not even giving it away. Just saying that, like I mean a chat is like, we need to have, it's like, yeah, you know, like you have a one -on -one performance review with a tool, hey, wait a talk, try something else. What you're doing so far is not working. It's amazing. Like one fitness function that I'm keeping in my head is I really want to find the thing that it cannot build with Vizier. So I, as part of the Vizier community, I have my own profile. We'll share the link with people.

39:06You can see six or seven things that I've built that I consider to be pretty impressive. So for example, I was flying from Tokyo to San Francisco. The internet was horrible. What I like to do during flights is I like to monitor our own flight on the flight. So I opened flight radar or whatever. And I was extremely bored as well. And I noticed that flight radar, I don't know which one it was. The flight radar is like a four or five of them. They were very bloated. They had ads. They were not what I wanted the flight radar to look like. So I built my own during the flight with the worst internet connection that you could imagine in the world, integrated into a flight data API called Edge Aviation.

39:55So I, what it is what I told Vizier, we're gonna build the best flight radar on the planet. I didn't, I wasn't prescriptive about how. So it used a tool called MapBox and a JavaScript library called Leaflet. I didn't tell them that or her, and Vizier what it is. And subsequently, once we cooked on the design, which looks, I would say beautiful, I then got more ambitious and I said, all right, let's make it real now. And by the way, that's actually how I would work. So it's how I like to work. I like to work experience first and thus also how Vercel was built. Let's start with the front end. Let's start with the planes on the screen.

40:43And by the way, there's a lot of subtleties here. For example, there's so many flights going on at any given time that there's just too many. So I had to work with Vizier on improving performance. And once again, I wasn't prescriptive. I just said, we have a lot of flights, chief. Let's, did you say chief? I actually do say that a lot. And this is I think when I shared it on X, it blew a lot of engineers minds because it created a canvas based. Canvas is the sort of underlying rendering surface that very sophisticated products use like Figma. And it created this awesome overlay on top of the map that can render tens of thousands of flights at any given time.

41:29And then I told it, let's make it a full stack application. I, okay, plug it into the the flights API. So that's an example of like we cooked and there was no limit. And so I'm always in the lookout, the service that I'm providing to the Vizier or coffee is, I'm part of the team that it's really trying to break this and say like, can it not build something? Or, and even when it does build it, we're very obsessed with quality and performance. Like it has to be real. That's the, that's our commitment to our users. And how much of this cost? How much time does it take to make something like this?

42:05So the, the flight rateer example or Vizier? Yeah, the flight rateer example, specifically, just like, I mean, that one probably took less than two hours with the worst internet, sorry, in our lens, I love you, but like, you give me a hard time. And, and what did that cost like, like 10 bucks? Like, what would you estimate? I mean, I pay for the $20 or these are just 20 bucks. So, okay, for a month. So it's like a month, but you use it for two hours, 20 bucks. Yeah. If you had engineers building this, how much do you think that would cost? How long do you think that would take? I mean, weeks easily, easily.

42:40And that's like $10 ,000. Maybe the most cracked engineer out of our cell couldn't knock it out. And like, without using any AI, could knock it out in a couple days. But then what about the design? What about like me? Because I'm the bottleneck, you know, the engineer. And this was amazing about this collaboration, right? It's like, I'm providing the product guidance. I'm saying draw a dashed line between the, by the way, this is a, these are just blew my mind so hard. I said draw a dashed line between the two destination airports. And these are said, well, I have to account for the spherical, or what is it?

43:21It's a pseudosphere for the curvature of the earth. It's like, okay, these zero super genius, like whatever. And so that's what I mentioned about like how you can kind of go back and forth. It's like a product copilot is like an all -knowing being. One of our users yesterday submitted feedback to the tool. And it was positive feedback. We were very happy with those. They were saying like, BZ is like a super genius, five year old PhD with ADHD. So like, you still have to like, I'm not gonna, oversell this is like, you know, it knows everything about everything is like, it's everything perfect, of course.

44:03But it has this sparks of brilliance. Really, truly, I think I've been a big believer that AGI undersells what we're collectively building because we already have all of this is sparks of super intelligence. I don't believe that these zero is an AGI. If it knows everything about how to draw a dashed line according to the curvature of the earth and this high performance map of airplanes, like that's just super human. And it's, yeah, it's a joy to use. Today's episode is brought to you by LinkedIn ads. One of the hardest and also most important parts of B2B marketing is reaching the right people.

44:49I am constantly getting ads for products that I will never buy. And I almost feel sorry for the money that these companies are spending, pitching me on their spend management software or some kind of cybersecurity solution that my one man business just does not need. When you're ready to reach the right professionals, use LinkedIn ads. LinkedIn has grown to a network of over one billion professionals, including 130 million decision makers. And that's where it stands apart from other ad platforms. You can target your ad buyers by job title, industry, company, role, seniority, skills, even company revenue, all the professionals that you need to reach in one place.

45:27Stop wasting budget on the wrong audience and start targeting the right professionals only on LinkedIn ads. LinkedIn will even give you $100 credit on your next campaign so that you can try it for yourself. Just go to LinkedIn .com slash pod Lenny. That's LinkedIn .com slash pod Lenny. Terms in conditions apply only on LinkedIn ads. As you talk, it's interesting the way I'm thinking about this now. There's almost like three core skills in building apps with AI. There's figure out what to build. There's making a low good, like design. And then there's getting it unstuck. Yeah. And it's interesting how like these are going to coach.

46:07Coaching it. Yeah, or just like here's a, oh, here's the here's like the database error. I don't know. It's not figuring it out. On the, I guess, does that resonate? I've never thought of it. Oh, absolutely. In fact, I'll tell you a little bit of a story of something that I, so even going way back in time, next year's builds on React. React was this UI component library that Facebook created. Actually, with very similar goals, they had so many cracked engineers and they had to help them collaborate on an enormous product surface. So they invented or at least pioneered, I would say, the concept of this component as a unit of reusability, as a building block, as a Lego brick of how you build software.

46:48There's no coincidence that Ella Lemms love to work with React components, by the way. And one of the things that always stood out to me about that model is it basically enables people to scale in how they worked together. And one of the key design principles that they embedded into this thing is they called it a scape hatch. The API, when React doesn't perfectly model your problem with its component system, they give you a scape hatch. They say, okay, engineer, you're on your own now. There's no guard rails. And in fact, one of this scape hatches is called dangerously set in our HTML. They wanted the developer to know untardatory, but they did give people the API.

47:45That is a profound systems design engineering principle. And throughout my life, I've always thought about scape hatches. One amazing scape hatch that Bezier has is that you're looking at the code that we're generating with NextShast. You can edit it. You can even have other experts look at it. One thing that one of our demos last night came from this awesome company, Luma Labs. They're creating one of the most amazing video models in the world. And they use V0 and V4L extensively to build their application, their websites, et cetera. And the design engineer was talking about how he was on a V0 that had 120 or so iterations.

48:34So he deep into the latent space. He was in the matrix. And at one point, he got stuck, but you know what he did? He copied and pasted the code that we generated and he gave it to ChattuPT01. And ChattuPT01 thought about the solution. I honestly, I'd never even thought about this myself. I was so blown away. And it doesn't speak to, I love your third point of, you need to learn a skill of how to get unstuck. It's like a profound like life lesson. Like it's just more a generic life advice. You need to get Facebook actually had a principle. Don't get blocked. Like seek to get unblocked, seek help from other people.

49:14What's fascinating is that you can seek help from other AI's to get unstuck. And those escape hatches of like actually understanding and seeing the code underneath. And even being able to say, okay, now let's use Git, let's turn this into a less, more of a hybrid project, not just prompts, but also traditional software engineering. The fact that that door is open to you is extremely valuable. Let's actually make the Superconcrete and show people what this actually looks like in V0. So pull up, I'll share a screen and then we'll do a little live demo. We'll keep it brief. I find people are like, okay, I get it.

49:54But we'll make it fun and brief at the same time. There it is. I see it. Okay, help you ship. Yeah, of course, we're all about shipping. Okay, so as I mentioned, you write in English, you yap into the tool. I'll say for a demo, let's create a contact sales form in the style of, by the way, I had a typo, I don't care. The lives get it. It's self -subreme, the clothing company for an online store. Now, I mentioned that sometimes people get blocked on like there is like a writer paralysis at this step. So we added enhanced prompt. So now like you're tapping into the latent space of the model, which has a random component to it.

50:48And by the way, this is still not a substitute. It doesn't contradict what I said earlier. Obviously knowing the meaningful tokens, knowing what, like the right style is and what it's called and whatever is still highly valuable. So the first thing I'm going to notice is that as the model thinks, you can introspect its thinking. So do we add it this recently? And it's been, most of it actually by the deep seek revolution, I would say. So the fact that when you tell it develop a contact sales form, like what is it going to do? Like do we talk about scape hatches? Okay, it's going to use Shatsy and UI.

51:27It's going to use Tailwind CSS. It's going to use React. And this is your opportunity that if these year is not doing exactly what you wanted, this is your opportunity to actually go and sort of correct or influence or give feedback and so on. So you're going to notice it spits out a bunch of files and it gives me the thing that I wanted, right? So I'm going to assume out a little bit here. So a couple of things that stand out here that, I mean, as an experienced engineer, I can point out, the underlying component system that it uses is the same component system that the best tools on the plan are built with.

52:05This is called Shatsy and if you go to GROK .com today, they're using Shatsy and to build their UI. They're using NextShast. You're getting that caliber of code. The other thing that it did is people on social media talk about this a lot. When you use a global shared component system with the world, you don't want everything to look the same. So the fact that it was able to apply the style and it kind of knew what Supreme looked like was kind of cool. But now I'm going to say actually, because I'm building a financial institution, make it more serious, make it in the style of, let's say Charles Schwab, change type basis.

52:48So this is the iteration process of like, I'm going to go and get feedback to the model, I'm going to make it try different things. So once that initial generation was already created, now the model is actually acting more as an editor. It's going and making tweaks to what's already been built. And this actually scales to very large projects. You could have started with something much bigger. So in the meantime, I'm going to show you what Luma Labs created with B0, which is absolutely phenomenal. I love this last night. It already has 2004. I was telling you about the power of our community. So by the way, you can just click community here on the B0 sidebar.

53:28I'm going to fork it because they generously shared it with the world. Notice all the incredible animations here. By the way, they shipped this to higher and attract talent to their company. I recommend them, by the way. Like you should, if you're going to be a brand designer, take them into consideration. Notice that it's an interactive. Everything is AI generated. The user -owned AI image generation tool to create this beautiful frames. These are all AI generated as well. And it's interactive. So like there is the auto -playing functionality. This is actually a complex layout in animation system that they built entirely in B0.

54:11I was telling you that at one point, they even got some advice from O1. So shout out to OpenAI. I'm going to say, make it, sepia style colors. So this is an example of like, okay, I forked something. I already have a starting point. My bank grade content form is ready in the meantime. Another fun thing to do is, you can start with a screenshot. So I'll use another next .js user as an example, which is fortune .com. Shout out to them. They built a slick website. But let's say that I'm actually wanting to break into the news business. So I'm just going to paste a screenshot. I could have also attached a Figma file.

54:58And I'm going to have, VZero already knows. VZero can answer questions as well about the engineering design product world. So like I can ask VZero, what is a newsletter that I can tell you better, explain with a diagram, use Lenny as an example. So VZero is also a knowledge seeking tool. But we do strongly sort of like, quote unquote steered a tool to create things. So if I paste a screenshot, as you can see, it's cooking on creating a hopefully awesome news website. I specifically ask, because I think it's funny to explain a new sletter with a diagram. VZero can create again, like explainations, content, knowledge.

55:48The creator is Lenny. You were a former Airbnb product lead. I guess I should have used some examples from Airbnb by the way. But let's look at here at what it created with fortune. So notice that the, I'm just noticing now the, the cyber should have been on the center. Let's use the, I'm gonna zoom out a little bit. Let's use the refinement tool to center this. I called this by the way, one of the hardest problems in computer science is actually centering things. Let's see how it says. So that's right. Centering a div. And in fact, look at it, it was a div. So notice that I did a precise sort of inline prompt, in the difference between VZero and a lot of other tools is that, yes, you do have the code and code is very important, but I call it code last rather than code first.

56:47You're living in the product. So is centered that, another website that I love, also built with NextShast is Semaphore. So I really like their sepia style. So I'm gonna say use this, apply this style instead. Include. So you're sharing a screen, so you used a screenshot to design, to build the site and have a different screenshot to tell it will make it look like this. Yes, and so the idea is that VZero can grow a different aspect of what any subild. It can be functional aspects, it can be layout aspects. One of the things that's also very important to know is, we influence the model, so a lot of the things that you would have had to prompt, you might get for free.

57:38One that's important to call out is responsiveness. So as an example, if I notice that if I do this, it's going to like make it work quite well in mobile, is going to give me that hamburger menu. I can now tell it like, apply that style to everything. In the meantime, I'll show you, this is actually to me very, very impressive. And I don't know why today I'm so excited on the theme of Sepia, but notice that not only did it just the background, I hope people get noticed this, it applied it to the check boxes, and it applied the CSS, assuming this is CSS filter. Yeah, it applied a CSS filter, just for the sake of it because I'm a nerd.

58:28I'm going to look at it, but yes, it applied a CSS filter, confession time. I actually didn't know that there was a Sepia function in the filter property of CSS. There were many ways to accomplish this. You could have also read the images or the videos to a canvas and like apply all kinds of algorithms and whatever. I like that it did more elegantly than you would have. Yeah, exactly. So that's why you can't be too opinionated. With the tool. So another cool thing is I do like showing screenshots, but I do want to remind people that the idea is not to clone other people's websites, necessarily, right?

59:09Like it's just a good demo. So simple way to show off what it can do. Exactly, like take screenshots of your own things, right? Take screenshots of your artboards, take screenshots off, things that people post in Slack, and also don't hesitate to add functionality. Incredible. Thank you for doing the demo. I'm just trying to imagine having an engineer I'm working with asking them to do these things and not only just how annoying they would be, like make it seepia, but just how much time it would take from, okay, do this thing. Copyforch .com. It would be like days, weeks. Here it's just like, months.

59:39Months, if ever. Bigger never ships. That's right. Well, something that I noticed that I loved at the beginning when you were doing the prompting and that prompt improvement feature, is it basically is like best practices to make it look good and look better, which I think is one of the more interesting, I don't know, levers to working with AI is it just has best practices to help you build things that are beautiful. And also, it feels like there's this opportunity of just like helping you figure out if what you're building is at all a good idea. Like, what is the problem you're trying to solve?

1:00:14Like it feels like there's a PM1 page or step that should exist. Like, how do you know this is a problem? What are the views there's told you? How many people have told you this? Things like that. Yeah, there's something to be said about the fact that over time we're more and more peaking into the mind of the AI. That in itself is becoming a killer feature. So the deep seek is stream the thinking tokens moment was a very big moment for industry, I think. Because open AI did have the technology, but they decided that for competitive reasons, which, you know, it's a reasonable thing to think and open intended.

1:00:55They were going to withhold it. And also, it wasn't clear that there was going to be product and user and product utility. But when deep seek hit, it was very obvious that people really liked the idea of understanding how the AI thinks and influencing where it should go. We've got an actually amazing feedback and bug reports where people actually specifically point out, look, this is where the AI went wrong. Please fix it. So the more people we get on this product, the more thumbs up, thumbs down, the more user feedback we get. And by the way, I'll tell you like, for people out there building products, my number one guidance or piece of advice I would give to any sort of founder was, create a lot of opportunities for people to give you feedback inside the product.

1:01:45I drew inspiration from Stripe. And this was amazing for the early days of our sell. There was a feedback button with a very slick inline form with four emojis that would allow you to decide how you were feeling about the feature of the product at that very moment. And that would go straight into Slack. And we were building day in and day out, just like streaming user's thoughts right into our consciousness.

1:02:14And maybe we would get, I don't know, like, tens, hundreds a day, especially in the early days, maybe like a couple of day and whatever. When you're building AI products, it's a constant stream of user feedback. So for people that are thinking about not building AI products, you know, it's gonna be hard to compete with something that has such a tight feedback loop with users. It's the whole idea is to capture user's feedback. So the next iteration of the model that prompt the fine tuning, the examples, the rag is better. And one of the things that a person has done as a result of this insight is with open source, a lot of what makes D0 work.

1:03:01So let's say that you wanted to create the D0 for doctors as an example. You can go to versel .com .sash templates and you can clone a chat GPT template that basically follows all of the best practices in the world for like really high performance awesome UIs and now you can go out and build your own AI products. We've also open sourced the AI SDK, which is the foundational plumbing of D0. It allows you to connect any model and generate UI from its responses. Not just output text, but actually generate UI. So maybe because I love showing stuff, I'll just really quickly show you this is a great excited about it.

1:03:46So if you go to chat .versel .ai super quick, you're gonna see this is the open source chat GPT demo that we've built. It can, you can ask questions like old school LLM but also you can ask, let's actually finish this, let's ask what is the weather in San Francisco? We call this, Generative UI. It's responding not with just plain text is creating components as a result. Last but not least, and this is a sort of easier style opportunity. Let's ask you to help me write an essay about Silicon Valley. It's going to create a canvas or artifacts of style experience. And everything is generative, but also users can edit, refine, et cetera, et cetera, et cetera.

1:04:36This actually reminds me of something I've been thinking about. There's all these startups that are building vertical AI tools. This is like a little bit of a tangent and there's always this AI stuff for lawyers, AI stuff for doctors and nurses. And the pitch there is that these are gonna be founders that know a lot about the specific problem in this totally market. And so they'll build like the tools that are very specific to that. Yeah, I'm absolutely convinced that expert AI tools are the future. There's an amazing product being built on Vercel called chatplrd .com. It's the VZero for writing PRDs.

1:05:11And it's gonna get at VZero integration soon. So you can write your PRD with AI and then you can deploy to create it with AI. That's just an example of like a vertical that you can go after. There's also open evidence. It's like the chatgbd for doctors actually. There is an amazing startup building x -ray AI tooling. So the ideas I think are infinite and what I've seen from users of AI at Vercel, like for example, our legal team loves this tool called getgc .ai. They could hear you go to chatgbd to ask legal questions. But someone out there decided I'm gonna build the best legal AI tool in the world.

1:05:54It's gonna be up to date. I'm gonna obsess about this problem. The CEO herself is a lawyer. So it's gonna be hard to compete with that, I think. But here's what I'm thinking. This is like almost the opposite. And I'm curious to get your take. But let's not spend too much time on this because it's a complete tangent. No, no, I love it. So you showed me this, the way they're widget that you just built. Basically, it's like a little mini app that they I built as you're talking to it. Is there a world where when AI, when AGI is far enough and approaching super intelligence, can it just build you a hard, for example, in real time?

1:06:28Here's the best experience for a lawyer. Here we go, we got it for you. Totally, totally. I believe that eventually yes, but humans will always want to have some guardrails. The reality is that getgc is taking a double job. One is making the best tools for lawyers possible. But also putting their weight behind it, putting it saying like, we've actually used this and we believe that this is what the future should look like. There's a sense of direction and opinion about things. And I think left to its own devices, AI, I don't know, this is the double edge sort of like prompt and bellishment. Like AI doesn't always know exactly what we want or what we need.

1:07:13It's still very much a copilot, a partner, an assistant is not really running our lives. And I don't know that it would even want that ultimately. Okay, I'm going to go in a whole different direction, which is taste. We hear this word taste all the time. It feels like the thing that people are always suggesting this will continue to be an important skill to know what is good basically, to know what people are likely to find valuable and good. And I know, clearly you have great taste. You're building incredibly beautiful products. These heroes clearly, it's like the most beautiful by default builder out there as we've seen.

1:07:49So in terms of building taste, people are always like, how the hell do I do that? I have great taste. I know I do. I don't need to. How do you build taste? How do you think that is? How do you think you build taste and any advice for folks that are trying to improve their taste? Yes, I think it's extremely important to try lots of products. You need to get yourself out there. I think it's very important to go back to that sort of like, get into the world, ship things, don't be hesitant of self promotion in a way. So being very honest with yourself, like building something, getting it out, or see how people react, go back to the drawing board.

1:08:30I think it's about exposure. At Versailles, we have one of our sort of internal operating principles as increasing exposure hours. Try to quantify how much time you expose yourself to watching how people use your products. Even to watch how people use other products. And you've developed that muscle. Like, taste sometimes, I think we think of as like, this inaccessible thing that, oh, that person was born with taste. I see it as a skill that can develop. And again, the AI will help you a lot here because we try and capture some of the universal principles of it. But there's also trends in the world, right?

1:09:19Like, I'm not a super, like, couture guy, but like, you can see that like every year, like Paris Fashion Week has like a theme to it. And like, there is some innovations that have some breakthroughs, whatever. And so trying to stay at a different tier, or even trying to define different tier as well is certainly very exciting. I love how doable this is, increasing your exposure hours. Basically, what I'm hearing is use the best apps. Yes, there's like a feedback cycle component to it, just like show people. And they're sending this nuances, right? So I actually recently created a, I published it to my community at Proheon V0.

1:10:01I created a chat GPT style interface inspired by GROC. And I captured a few things that GROC does that are just so smart. So on mobile web, when you press enter on their input, they default to creating a new line. Because they know that the way that people are used to submitting things on mobile is not by hitting enter, like we would do on a desktop computer. You can tap the little icon and like your message goes out. On desktop, they inverted it. When you press enter, you're expected to submit. And I think if you got a new line, I think a lot of people would get frustrated that most people don't know that they can press command enter to submit and whatever, and like it slows everything down.

1:10:45And you can basically prompt for those things, right? But you have to pay attention to the details. And you have to decide what you want to see in the world. And sometimes that means like either defining best practices or seeking the best practices and learning from others. Another aspect of exposure hours is that you tend to overrate how well your products work. It's very important to give your product to another person and watch them interact with it. Expose yourself to the pain of reality. And the more you sort of like submerge yourself in the real deal, nitty -gritty of what happens when people use your interfaces and whatnot, I think you'll come out stronger, more grounded, hopefully more humbled.

1:11:34We don't like pain though. And I like that this is a push. Create some more pain in your life. Show people that you're building. Is there like a, do you have like a new heuristic or a number of how many exposure hours per week per month you want your team to have or is it just more is always better? Yeah, more is always better. I mean, because a inertia is to get inside your head and the inertia is to think everything and assume that everything is going good and there are no errors. Of course it's fast. It worked on my machine. I think it's always a push for more. I do sometimes little things like I ask my team to color my calendar.

1:12:12So I say like I have to have certain amount of like one and once with my team, in represent of my calendar, kind of like meetings so that I can sync with people and see how the company is doing. Then I want to have customer meetings. In doing those customer meetings, I push myself to use the products. In fact, with our enterprise customer, something that I do is like, I try to forget how things are build, would feature of next year's reverseseller using whatever. I just frequently use their products. And I want the product to be great. That's all. And then I could try to work backwards into the signal.

1:12:49So a form of exposure hours for me is seeing what kind of success our customers are having in the real world. And so, but again, it's just eheuristic. Like maybe one third of my meetings this week were customer meetings and I tried and watched them do. Another really quick one is we invite people frequently to demo how they use the product at live. Sometimes to the executive team, sometimes to the whole company. And we always inevitably discover something interesting from the customer about maybe there is something that they're in pain about that we didn't know about or maybe something was not as intuitive as we thought.

1:13:31And I find with these sorts of things, when you do them, when you talk to customers, you have them show how to use the product. You always like, why have I not done this off more often? Why am I, what am I thinking? This is just so mind blowing usually. Yes. I want to talk about limitations of VZIR at this point. So what should people know about just like what VZIR can't do? Like can you, if you have an existing code base, can you plug it in and start doing stuff? Or is that coming? What else should people know? Just like, okay, it's not going to do this yet. But yeah, you can import code basis through zip files and get this coming very soon.

1:14:04It can do full -sac development, it can connect to APIs. In the next couple of days, maybe even before its block as is out, we'll have this very tight integrations so that if you need a database, or if you need an AI model, or if the AI decides, it needs that, it'll just seamlessly install it from the VZIR marketplace. And the VZIR market base has already curated some of the best infrastructure products in the world to store data, to search data, et cetera. So it's going to make the product even more powerful. I'll say, again, I kind of did that exercise, and I do that exercise every day of like, I have a wild idea and try to see if it can come to life.

1:14:46It's very powerful so far. AI's are still very much a work in progress. They can make mistakes. We have it as a little disclaimer underneath the input. You will find errors are fitness function and we've seen such a stronger relation between user love and retention. These are actually their retentive product compared to other AI products that I've built in the past, or little demos that we've done, or whatever. If people subscribe and use it every single day and are very sort of, if they notice a bug, they're very, very like, jeanery about it because they're depending on a day and in and out. But I'll say errors are still possible, right?

1:15:28Like, every once in a while, you might get a runtime error or whatever, but a lot of the technology that we've added is so that VZero is very agentic. It has a lot of agency in how to act. So you're going to see very frequently that if it runs into errors, VZero tries to solve them itself. And then last I will say, when products get really big, AI today is just not as good at dealing with massive code bases. But going back to that idea of the React component, because we break down things into files and components, we tend to do quite well in that dimension. In fact, one thing that NextShas was sort of known for is that in order to sort of project, you just create a file, and NextShas were route to that page.

1:16:27It was, if anyone is familiar with PHP, it's kind of like how PHP worked. And so it's so good that LLAMs are good at working with files now, because it fits very naturally into our world. And if you can scope down, when things get really big, if you can give it a smaller task, to work on a specific component or a specific file, you decrease that likelihood of the LLM not being able to reason over or very, very, very long context windows. I want to go back to design. We talked about how VZero is really good. It's just great design by default. It's a kind of lean into that more. If someone wants to improve the design of their product, most people are not designers.

1:17:07They don't really know how to make it look good. They don't know what to ask for. They need just tips and breast practices for making their app even better, look even nicer. Yeah, it was really interesting. At the other day, I met with a CIO of a large bank who on the side does a lot of coding, or tries out new technologies and whatnot. And I showed him VZero and he immediately became a VZero addict. He texts me everything with feedback. He moved two websites of his own from another website builder type provider, two VZero and Vercel, deployed them, given the domain name, they're live in production.

1:17:44And then he said, look, I have this challenge. I have this music festival that I organize with a couple of friends. And this is what the designer gave us. And he had this kind of like brochure. It looked very much like a print style design. And so he gave that to VZero. And the first result he was kind of like digging me for is like, look, this is in look good. And then because I have experience with the tool, I said like, why don't you just give it the feedback? Like little age, you know, you were asking yesterday, like earlier some of the things that I've learned from the product. And now is, or the best practice, what would I recommend if we were sitting next to someone?

1:18:24Not only you should not hesitate to give the AI feedback, it's so interesting, dude. Like sometimes people will press the feedback button to tell us what they wanted VZero to do. And literally all we had to do in many cases is can you just tell VZero that? And so he sent me this message saying like, yeah, I just don't like the design. And I give him back a prompt that I would have given. I said like, I don't know what I said is specifically better. Make it more jazzy. Get it, make it more, make it pop, make it. And so trying and again, it goes back to like, try to draw inspiration from variety that the AI already knows about.

1:19:08So in a couple of problems, we ended up something that was in his mind better than the original print design of that brochure that concert lineup. And at that time, again, I'm even learning about what VZero is capable of and the best way to use it. But with design, I think unleashing its creativity with and seeing things and playing with it is definitely super helpful. So one thing I'm hearing here is just tell it, make this look better. Or I don't want to talk. Make it pop. You can, I totally see. But and if you can use tokens that are relevant, so neo -brutalist, minimalist, newspaper like vintage, make it look like a telegram.

1:19:58Like you can sort of like try and reach for things that you, maybe it would not naturally come to mind and you would be surprised about how well it can transfer those ideas into reality. Incredible. It's too easy. Maybe to close that conversation, we'll see where this topic goes. I have this tweet that I loved that I super resonate with. The secrets of product quality is blood, sweat, and tears. I completely agree. I think that's why I think my newsletter's been successful. I spend so much time in every newsletter post more than I think anyone spends on these letter posts like 10, 20, 30 hours.

1:20:33And that's why I think it works. Is there anything more behind that tweet, anything you've learned, just the importance of working hard? I guess, to great, great stuff. Yeah, I mentioned, I suppose you're always a good example of like, look, it can be painful. It can be painful to see your baby break in front of everyone and noticing all the other thing is that a great product is made up of 1 ,000 little details, right? And so you're never really done. There's a humility that comes from the process also of why the best product builders will say, nine knows for every yes. Because when you say yes, is like adopting a puppy, a feature is like adopting a puppy.

1:21:16It grows into a beast that you have to take care of and it's very demanding and loving, but also, you know, it's a lot and poops everywhere. So you have to have a creative restraint and while you also have to have a give, you have to withhold sometimes with the respect of the real world complexity that emerges. A little thing that I kind of obsess about, I'll give kudos to the mid -journey team. I really love how mid -journey works in mobile web. I don't know if they have an app yet, like a native app, but like the mobile web site is phenomenal and to get it to be that good, by the way, it's possible.

1:21:58It's actually possible to get, make great things on mobile web. But it needs that sense of love and restraint and obsession and testing a lot and using your own products a lot. Duck fooding is a great mechanism, obviously. So we use the heck out of Versaille N. V. Zero to make Versaille N. V. Zero and hopefully that helps us do better. But there is a lot of blood, so it enters the process. Yeah, you can tell how much you use the product. Like it comes through and everything you say. Let me actually ask about this. You talked about how you said you have 600 engineers. No, 600 people total. 600 people total.

1:22:41150, so. How is AI changing the way they work? Is there anything there? I feel like you guys are the cutting edge of how products are built. What's happening? Is it just everyone's on Versaille N. V. Zero to build stuff? Yeah. Yes, but actually it's more profound. I think it's the everybody can ship. It's the we build with AI principles in mind. I actually give a shout out to the Luma Labs engineer who said, well, I'll use AI for everything. I'll use AI also generate the images for the website. And I'm seeing, for example, our designers that are working on our next conference generate all of the animations with video models.

1:23:27I'm looking at our marketing team, our creating demos of how the infrastructure works with V. Zero that are better than any static diagram or landing page that I've ever seen. One of my most viral zits or expos is something that one of our designers created, which explains how our computer infrastructure works with an interactive demo. And until until he created that, by the way, he designed it and created it and we shipped it all in the tool. First of all, it wasn't part of his day to day job to do that. He, these years making you such a powerful generalist that you can step out of your conference zone of like, well, my job was to do only this.

1:24:17You just create. We have a ritual every Friday. We had it this morning called demo Fridays. And so it's very important to create the space for people to step out of that conference zone and use AI. So us giving permission to be able to build and ship things as part of that cultural backdrop that makes this thing possible. We had a demo today as part of demo Friday of our VP of sales engineering, also creating an amazing tool that he's going to use to help prospects under send for sale with V. Zero. So I've heard from DevOps and infrastructure engineers how much they use tools like cursor to work on the low levels of the versatile infrastructure.

1:25:03So I think very quickly we're seeing AI being embedded everywhere. I just heard that product requests from a customer and said, I was saying, okay, V .R. sell you sell domain names. Let me let me come up with new domain ideas with with AI. So I just see a future where AI becomes synonymous with software. I do look forward to it because we need to stop talking about AI at some point. I foresee it's probably not going to happen, but it is useful to remind people that AI equals software now. And we're a software company. We build software and we use software to build software. So. And AI is just a part of that.

1:25:42Yeah. Guillermo, what a beautiful way to end it. Is there anything else you wanted to mention anything else that you want to leave listeners with before I let you go. I leave you with my vision of the future, which is we have this billboard in San Francisco, which is everybody can cook. It's also part of the Rata 2E film, one of my favorite movies. I look forward to a future where everybody can get their ideas out there. If you can dream it, you can ship it. And also that when you use products and when you see the creations of other people and the things that they put out into the world. That we are collectively making the world better.

1:26:25So anything you experience hopefully gets faster, higher quality, fewer bugs as we go along. And I think we're all contributing to that. And I look forward to that and look forward to everyone's feedback on how Versel can play a part in that future. So to build on that, we can focus on online. Where can they just go to Versel .com is at v0 .com. And go to v0 .dev. That day. You get started. I did mention that if you want to build your own v0, this is more advanced. But check out our templates on Versel .com is such templates. And also I'm Brouch G on X. So you can DM me or tweet at me at any time.

1:27:11Amazing. Gare, Ma, thank you so much for being here. Thank you, Lanny. It was so fun. Everyone. Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show at Lanny's podcast .com. See you in the next episode.