How I built a $3B AI Startup + 7 AI Business Ideas
Vercel, AI Business Ideas, and Open Source - May 28, 2025 (8 months ago) • 01:14:09
Transcript
| Start Time | Speaker | Text |
|---|---|---|
Guillermo Rauch | Myself, I've been an engineer for like, you know, a couple of decades now. I no longer write code; I only prompt.
| |
Shaan Puri | Alright, so you founded this company, but your story's crazy. You're a high school dropout, grew up in Argentina, and have been building and hacking on things since a very young age. You sold a company kind of early on; I don't know if it was a big sale or a small sale.
Then you built this product that has just taken off. Every front-end developer I know loves it. It's valued at, I don't know, $3 billion or so, give or take. You've just done this incredible thing.
And now you have this AI tool that's also super on trend and is doing really, really well. It's a really cool agent that builds sites for you. That's my version of the summary of your story. It's a great summary.
| |
Guillermo Rauch | Maybe the only thing I'll add is the crazy way that I've been able to go from being a teenager in Argentina to today. It has been a lot of open source. I've been involved in creating a lot of technologies that have become foundational in the tech ecosystem. I felt like that, and the web, has sort of been my ticket to success, of course, over decades of hard work. | |
Shaan Puri | Well, explain that. So, why did...? | |
Guillermo Rauch | You drop out of high school. I've never been a fan of the high school dropout moniker because I actually really loved the high school that I went to.
It was a high school in Argentina. It was a free public school that had an entry exam. You had to study really hard to get in, and I worked so hard to get in, entering in position number 10 out of thousands of students.
But I had two competing interests. I was becoming popular in this open-source ecosystem because I was creating libraries for JavaScript and front-end development.
| |
Shaan Puri | You're becoming popular in open source, but you're only 15 or 16 years old. So, when did you start?
| |
Guillermo Rauch | I started coding very early. I would say seriously, I began when I was 10 years old. I was creating websites and shipping them. I started doing work online, helping my parents with our home finances.
| |
Shaan Puri | Was it just a lucky break, or what got you started?
| |
Guillermo Rauch | Lucky break in some ways, but open source. I was contributing a lot to online forums, helping people out. The lucky part was I remembered this guy whose name I guess he'll never know. He was like "Dark Shadow 123." He said, "Hey, you seem to really enjoy helping people out by writing tutorials and guides and things like that. There's this website, it's a freelancing website. You could just sell your services here because you know so many things about Linux, PHP, and programming."
So there was a bit of a lucky break in that I figured out a business model for myself really early on. I got my first check when I was like 11 years old and started. I had a client in the Netherlands when I was like 12 or 13.
| |
Shaan Puri | Are you pretending to be an adult, or are you openly like, "I never come up"? You never...
| |
Guillermo Rauch | I wanted really badly for it to never come up, and I'm really, I guess, lucky that at the time, like, even Skype was not a thing. | |
Shaan Puri | So it. | |
Guillermo Rauch | Was like actually kind of rare that you have to get on the phone. So, I really took advantage of that.
When I got into this high school, my reputation for doing all of this work and my reputation in the open source world were both growing simultaneously. So, as my grades were decaying, my sort of online net worth and contribution in notability in the world was growing.
I would write articles that would get to the front page of Digg.com. I would write open source software that would get a lot of traction. I would get written about.
| |
Shaan Puri | Are you a genius, or were you just being extremely helpful? I’m just saying it in a dumb way. Was it that nobody was writing the tutorial on how to host your WordPress site or whatever?
| |
Guillermo Rauch | Right, I sort of went away. | |
Shaan Puri | Like you were figuring things out, and it was really cutting-edge stuff. Where were you?
| |
Guillermo Rauch | Yeah, when I advise young people on how to bootstrap their careers, I say start by teaching anything.
I started with how to compile a project called RPPPoE to get internet connectivity in Linux. It's just like writing down the tutorial; the community will do a hundred times better job, right? Or, at best, it becomes training data for an AI to then explain it back to people.
But then, over time, I started coming up with my own breakthroughs. My quote-unquote "big break" was when I started contributing to a library called MooTools when I was 15 or 16 years old. This library got picked up by Facebook to become sort of the inspiration and foundation for their JavaScript infrastructure. I received a job offer from Facebook when I was about 17 years old, and my contributions to that project started to become more important.
| |
Shaan Puri | You got a job offer from Facebook when you were 15.
| |
Guillermo Rauch | Yeah, so I was probably 16 or 17, during the super early days of Facebook. In many ways, you could play out another timeline in which I was in America already and was an early engineer at Facebook.
Did you turn it down? Well, when they discovered I was in Argentina at that age, I was like, "Oh yeah, maybe we should look for someone else." But that same project kept opening up doors for me because other startups started using the same foundation. They were like, "Who should we hire?" The first thing you think of is, "I'm going to hire the people that contributed to the project." We do that ourselves today with our open-source projects like Next.js. We go in and say, "Okay, who's contributing? That person seems really interesting."
So when I was about 18, that's when this startup from Switzerland reached out and said, "Hey, we want to hire a MooTools developer." That's when I basically just dropped out of high school. I had my first real job offer from a company in Lausanne, Switzerland. For my parents and for myself, it was kind of surreal, right? I'd never left the country, and I was leaving Argentina for the first time with a job offer in hand in an amazing country. You know, it was kind of surreal.
| |
Shaan Puri | Yeah, we hired a kid when I was doing a startup in San Francisco. We had a guy who was in eighth grade, and he emailed me, Johnny Dallas. He said, "Hey, I love to code. My dad met somebody—I met Pete at a dog park, our sys admin guy—and he was like, 'I don't know anyone else who codes. Can I just come hang out for the summer? I just want to be around other programmers.'"
I was like, "Oh man, amazing! Yes, for sure." He comes in on the first day, and we actually give him a test. We're like, "Hey, we want you to make this little onboarding quiz in HTML. Just make a quiz, like multiple choice, take them down to flow, and land them in one of these four buckets."
He just sits there until like seven or eight at night. I feel bad; I'm like, "Oh," but he's not asking for help. I just want to see how it plays out, and he actually ships the quiz at the end of the evening. I was like, "Alright, this kid's legit."
By the time he's in tenth grade, we're like, "This guy's working basically full time for me." After school, he's coming in. I had a conversation with his mom; I remember it was downtown. She said, "I can't imagine my son being like a high school dropout." I said, "Do you know LeBron James and Kobe Bryant? Your son is going pro." So I think that should be a little bit of... | |
Guillermo Rauch | Yeah, totally. | |
Shaan Puri | If any kid is listening to this, it worked like a charm on that mom. Try this on your mom or try this on anybody where you can really go pro early.
| |
Guillermo Rauch | Yeah, totally. I always give people very caveated advice. I tell them, "Look, I went to a high school that was giving us college-level content." So, I actually developed a good foundation, a good "world model." I became quite competent in a lot of sciences, and we had really good chemistry classes, physics, math, etc.
I'm actually really thankful that I had a good foundation. By the time I decided not to continue down the normal "educational career," I had a really good alternative. By that time, it was really clear that my skills were going to take me somewhere. I didn't know specifically where I was going to end up, but I was eventually going to end up in San Francisco building companies. It was a bit of a leap of faith, but I was well substantiated in existing evidence of success.
| |
Shaan Puri | Hey, let's take a quick break for a message from our sponsor, HubSpot, who's making this episode possible.
Listen, if you're trying to build something big—I'm talking about a hundred million dollar or a billion dollar company—one of the most important things is to focus on the market. Where is the opportunity? You are like a surfer on a surfboard, trying to find the biggest, best wave possible to ride.
Well, HubSpot has put together a cheat sheet that studies and reverse engineers what's working today. They are looking at the top performing industries, the funding routes that work best, and the growth strategies that the fastest growing companies are using.
So, if you want a one-pager that is a cheat sheet on the big company playbook, you can get it right now. Just either scan the QR code that you see on the screen or click the link in the description below.
Alright, back to the show. You were talking about MooTools and JavaScript. I read something great that you said: you kind of bet on JavaScript because you realized that for the back end, there are a hundred different languages you could choose, but the browsers only know JavaScript. Right? It was...
| |
Guillermo Rauch | It's like an unfair advantage. Looking back, I think you always arrive at success by finding asymmetries, or finding alpha, or finding unfair advantages.
JavaScript has this unfair advantage, and I can explain how it won as well. But right now, every single device on the planet, on the client side of a web browser, can run one language, and that language is JavaScript. It can't run Python, it can't run C++, and it can't run Java.
The way that JavaScript got there was actually by beating a lot of alternatives. So, way back in the day, we had Macromedia Flash as a proprietary plugin that Steve Jobs put to rest eventually with the iPhone.
| |
Shaan Puri | Was he right on that, by the way? Was his...? | |
Guillermo Rauch | Absolutely, his taste is correct. We had Java applets, right? So there was an idea that the JVM (Java Virtual Machine) was going to be that universal language. The fact that when I started going deep into JavaScript, there was still a little bit of a perception that it was like a toy. | |
Shaan Puri | Why was that? Why would... because, okay, I'm a mostly non-technical person, but I love UCC. I love situations like this where there were a hundred contenders and one win. But it doesn't win because it had the obvious traits.
| |
Guillermo Rauch | of | |
Shaan Puri | What you would think would win, but there's this other thing that actually proved to be a big advantage.
| |
Guillermo Rauch | Totally. | |
Shaan Puri | Yeah, we were just talking about Midjourney before this. We're like, "Midjourney, it's not like product design." It's not like some designer sat down to create it. Then you'll open up Discord and there'll be a hundred random channels, and in it, there'll be strangers making images. But actually, bad was better. Bad was good. Yeah, because you learned how to use it. I've...
| |
Guillermo Rauch | Heard of the saying, "Worse is better"?
| |
Shaan Puri | Not worse. I haven't seen worse. Better, I've seen the Paul Buchheit one where he's like, "If you're great, you don't have to be good." I think it's maybe similar, but can you explain "worse is better"?
| |
Guillermo Rauch | **Worse is better** was a paradigm in the early days of the internet. Specifically, that essay spoke about "worse" in the sense of being less powerful. Sometimes, when you constrain a technology, you make it a lot more predictable.
There were some advantages to markup languages that made them "worse but better," and they became the successful foundation of the internet.
But there's a broader perspective that Brendan Eich, the creator of JavaScript, has pointed out. He noted that sometimes the success of technologies or startups can be best understood through the lens of evolution, natural selection, and Darwinism, rather than the obvious "intelligent design."
There was an article that he shared at one point called, "What Would Charles Darwin Think About Clean Slate Architectures?" | |
Shaan Puri | Okay. | |
Guillermo Rauch | And so, JavaScript was almost like a piece of DNA that evolved and became more sophisticated over time. It started out looking very simple; it was a piece of code that would inline into the markup.
When you think about markup in the sense of MySpace codes or HTML, it was like, "Can we bring HTML slightly more alive?" We needed something that was very minimalistic that we could put right there inside the markup code. From there, it became more powerful and more sophisticated.
However, most people, when they looked at it in the beginning, thought, "Well, it can't be fast, it can't be typed, it can't be correct, it can't scale, it can't have a module system." But all of these observations that people were making were not actually technical. They were saying, "Well, at that time, as it exists, it cannot do those things."
A lot of the alpha that I created in my career was by actually taking it seriously and saying, "Well, we can add this thing." A very simple example—it's a little technical—but JavaScript lacked classes. It couldn't do object-oriented programming in the conception that people had with things like ActionScript, C++, or Java.
But in the MooTools team, we figured out a way of "faking it." We created a function called "Class" with an uppercase C. When we presented it to developers, they were like, "So you added classes to JavaScript?" Well, yeah, we kind of added them. That actually was one of the salient features that stood out to the Facebook team and many other teams in the Bay Area.
By evolving that thing and actually betting on it, that became the asymmetry of my early career. Again, the experts at the time—this is why I think there has been a shadow of doubt cast on experts for many years now—were familiar with this. Successful angel investments had the same characteristics: "Look at how rough that entrepreneur looks," or "Look at how bad the homepage looks." You have to be able to project out what it's going to be in the future.
And now, JavaScript has eaten the world.
| |
Shaan Puri | Yeah, one of the best blog posts I ever did was when I went back to the Wayback Machine. I said, "Here's what the first website of Airbnb looked like. Here's what the first website of Uber looked like."
Which is actually brilliant and very important for entrepreneurs to look at. You kind of know intellectually, like if I ask you, you'd be like, "Yeah, I probably started off rubbish." But literally, go look at it, read it, and look at the thing.
| |
Guillermo Rauch | Look at the pixels. | |
Shaan Puri | Look at the actual pixels because it does a couple of things. First, you're overthinking it. Whatever you're doing, you're probably overthinking to start the launch.
Second, it shows how far you really have to go to iterate and make things better. It also illustrates how narrow of a wedge you take. Like with Airbnb, their initial concept was for a design conference where the hotels were booked. They offered an inexpensive stay to sleep on another designer's couch.
| |
Guillermo Rauch | Yeah. | |
Shaan Puri | Couch or air bed?
| |
Guillermo Rauch | Yeah. | |
Shaan Puri | And you just take so much from those initial web pages.
| |
Guillermo Rauch | Yeah. | |
Shaan Puri | And you're right. Once you hang around in Silicon Valley enough, it humbles you. The things that look like toys, or the things that start narrow, or the things that seem limited on six dimensions but are really good at this one thing, they can't be underestimated.
| |
Guillermo Rauch | And it does raise the question of **genius versus accident**. There are just so many good stories about the creation of JavaScript.
One was the extreme time constraint that Marc Andreessen put Brandon Aik under. The reason it's called JavaScript is that they needed to market it as Java; they just added "script." There's no relation whatsoever between those two things.
Another point Brandon always mentions is that it took him ten days to conceive the language, primarily because of deadlines. So, they had to ship something that makes pages come alive.
For those who are not technical, the way I sometimes explain what we do with running JavaScript on the server and running it on the client is by referencing a newspaper in *Harry Potter*. It's like a regular newspaper, but when you open it, it comes alive.
| |
Shaan Puri | Pictures move, yeah. | |
Guillermo Rauch | So, think of getting the newspaper as the server giving you the server-rendered, pre-rendered artifacts. It comes with all the letters, all the static images, etc.
But then, it's really cool that you can open it, and what we call it technically is that it becomes "hydrated." It becomes alive, and more code can run on your side of the equation. That can be a very enriched experience.
So, that's the power of JavaScript. | |
Shaan Puri | That's a great analogy.
| |
Guillermo Rauch | It can run at the production line. It could be manufactured, it could be printed, but then it can be shipped to you and then it can become alive again.
So, in those ten days, Marc Andreessen wanted to basically pitch that. One of the initial names that I think was "Live Script" and then they renamed it to "JavaScript" to market it more for enterprise traction.
It'd be akin to today, like we want to call things "agents," and perhaps they're not agents or whatever. Like, here we go.
| |
Shaan Puri | To use the evolution example, it's like the skin color blended in with the trees. It's like JavaScript just blended in with Java. That's right. | |
Guillermo Rauch | And that. | |
Shaan Puri | Gave it an evolutionary advantage versus getting eaten because it stands without.
| |
Guillermo Rauch | The other interesting observation is, okay, so you have ten to twelve days to ship something, whatever it is. What is the minimum surface that you can ship on top of which evolution can be bootstrapped? I think that's much better than trying to aim for completeness of that initial version.
What I ask myself when I study the success of others is: Were they really clever to think, "I have to delete, delete, delete"? Like, I think you were referencing Rick Rubin's reducing philosophy. I'm a reducer, not a producer. So I think if you're truly brilliant, you'll find that you have to delete and delete and delete. That's one path to success.
Because you know what the more complete picture would look like, but you can exercise that restraint, or you could just stumble upon it through deadlines. Right? Like you have the classic Y Combinator model: "Let's make you ship a startup in three months." That acts as a forcing function for the reduction of the surface.
I think there's something about human nature, or perhaps the pressure that people put themselves under, which is that they feel they have to add more. They have to make that homepage, add more images, add more gradients, whatever. But then, through the exercise that you talked about, you can go back and see, "Well, actually, things look pretty simple." They focus a lot on the content and the essentials. | |
Shaan Puri | So, you're doing MooTools. You go out there to Switzerland. How do you get to Vercel, and what insight did you have to start this company that has become this juggernaut?
Right? Like, yeah, I think I've DM'd you like every year for three years being like, "Hey, can I invest in this thing?" Because it's such a juggernaut that it's obvious to me that when you have this sort of developer love...
| |
Guillermo Rauch | Yeah. | |
Shaan Puri | You really can't help but win. You were on the right waves to be on for the last few years.
| |
Guillermo Rauch | The pattern throughout my career has been iteration. **Velocity** is the most important thing to optimize for.
In my previous startup, which I sold to WordPress, anytime I would start a project, I would begin with the mechanism to ensure I was the CTO of that startup. I focused on creating a system that allowed my colleagues, my engineers, and everyone in the company to ship really fast.
My obsession became: how quickly can you go from idea to shareable artifact? From idea to URL? Today, it seems obvious that version zero is so successful, and that you can prompt and get a link, going from prompt to application. But I didn't have AI at the time. What I had was the ability to streamline the deployment pipeline of ideas.
| |
Shaan Puri | So, just to summarize what you said: the most important principle for me when I work on a project is that we're going to be able to ship fast and iterate fast.
| |
Guillermo Rauch | Ship fast. | |
Shaan Puri | In order to ship fast and iterate fast, we need our pit crew.
| |
Guillermo Rauch | Yes.
| |
Shaan Puri | You know, if we're a Formula One car, we know our pit crew should not take six hours to change the tires or to get it refueled. We need to get back on the road to go for the next lap.
So what you were saying is you would focus more than the average bear on that pit stop between ideas. Yeah, and actually, versatility... your body exists.
| |
Guillermo Rauch | So, I had the insight that instead of just assuming that the world takes hours to ship or that you can ship once a week, I thought, "No, let's look at the web." The web is so fluid for all the reasons we just outlined; it's just so alive. Why can't we be shipping a hundred times a day? Why can't we try lots of different things?
I actually just interviewed recently for Offsite the founder of DoorDash. He was talking about how DoorDash started out with one HTML page and six PDFs of the restaurants that they were going to deliver for. They just brought the idea online as soon as possible. For some people, it becomes obvious that the most important thing is to get the idea out into the world as a URL and see if it sticks.
So, I wanted to create a platform where that was the norm. That was sort of the inception idea for Vercel. Before Vercel, I was at WordPress. I noticed that WordPress as a company had become quite good at deploying one app, WordPress.com. But if you were working for the company and you had a new idea, you were kind of getting stuck. You had to go to the IT team and say, "Hey, please can you set up a server for me? Can you give me an area here where I can come up with a new application or a new idea?"
So, Vercel started out as a way to reduce that friction from idea to live in seconds. In fact, it became such an obsession that we started measuring each millisecond. Like, you have an idea, you write it down in JavaScript or HTML or whatever, you press deploy, and how quickly can we get it online? It was literally a little obsession.
I love the Formula One metaphor because it's about shaving down the seconds that were stopping the scar. This scar is not just one app or one idea or one person; it's literally how the entire business works. This is why Vercel has been so successful with the largest companies in the world, like investment banks, and it also powers most of the Y Combinator startups that are creating new ideas in the manner I just described. It's like, "Hey, I need to try something out, and I need to hit demo day. What is the quickest way to deploy?" It's Vercel.
| |
Shaan Puri | So, it sounds like this might be one of those startups where you didn't need to pivot a ton. It seems like you had the correct idea; you knew the pain and the problem correctly. It sounds like you had the right idea for a solution, and then you obviously made it better.
| |
Guillermo Rauch | But is that true? Yes and no.
When you achieve some level of success, people start studying your success a lot. People ask me a lot, "Did you pivot? Did you not pivot? What can it be, the chicken or the egg?"
The reason that they ask me this is that we have an extraordinarily successful open-source project that Vercel created called **Next.js**. Next.js powers a lot of the modern internet. Like you talked about, **Midjourney** is built with Next.js.
| |
Shaan Puri | And by the way, you make money off of that? I didn't understand. It's just that you open source it. It's open source and widely adopted. Does Vercel benefit from it in any way, really?
| |
Guillermo Rauch | Or it's because it's in the service of, "Okay, how can we get that Formula One car going?"
Okay, start with Next.js. You're going to cut down on the whole assembly of the car.
| |
Shaan Puri | Okay. | |
Guillermo Rauch | The alternative to Next.js is that you have to procure the chassis, the wheels, and the engine.
Yeah, like this is actually what was happening to engineers when I started the company. I was like, "Okay, so how do you start a new idea?"
"Oh, well, I go to Home Depot and I shop for like 200 different kinds of wheels. I grab the wheel, and I go to this thing, and then I assemble it for my one car. Then maybe I get started running it, and then maybe I see if I have product-market fit."
| |
Shaan Puri | So, you created Next.js, and I believe the story is that you were trying to build something. To use React, you thought, "Oh shit, I gotta go get the engine and the testing. I gotta go get all these pieces."
Alright, I'm just going to build this kind of template for myself once.
| |
Guillermo Rauch | The website for the company is kind of like the experience of DoorDash. I had to build zeit.co, our domain name at the time. I said, "Okay, to get started with React, I have to do like... get a PhD."
So, to your point about the idea, this idea is so powerful that it cuts down the time for humanity to go from some hypothesis to a production-grade deployment. We're going down from weeks of setup to seconds. It seems obvious, but what was not obvious and felt like endless pivots was narrowing down the scope.
The reduction that you just talked about started out with, "Oh, you can deploy anything! You can deploy Java." It's so contradictory even to my genesis story, right? You can deploy Haskell, you can deploy PHP... and then we realized, "Wait, why are we doing all this?"
Clearly, we believe that the modern web will be powered by frameworks like Next.js. We believe that there's an alpha in the market and in the world of democratizing this idea of using both JavaScript as your sort of backend and frontend language, which massively simplifies software development.
So, I will say we didn't pivot in the sense of going into a different space, but we simplified the offering way, way, way significantly. | |
Shaan Puri | You have this interesting seat where you get to see what a bunch of people are building on your platform or using V Zero or whatever. You're also on the edge, I would say, of being a tinkerer, hacker, or technical person who kind of sees what's possible. But you only have so many hours in the day.
So you tweeted out this thing that said "free AI ideas." Yeah, and it's like if Gordon Ramsay opened up a lemonade stand, it's like, "Oh wow!" You know, Steve Jobs was like, "Hey, free product lessons for toddlers."
| |
Guillermo Rauch | Yeah. | |
Shaan Puri | I'd be like, "Get my daughter in!"
| |
Guillermo Rauch | That class. | |
Shaan Puri | Right, so when you say "free AI ideas," I want to show up. Can we run through some of your AI ideas?
| |
Guillermo Rauch | Yeah, and I'll give you a little bit of background. I've always been obsessed with democratizing the web for everyone. Anyone with an idea has to be able to create. That's why we created V Zero. V Zero is like ChatGPT, but for creating web applications. Instead of giving you text, it gives you a fully working web application. | |
Shaan Puri | You say, "I want to make an app or a site."
| |
Guillermo Rauch | Make me the next DoorDash. You can literally type that in.
| |
Shaan Puri | It will make one when you deploy it. | |
Guillermo Rauch | And the difference, I would say, is that there are many players trying to build this. But one of the things that I'm really excited about is that it's banking on the lessons of the last ten years of building the world's most popular framework for JavaScript and creating the production-ready infrastructure to support it.
For example, I'll name a cool brand that uses Vercel: Ramp, Supreme.com, Brex. In fact, we have so many successful companies in every market segment that you could imagine. But all of those companies needed expert engineers.
In order to build a really cool website like DrinkAGwater.com, you need to learn Next.js. Now, the magic of Vercel is that we cut down the learning curve from needing to know all of the foundations of computer science and all of the foundations of AWS and how to deploy software. We brought it down to just taking a React course and using Next.js.
But with AI, we can cut that down even further. It's just about speaking English, and we will steer the model towards what we think are going to be the world's most successful outcomes.
We care deeply about performance. When e-commerce websites deploy on Vercel, they vastly outcompete everything else. We recently heard about a public company that sells billions of dollars worth of consumer electronics a year that improved their conversion rate by 30% in some markets and up to 90% in others. Why? Because the website is faster.
So imagine if you could go to an agent and say, "I want to create the next big competitor to Apple.com. We're going to make it faster, more accessible, beautifully designed, and deployed on this enterprise-grade infrastructure."
| |
Shaan Puri | You get those out of the box. You get those for free.
| |
Guillermo Rauch | Out of the box, it's true to the spirit of the company. It's about how we can get that Formula One car running as soon as possible. So, it's opening up creation and deployment to basically every human on the planet. | |
Shaan Puri | Okay, but you were saying that's context for the AI ideas. Because what?
| |
Guillermo Rauch | Because of two things. One is that with V Zero, any idea that I have, I can bring it to reality. But then there is a meta aspect.
We're being very successful with V Zero because we created AI for web engineers or AI for people that are interested in shipping to the web. There are going to be so many other verticals that are going to be similarly disrupted, and it seems really obvious to me because I'm on the, you know, the inside. I'm behind the scenes of building things like V Zero.
I think there's going to be, for example, why don't we have a V Zero for creating video games? A studio that combines the best of both worlds of Software 1.0 techniques and Software 2.0 techniques. I don't know if you're familiar with that framework, but...
| |
Shaan Puri | **Explain.** | |
Guillermo Rauch | So, Andrew Karpathy, who's the leader of AI at Tesla and later OpenAI, came up with this incredible essay that I think is now considered canon. Everyone must read this because it's so ahead of its time. He called it "Software 2.0."
He says, "Look, Software 1.0 was what I grew up with: programming languages, data structures, algorithms. You learn how to make things more efficient by writing better for loops and recursion. You're in control of everything, and everything is very deterministic and predictable."
In contrast, Software 2.0 still uses the foundations of computer science, but we're making the process a lot more stochastic and probabilistic. Instead of writing every piece of circuitry in the programming language, we're relying on training models with data. The output of what those models do might resemble what a Software 1.0 program might do on its own.
The best example would be using ChatGPT-4 with image generation to produce incredible diagrams. There are ways of generating those diagrams with traditional software engineering, but this AI is almost like a miracle general-purpose program that can do anything based on what it's been trained on. So, let's call training and neural networks and AI "Software 2.0."
| |
Guillermo Rauch | O, and it has all of these magical emergent properties that were not thought through by the engineers. The engineers didn't have to go and write every "if-else" branch that is under the hood and think about every corner case. That's why it's so exciting to people.
Because every time a new model comes out, we're all like... we call it "discovering the latent space." We're all trying to figure out what is even possible because even the creators of the models don't know what's possible. This is in stark contrast to Software 1.0. Software 1.0 is like there's a PM giving you tickets, right? I know exactly...
| |
Shaan Puri | What this can do, I know exactly what it can't do.
| |
Guillermo Rauch | Yes.
| |
Shaan Puri | I know if you push this button, you're going to get this exact result. I know if you push that button, you're going to get that exact result. The new model, you're saying, even the makers of it are like, "I'm not sure exactly what it could do or how well it could do." You push this button, you're going to get something.
| |
Guillermo Rauch | Yes.
| |
Shaan Puri | It could be phenomenal. It could be, you know, a little bit unpredictable in some ways. | |
Guillermo Rauch | But what I offer is that the successful AI products of the future are going to be an intersection of these two worlds: the software one. Parts of the product are still going to be highly valuable, if not zero. It would just be purely an autonomous AI.
But think about it. It's still writing Next.js code. It's still using UIs that make it friendly for people. It still has a community of templates that you can click with one-click deploy. It's still banking on the ten years of investments in Vercel infrastructure and our partner infrastructure, like Supabase and Neon, and the databases that we bring in.
So, I think people are going to be able to do this in many other verticals. Video games is an obvious one to me because you will want parts of the game engine to be just like Unreal Engine is, right? You'll want something like Next.js but for video games that is under the hood.
Then, you want to open up video game creation to as many people as possible. You can start with a prompt, and you're going to be like, "I want to create something that is like Pokémon Red, but the art should resemble this, and it should be in three-dimensional space, and it has 10 missions." So, you start describing in English what you actually want.
| |
Shaan Puri | Is this possible today? Like, what is the tech? The tech could do this today.
| |
Guillermo Rauch | Oh yes, in fact, I'm like, "Why?"
| |
Shaan Puri | Doesn't someone have done this? | |
Guillermo Rauch | Please, like, where are y'all at?
So, I'll tell you some of the things that we've done to help facilitate this world. Using the game engine metaphor, you can think of V Zero as one video game that Vercel built. You can think of Vercel as a game engine.
There are going to be many, many other video games that people are going to create. If you have an idea for an AI agent, you can deploy it on the Vercel platform.
You can do V Zero for doctors, you can do V Zero for video games, you can do V Zero for lawyers. So, any vertical that you can imagine, we even templatized. We've open-sourced a lot of what makes V Zero so great so that entrepreneurs can come in and say, "Look, I see an opportunity for AI to disrupt this space."
| |
Shaan Puri | Let's take the video game example real quick. I think you may... did you make a Doom thing? Or did you port Doom? Or was that an AI-generated Doom?
| |
Guillermo Rauch | No, so that was the one you're talking about: my doom captcha thing. For context, I hate captcha. First of all, like captcha is...
| |
Shaan Puri | A thing nobody likes: CAPTCHA.
| |
Guillermo Rauch | Where you go to a website and it tells you, "Please tell me how many..." | |
Shaan Puri | Stupid, because you're like stop lights... the bike. | |
Guillermo Rauch | That's right, it's the bike or select the staircases.
| |
Shaan Puri | Yeah, right. | |
Guillermo Rauch | So, I created one which was instead of selecting staircases or stop lights, you have to kill three enemies in Doom.
Funny enough, the idea came over Christmas break. What I did is I went to V Zero and I said, "I want to create a CAPTCHA that looks exactly like Google CAPTCHA" because it needs to look familiar for people for the joke to work.
| |
Shaan Puri | Pops up. | |
Guillermo Rauch | Yeah, yes. And then I basically took advantage of the fact that **Doom** has been open-sourced. So, what I did is, and this is kind of like a maybe a little bit more advanced engineering here, but I took a WebAssembly version of **Doom** that can run inside the browser.
I prompted my way to basically spawn the user in a very specific level. This actually involved hacking the C codebase of the game and then setting it up so that it was a constrained version of the video game. You kill three players and you pass the captcha. So, it's almost like creative engineering, you could call it.
This is something I think is also going to be big in the world. You could argue that someone could create an entire platform for just creative coding—AI for creative coding—where the next generation of artists are going to be playing in this dynamic medium. They're going to be offering up things that are highly interactive.
So, you could create the V0 for interactive art. It's kind of meta, right? But I started with V0 and I created this thing, and it went viral. There are like three or four news articles written about it. But it literally took a couple of hours of prompting. I've been an engineer for like, you know, a couple of decades now, and I no longer write code; I only prompt. | |
Shaan Puri | Last one on this video game thing because it kind of blows my mind that that's possible. I would have assumed it's not possible yet. Can it only make very simple Flappy Bird-style games, or can it make something like Fortnite? You know, where are we at today in terms of what's actually possible?
| |
Guillermo Rauch | Yeah, so there's two levels to this. If you go to V Zero today, you could prompt and create a one-off video game. But the next level is that I think there are going to be entrepreneurs who will create the next big AI platforms and deploy them to Vercel.
So, it depends on where you want to play. If you just want to create something that looks like Fortnite, you could do that today. You could just prompt; it's possible.
Really, yeah. I mean, I think you'll probably go down a journey similar to what JavaScript went through. You're going to be able to get something basic going, and it might take you the next ten years to perfect it to the level of what Fortnite is today.
But this is the beauty of things like V Zero. Like, anybody can cook. You can start, you can get it out there, kind of like DoorDash was six PDFs and a website. You can get the V Zero of Fortnite out into the world.
But I think there are also going to be ambitious people who maybe have skills that are more on the game engine side, kind of like I created Next.js, and can say, "Look, there's an opportunity to create a framework that works really well with LLMs that enables broad, massive-scale game creation."
The things that you can facilitate in that world are, for example, for a game to be successful, it needs really high-quality textures and really cool art. So, you can start creating a platform that facilitates bringing the art in. That's why I mentioned you will need software one.
| |
Guillermo Rauch | Skills are needed to create a platform and establish the connective tissue that facilitates this highly opinionated workflow. Creating video games involves a lot of art and emergent ideas. The game creator has to come in with a concept, but there are also many predictable elements.
For example, there are different kinds of games, such as the 2D platformer. What does a 2D platformer need? Well, all of the runtime infrastructure and the components that make a game work already exist on the internet. The AI is perfectly capable of orchestrating these elements.
What a potential game creator would need is an easy way to generate game assets. If I were an entrepreneur in this space, I would connect users to various image generation models. When you come to this game creation studio, I would guide you by showing you all the things you'll need and the necessary integrations.
Sound creation is another good example. Eleven Labs allows you to create sounds with language models. In fact, I've coded a few things that required sound, and instead of searching for open-source sound effects, I just went to Eleven Labs and prompted for what I needed. I can do that because I am familiar with everything that exists in the AI world, including Eleven Labs. | |
Shaan Puri | Creative engineering. | |
Guillermo Rauch | Exactly. But imagine a would-be game creator who doesn't know that. They just want to go to a platform that has already built-in sound creation with AI. So, behind the scenes, you can sort of plug this in.
To summarize, I think there are all of these permutations of technologies that already exist, making new platforms possible. Entrepreneurs don't need to train foundation models; they just need to go in and put the pieces together into opinionated workflows.
The things that are becoming obvious to people today are AI for legal, AI for developers, and I think those are kind of like the zero to one. It's what feels very immediate and emergent. I sometimes call it unbundling.
When Chargebee came out, people were like, "I can ask you for a draft of an NDA." So, people said, "You know what? I can take that slice of an idea and turn it into a legal platform."
What we realized was that you can go and ask it for web UI. ChargeGPT was quite good at outputting React, HTML, Next.js, and so we went and said, "Oh, this can be V0. This can be a whole platform for web development."
So, what I challenge people to think about is: what are those clusters of queries that people are going to ChatGPT for that can become entire platforms? I'm sure there are a lot of people who are going to these things and saying, "I want to create a video game," but this is just one of the many ideas that I shared on that thread.
| |
Shaan Puri | Right, so let's go back to this Typeform idea. This seemed like a super simple one, which I like because you don't need to build something world-changing, blah blah blah, just to be able to...
| |
Guillermo Rauch | To yeah, yeah. So that one arose from the fact that if you look at why Typeform has been so successful, it's an interface innovation. It gives you one question at a time. It feels friendly. I mean, I hate responding to surveys, but the most palatable way you can make it for me is if you give me small chunks. | |
Shaan Puri | One of the small multiple choice questions versus, like, "Boom! I'm going to have to answer this whole long page."
| |
Guillermo Rauch | Yeah, here's an IRS form. Please fill it out. In fact, the IRS would probably have way higher completion rates and people paying taxes on time if they created a better user interface. Essentially, this is the use of TurboTax, by the way.
| |
Shaan Puri | TurboTax is actually a phenomenal user experience. Mhm. It's like Typeform; it's one question at a time. They don't ask you things that you don't know the answers to, which is the problem with taxes. So smart! They don't... I mean, like, what do you owe? No, they just... and this.
| |
Guillermo Rauch | Is one of the golden rules of onboarding, right? Forget about forms. Onboarding should focus on one thing: remove all distractions. Remove all the links that might take me out of the flow.
I give people this little hack sometimes: if you're in a flow where you want people to complete a task, why are you making the logo clickable? And why are there like six footer links taking me to the *fucking* founding story of the company?
| |
Shaan Puri | Right. | |
Guillermo Rauch | Why a distracted person? So, I type "form kudos" like they nailed that. But I think there might be an opportunity on two levels. One is, I keep using this formula: if I need to create a form and very quickly send it to you, I could have the B zero for form creation where I prompt my way to tune the form.
| |
Shaan Puri | Yeah, because it could even come up.
| |
Guillermo Rauch | With the questions for you, exactly.
| |
Shaan Puri | I'm coming up with a plan for the bachelor party. I'm trying to get my friends' input on what dates work for them, so I'm sending out a quick survey.
| |
Guillermo Rauch | To my friends, in fact, I'm sure someone will go to V Zero listening to this and start creating the AI form creator.
| |
Shaan Puri | Right. | |
Guillermo Rauch | I think it's a really good idea.
| |
Shaan Puri | And by the way, I think there's an add-on to this. I had a similar idea once, but I was thinking we own this business called Somewhere.com.
| |
Guillermo Rauch | Great domain name.
| |
Shaan Puri | Yeah, we paid a great price for this domain name. I'm not sure about that one, but you know, you find talent in LATAM or South Africa, or just like your story, right? Talent is everywhere; opportunity is not always everywhere.
So, when the website works today, you land and you click, "Okay, I want to start hiring." Then there's a long form, and it's all there. It's a bad experience. It asks for your name, what you're hiring for, your budget, and whether you need full-time or part-time help. Then it says, "Great! Now book a call," and then you're going to talk to the sales guy.
I'm like, you know, with AI, what this should really do is you should land and be like, "Hey, what are you hiring for? What do you need?" Then you say, "I need a designer." Cool! We got plenty of designers. We actually just hired designers for Company X, Y, Z.
Tell me, are you looking for a graphic designer or... blah blah blah? Totally! Yeah, what are they going to do? You kind of just quickly tell it, and it says, "You know, we recently placed somebody like this. Here's a candidate profile pulled from our system."
Then it asks, "Would this be the type of person that would fit what you're looking for? They cost this much; that's pretty affordable to hire someone of this level of talent." And you're like, "Yeah, yeah, that would be great! More like that."
It's like, "Awesome! What's your email? I'll send it to you." Or like, it's a salesperson. It's not a phone number, and it's a salesperson that does what all salespeople do. They ask questions, they follow up, they note what you say, they respond intelligently, and they follow up where they need. They disqualify you if you're not a good fit, and if you are a fit, they're basically giving you bits of proof and promise along the way to get you to say yes.
I can't believe this doesn't exist.
| |
Guillermo Rauch | It's a more dynamic interface, right? So, you said something really interesting, which is, I think when people have ideas, it's like they're staring into a vector space. They know that there's something there that they want to do.
This is for everything. You may have an idea for a survey that you want to send to your customers, and you're like, "Yeah, I know that it's mostly about getting product feedback and how happy they are and what they do for work." But you might be forgetting that there's a very important question that people who do these surveys typically ask, and you just don't know about it.
This is why an AI-first type form would make so much sense. Because when you prompt it, "I wanted this form for this thing," it'll know.
| |
Shaan Puri | Right. | |
Guillermo Rauch | And this even goes back to... | |
Shaan Puri | And it'll, by the way, summarize all the results for you at the end. Because it'll be like, you know, otherwise it's a full manual step I have to do. Yes, great! We got 300 responses. Alright, I gotta go through those.
| |
Guillermo Rauch | And this is why AI is going to disrupt everything. Because of what you just talked about, I can also bring AI to the results process.
My other point was that you can even bring AI to the submission process. Instead of being rigid and making you select between 20 options and then pressing "other," there might be innovations. Maybe it's a purely conversational approach, or a hybrid conversational model, and what we call generative UI, which is that it can generate responses on the fly.
Exactly, it chooses the right format to answer a question in real-time. It might also learn something else that humans do poorly: when asked, "Where did you hear from us?" we often write down outdated platforms like AOL and Google.
You're like, "Wait, does anyone use this anymore?" That should also be given to the AI to choose. For example, if a customer is coming from Argentina, the AI should recognize that no one uses that name on their link in Argentina.
| |
Shaan Puri | and you | |
Guillermo Rauch | I know it from GIP headers, but this is a... | |
Shaan Puri | **Idea** | |
Guillermo Rauch | There were a lot of people nitpicking my idea in the thread, which is like, "This is a freaking tweet storm." I'm not describing the entire company, right?
They were saying, "No, but UI is still better because so and so." I was like, "Sure, but again, it doesn't mean that the UI is as rigid as it is today and that the eight choices of checkboxes are rigid as well."
I remember there was this company called Wufoo in Y Combinator batch one. They had created a beautiful form builder, and the form builder was all drag and drop. You had to select the type of response, and all that stuff is going to go away.
In fact, that would give you a rule of thumb: if drag and drop is involved as a primary interaction mechanism, it's probably ripe for disruption. Because no one wants to actually drag and drop stuff; you just want to say, "This is my idea, just build it."
| |
Shaan Puri | Right, right. Yeah, that's like a tell. It's like a poker tactic.
| |
Guillermo Rauch | Drag and drop is like the maybe to use the software one.
O and two. O metaphor. Drag and drop was making one. O more palatable and accessible to more people. Visual coding, visual programming was that as well.
How can we make one. Zero more accessible? Well, we invented dragging stuff and showing it in two-dimensional space.
| |
Shaan Puri | Let's do some more. So, you had one that was called "AI Camera." What's the AI Camera idea?
| |
Guillermo Rauch | This was back to like embracing the wrapper. Embrace the fact that models are **phenomenal**. There are a few hyper-online people, like me, that know all of them or try to know all of them. For example, Deepgram v3 just landed on Hockey Bay and whatever. You're paying attention to that.
| |
Shaan Puri | That level of depth.
| |
Guillermo Rauch | Yes, but the average consumer just wants to take awesome photos.
This came up because we were at a very big meeting for a bank that I wanted to use for a sale in New York. The champion on the bank side was like, "Let's take a photo with the beautiful New York backdrop."
It was so awkward that we tried to take the photo 20 times. It's hard to take a photo with, like, where's the sun in the background? We want New York to be visible, but we also want our faces to be visible. And then, of course, someone blinks during the photo.
| |
Shaan Puri | If there are men involved in a photo, it's a terrible photo.
| |
Guillermo Rauch | Yes, it's just a general rule.
| |
Shaan Puri | Taking it... standing in it... posing in it. Then we don't know how to do any of it.
| |
Guillermo Rauch | So, think of the input from the shutter. Think of the click of the camera. This is a software camera, like a camera app, as the input into the prompt rather than the output.
If you're extremely good at taking photos, it could just be the output, or maybe it's lightly tuned and filtered, and whatever, and it becomes the output. But maybe just embrace the fact that it's an idea to give the AI, right? The AI will know that if that photo was taken in that place, the goal is to obviously show people smiling. No one should be blinking, the backdrop needs to look amazing, and you know, maybe it actually gives you five permutations of what good looks like.
We know the shades, the lighting, and maybe it removes objects for you. I also mentioned that there are software 1.0 techniques and software 2.0 techniques to embed into everything. You shouldn't believe that you need the perfect model that'll make that photo perfect.
Either you can give people a workflow. What I had imagined at the time was that it's going to look like Instagram because I love how Instagram allows you to take the photo and then pick a filter. It's called like "San Ramon" filter and whatever, like "Oakland" and "hipster" or whatever.
So, imagine it's giving you permutations, but also maybe it gives you the tools to select something. I mean, this is just purely brainstorming now, by the way. You had me feel like full-on brainstorming.
There's a model that is really good for like what Apple uses for removing objects from the scene. So maybe by the time it gives you the produced photos, all of the objects are already movable. What happens a lot of the time is someone blinks, but also there's an object that you don't want in the scene, like your baby was crawling in there, or your dog pooped, and you want to take that out.
| |
Shaan Puri | You have the best one, Argentina. What's the name of that beautiful waterfall? Like, the craziest waterfall place... Falls? It was a falls.
| |
Guillermo Rauch | Take out the tourists. | |
Shaan Puri | I go there with my fiancée, and we take this photo under the waterfall. We're kissing when somebody takes a photo, and there's this dude in the background. He's got his shirt off and he's taking a photo for the boys. It kind of ruins the photo.
My wife thinks it's so funny that I want to frame it, but she's been trying to find a Photoshop expert to get rid of this guy.
| |
Guillermo Rauch | % | |
Shaan Puri | And you're right. That would be like her dream: the magic camera that says, "Let me guess, you want this guy out of here," and we have the model that is good at detection and then good at removal automatically.
| |
Guillermo Rauch | Yeah, and this is all possible today. You know, your job will mostly be to combine models, create pipelines of models, and even do prompting, right?
| |
Shaan Puri | Do you think that Apple will just make this kind of default in the camera?
| |
Guillermo Rauch | I mean, Apple could have shipped Instagram. Apple could have shipped so many things. Like, Apple could also ship that annoying popover of facial effects on Zoom, but it still hasn't been fixed. It's still broken.
| |
Shaan Puri | When the balloons come up.
| |
Guillermo Rauch | So, I know I'm in a post-worrying-about-what-Apple-does world. I'm actually more worried about their constraints on developer freedom, how they tax you, and only let you run one browser engine. That's kind of my mental model.
Apple is like in the IBM phase of "let's preserve what we have at all costs" and litigate. They're almost becoming like Oracle of our generation. It's purely all about terms of service, and one prohibits this.
Hopefully, I'm saying this so that the company becomes better in the interest of open public feedback. Tim Sweeney from Unreal and Epic Games just had a similar comment on the Lex Friedman podcast. I think there needs to be a world where developers can just ship. That's the main idea of Vercel, and Apple has sort of been constraining that.
I'll tell you, there are so many cameras that you can ship to the phone that people love. Right? So, there is Halide camera, I think it's called. I don't know how it's pronounced. There is obviously Instagram, which is a camera, right? So, Snapchat is a camera.
So yeah, and also, but it's also time...
| |
Shaan Puri | Like, you know, these things come in waves where about every seven, eight, or ten years the window reopens on these kinds of things. So if you just look at Instagram and Snapchat, when they came out, it was sort of like this eight-year gap after Facebook came out.
| |
Guillermo Rauch | Yeah. | |
Shaan Puri | And like, there was enough new stuff—either new social norms, like people taking pictures everywhere, or new technology, like smartphones—that all of a sudden, there was an opportunity for those. It's now been another ten years post.
| |
Guillermo Rauch | Yeah. | |
Shaan Puri | Snapchat and Instagram are for somebody to build a better camera.
| |
Guillermo Rauch | And there's a wedge through utility. Like, you can build a better camera. You can build a better... oh, just import your photo and start editing.
The exact approach... I mean, who knows how this is going to happen, but it has to happen. It has to happen that I can take great photos even if I didn't capture everything perfectly. The AI is already there; just make it happen, people.
| |
Shaan Puri | Yeah, you could even actually just do the camera roll part, right? Because, you know, Google Photos and all these apps try to do this like, "Hey, we made a memory." My mom loves these. It's like, "Oh, I forgot about that." It's stitched together and they put corny music on top. There's no cool factor.
| |
Guillermo Rauch | To those. | |
Shaan Puri | Those are all pretty. | |
Guillermo Rauch | Totally. | |
Shaan Puri | Bad, but it's your photos and your memories, so it's still good enough. Imagine if you had taste and you did it right. Imagine if you had Kevin Systrom and you realized, "Oh, there's a gold mine on people's camera rolls that I could just be generating. I could just be mining that and creating actually good content."
| |
Guillermo Rauch | Yeah, there are so many thoughts that come to mind. One is, **brilliant things happen when people focus and obsess over a problem domain**.
The operating system makers have so many irons in the fire, and there has to be someone who just loves to take great photos and wants to democratize that with everybody else, right? So that's kind of how I think about it.
There are so many angles: there is memory, video creation, there is the photo aspect, and there are some aspects of shareability of the photos.
So who knows? I mean, I think my prediction is more so around something amazing is going to happen in this space, more than what the specific thing is.
| |
Shaan Puri | Alright, I want to do a couple of your other ideas.
Oh, this one was clever: **absurdly smart auto-complete**. I didn't know when you said that, I was like, "Auto-complete? Who cares?" Then I saw your kind of brief description, and I was like, "That's actually brilliant."
Can you explain what this idea is?
| |
Guillermo Rauch | Yeah, I think when it started to become obvious that LLMs (Large Language Models) were going to transform software engineering, it was when we started typing code into a code editor, and it just suggested stuff based on the whole project—the whole sort of corpus of code that has existed in humanity.
The next wrinkle was that things get better the richer the context. So, if you only give the LLM one line of code and nothing else, it'll still produce something useful, and it'll blow people's minds. I think we're in a time in humanity when things blow our minds for like a week, and then we get immediately bored and used to it.
But the next wrinkle was: what if you put the content of your clipboard into the prompt? Because developers typically have what they just saw in their clipboard, as they intend to search for it or make a mutation. So, what's in the clipboard is likely to be kind of what's in your mind. Could... | |
Shaan Puri | Help. Yeah.
| |
Guillermo Rauch | Right, so auto completions get better, and they improve with more context. They also utilize smarter models.
With search, when I am in a different file, I'm likely to be writing code that is related to the dependency of this other file.
So, long story short, things get better with better context. Auto completion is sort of the first manifestation of how LLMs (Large Language Models) and AIs can enhance your cognitive ability.
| |
Shaan Puri | Yeah, I mean, it's just-in-time expertise. Yes, right? Like, in line, right while I'm writing it. Yes, give me the perfect thing right there without me having to leave, go ask, or think of a question. Just give me a suggestion right there.
So now, what's the... that worked in code you're saying? What about the rest?
| |
Guillermo Rauch | Yeah, exactly. I get frustrated with how bad spell checking is. If you look at macOS's spell checker, it's gotten better over the years, but it just doesn't know the things that you literally saw ten seconds ago.
For example, someone tells you, "Hey, can you send me the B zero?" in Slack, and then you go to an email and start writing "B zero." Apple goes, "I think you meant 'via'." It comes up with a new word, and I'm like, "No, no! I literally just was talking about this. How could I possibly want to not write that down?"
So that's where the inspiration came for it. I think there are a lot of ways to go about this. One is to enhance the operating system. It's a non-trivial task, but I think this general idea applies to so many things. We forget that the LLMs (large language models) are here. The context is not evenly distributed, meaning if you just put the right things into the prompt, magic will happen without changing the actual sort of engine of intelligence. | |
Shaan Puri | And so that example you gave of like the V0.1 sounds minor, like, "Oh, who cares? You just push backspace twice and just fix it." But what you're saying is, if the AI knows what I'm talking about, what I'm thinking about, what I'm working on, and what we're doing, then in the same way that in code it auto-suggests the right code because it knows about my whole, totally multiple files and the projects, it'll do the same when I write that email. It won't just give me the generic thing; it knows how I talk, it knows what we were talking about, and it knows what the plan was that I'm trying to relate to this guy.
| |
Guillermo Rauch | It's like the *Black Mirror* episode "The Entire History of You."
| |
Shaan Puri | Is that the one where the camera's like... is like your memory is an external memory, basically?
| |
Guillermo Rauch | Yeah, the other thing that's really interesting is that these systems are all about **next token prediction**. Right now, we're not fully exploiting that because we're not putting in the sequence of everything we're doing and everything we're thinking about.
Every time you go from app to app, that context is getting lost. We typically tend to work sequentially. What happens is I read an email that is about a problem, and then I go to another app. I'm likely going to discuss that problem or I'm going to try to look for the person that is an expert in that problem.
So, the other way to go about this is: how can you ingest this series of apps, integrations, and systems that people use to do their work and connect the dots?
| |
Shaan Puri | There's one more that you have here. You said "more granular vs. zero." So this is almost like, "Hey, come disrupt us."
| |
Guillermo Rauch | Yeah, it's what I talked about with the game engine, right? Just like Chargebee got broken down into individual components, we have a big effort at Vercel. We want to be the platform of platforms. We want the next Shopify to be born on Vercel.
What I anticipate will happen is that if an entrepreneur says, "I want to make it easy for people to sell online," they're probably going to start with AI. They're not going to build the same Shopify that exists today. They're going to create something that starts with intelligence.
It starts with maybe importing a photo of the product that you want to sell. Maybe it starts with a prompt of what you want your store to be. Perhaps it's so smart that if it's a physical store that already exists, it knows everything about it.
| |
Shaan Puri | It’s right. | |
Guillermo Rauch | It's just like one button: create the store, and it ingests all the products, SKUs, categories, and whatever.
So, similar to the game engine thought experiment that we went through, that would be a more granular V Zero. It's like, okay, V Zero can create anything, but the reason that this would work out is the same reason that Shopify worked out.
There are things that are very general, like AWS and Vercel. You know, Vercel serves like AWS on steroids in a way. We're making it so easy for people, but it's still a broad platform on top of which any idea can be deployed.
I really believe that AI will transform everything. It'll transform website building, e-commerce, and form building. A lot of the exercises that I do is that, look, there's going to be V Zero for legal. Our general counsel uses a tool called GC.AI. GC.AI is essentially V Zero for lawyers, and it was built on the Vercel platform.
And the other wrinkle...
| |
Shaan Puri | And it's... | |
Guillermo Rauch | Used to build their website only, or a government representative draft. So, like, you can prompt your way to, you know, saying, "I need this contract between these two parties," or "I need to review a contract," and "I need to import documents." That's kind of what I mean by the more granular visitors. It's more like a theme but a specific idea. | |
Shaan Puri | One of the other things I wanted to ask you about is something I brought for you. I thought this was incredible. This is a piece of, you know, internet history.
So, explain what I just handed you and the backstory of this.
| |
Guillermo Rauch | Thanks for printing this out because I use this example so much as like the magic of Silicon Valley. I ended up at this party, you know, the classic like it could have been an office warming party or something along those lines, or like a meetup. People were having conversations, passing out drinks, and I met this gentleman named Brian Armstrong.
He did something that I find myself doing a lot. I walk up to people not with networking agendas or random ideas; I walk up to people with content. | |
Shaan Puri | Okay. | |
Guillermo Rauch | I want to show them something.
| |
Shaan Puri | Okay.
| |
Guillermo Rauch | I want to show them an app. I walk up to people and strangers.
| |
Shaan Puri | What do you mean? What are you talking about?
| |
Guillermo Rauch | Well, people in the context of networking on the internet.
| |
Shaan Puri | Like I... | |
Guillermo Rauch | Might have been in a van, whatever, but I could do it with an Uber driver. I have no limits; I take no prisoners.
So, he walked up to me and he's like, "Here's a... I'm working on Coinbase. It's like a bank for digital currency, and I'm building an app." He shows me the app and then he goes, "If you install the app, I'll send you a Bitcoin."
Just think about how crazy it is! If you install an app... like, first of all, twist my arm! I love trying out new things, right? And I loved Bitcoin at that time. I'll give you a hundred thousand dollars or whatever. What year was this?
| |
Shaan Puri | Right now, the date is on this email. What year was this that we're talking about?
| |
Guillermo Rauch | Yeah, so, 11/09/2012. It's the magic of the Bay Area because you can just walk up... it's on both sides. You can just walk up to people and show them things. You can see their reactions, you can get their feedback, etcetera, etcetera. And on my side...
| |
Shaan Puri | And there are some people on the street that will show you some things you didn't ask for too. Because that's just another part of San Francisco.
| |
Guillermo Rauch | Yeah, for sure. It comes with a lot of diversity. But on my side, you know, it's like, well, you can be there on the receiving end of new ideas, new opportunities. You can invest. Like, if I had been an investor at the time, I would have been like, "Hey, this is Brian, guys. He's really smart and he's hustling hard. Digital currency? Might as well give it a shot!" Right? So, yeah, things are serious. | |
Shaan Puri | He sent you this email. He sent you the Bitcoin.
| |
Guillermo Rauch | So, they put the thing in there. It says it's worth $10.81.
| |
Shaan Puri | $103,000 for the app install as of this morning.
| |
Guillermo Rauch | And then I actually ended up following Coinbase for many years. At the time, I think I was poor or I had just begun; I can't remember exactly when I sold my company, but I couldn't angel invest most likely at that time.
That was the other thing that I started doing when I sold my first company. I put it all back into the game, which is kind of crazy. In fact, I talked to this guy who I really respected in the JavaScript community. He was also starting to do angel investments, and he said, "Well, the way that I treat Bitcoin is like an angel investment."
I put in a $20,000 check into Bitcoin, and I remember when I sold my company, I did that. It was probably also because of the serendipity of having Coinbase and the infrastructure that was nascent. I had been exposed to the magic of Silicon Valley, where you can still do these things; you can just make things happen.
| |
Shaan Puri | Did it also help that you were, you know, born and raised in Argentina? It seemed like maybe you questioned currency more than the average bear, right? Like, yeah.
| |
Guillermo Rauch | A concrete memory from when I was a young child that had everyone in Argentina stressed out was... and people can fact-check us and look this up, but it was like, I think we had three presidents in three days.
One president had almost, like, I think he rage-quit because there was an economic and financial meltdown. I think this was around 2001. The next guy comes in, and the vice president becomes president, something along these lines. He's like, "Oh cool, I'll try to fix it." Two weeks in, he's out too.
Then someone else comes in, and I’m not even sure if he was literally next in line. Another guy comes in, and I think this is the guy that I'm pretty sure is the third one. He goes on national TV, and they do the whole thing where they interrupt all of the ongoing channels. The president comes in, addressing the financial turmoil.
He says, "Look, there's a lot of noise about how your dollar savings will get lost, and they might get converted into pesos at a non-beneficial rate." He said, "Do not worry, Argentinians." This is on national TV, fully synchronized across every screen in the country. He said, "Do not worry, your dollars are safe. If you deposited dollars, this is the exact quote: if you deposited dollars, you will receive dollars. If you deposited Argentine pesos, you will receive Argentine pesos."
Literally a week later, it didn't happen. The dollars got converted into pesos, and then the currency lost its value. So basically, your money was stolen from you. The banks were in cahoots with the government to make these transactions happen.
Your money is effectively just like when people joke... I take it personally to see Michael Jordan. People joke, "Oh, Bitcoin is just a database; we could replace it with Postgres and we would get a lot more throughput." I say, "Well, not really." What we had in Argentina was Postgres, or maybe worse, maybe it was Excel or something. The government literally did go in and, in the currency column, said, "Select all, convert." | |
Shaan Puri | Right, like... | |
Guillermo Rauch | That is the database that is not immutable. I became extremely interested in Bitcoin because I have this concrete memory of my dad screaming at the screen, saying, "These people are so corrupt!"
Another thing that happened was all these protests from people who had large dollar savings. It was terrible because Argentina has been guilty of this before, many times. If you're rich, that's frowned upon; it's terrible. You probably got rich by screwing someone over. That is how the culture was largely configured.
So, it was very hard to empathize with these protests. People would literally be outside of the banks, protesting that their savings got stolen. However, the way it was manipulated by the media was, "Are you going to empathize with that rich guy? Oh, poor him, he's complaining about his huge dollar savings." This added even more insult to the injury, as there was no empathy for people losing their money.
In this context, Bitcoin seems so obvious. We need a globally distributed database that is immutable and has extreme security guarantees. This is your life; these are the things you might leave for your kids. This is everything you've worked your entire life for. You cannot trust any given actor. You cannot trust the government, the banks, your friends—anyone. You need to have cryptographic certainty. You can only trust math and the universe, which is kind of like the inherent properties of it.
You trust the universe in terms of the energetic demands, on top of which Bitcoin banks that it's so hard to mint a block. It's also so hard to manipulate and conduct cyber attacks. You're banking on cryptographic verification. If you want, you can be a node and verify the blockchain yourself with software that you run, as opposed to trusting the world and asking, "Is this legit? Is this the chain I should be looking at?"
This is why I've always been somewhat unsympathetic to non-proof-of-work systems. They create uncertainty about what the right chain to be looking at is. In that category fall Ethereum and a few others. | |
Shaan Puri | Two questions for you. One, given that you went through that, do you kind of denominate yourself in Bitcoin? Do you put a huge percentage of your own net worth in Bitcoin? Or how have you decided to do that?
| |
Guillermo Rauch | I put my entire net worth into Vercel. Having said all of this, I always think about the mental model that Warren Buffett has offered. He said, "If I could have a cube of all the gold on the planet, I would look at it and say, 'Oh, it's this shiny block of gold.' Or I could have a cube of all of the productive farmland in the U.S. and all of its companies. What cube would I rather own?"
But it is a false dichotomy. I want to own both. I want Bitcoin, and I want assets in productive assets that are going to grow over time, which I also want to support. So, I like the idea of, by far, first and foremost, placing a bet on myself in Vercel. That's kind of my primary net worth.
However, if Bitcoin is not some percentage of my net worth, I would be really worried because I would not have that rock-solid foundation that I missed when I was in Argentina. It didn't feel like I had access to something that could be so reliable and trustworthy.
So, I am a fan of a potential future in which everything is denominated under the hood. Everything is rebased on top of this system. That would be really cool.
| |
Shaan Puri | I, so when I do these podcasts, like the way people think that this is the... "Oh, how did it go?" and I'm like, "It's good or bad." Actually, for me, I get the win far before we sit down. Because when I know you're coming on, I go down a rabbit hole of Guillermo. I read your old stuff and I learn from you. That's where I get... I'm in this for the wisdom.
So I'm looking for the golden nuggets, the insights, the wisdom, the frameworks that you use, the stories that inspire me or that I can remember or tie to my own life. By the time I sit down here, this is all gravy. Now I'm here just having chicken nuggets. I need the lunch.
But one of the things that I saw in your old blog, I think it was like 2016 or something, it was a newspaper clipping. I forget who it was, if it was Edison or who it was, but they had this idea of the energy dollar.
| |
Guillermo Rauch | It's wild. I've never heard of this.
| |
Shaan Puri | I guess back in the day, was it Edison? I don't remember who it was.
| |
Guillermo Rauch | It might have been Ford.
| |
Shaan Puri | Can't remember... yeah, it was like two of the luminaries of the time. It was like Ford and somebody else. They were talking about this concept of an energy dollar. They were like, "Hey, we need a currency that's based off of the production of joules of energy, you know, joules of electricity or work." And then that'll be a... | |
Guillermo Rauch | More sort of like a rock-solid occurrence. It attracts reality.
| |
Shaan Puri | That's what.
| |
Guillermo Rauch | We're going... it tracks the universe perfectly, right? It's also related to the Kardashev scale and our ability to capture energy from our nearby star, as well as the types of civilizations that we are.
The other reason I've been thinking about this more recently is that we're clearly entering a world in which energy can be transmuted into intelligence. I mean, it's already the case that I can do so much test time compute that any problem seems tractable with enough cycles of GPUs. The only limiting factor does seem to be our ability to harness energy.
So, the unit of wealth, or the store of wealth, has to be something that is rare, right? It has to be something that's provable. The idea of tracking our fundamental store of wealth through energy... I mean, I'm very intrigued by that.
| |
Shaan Puri | So, well, dude, thanks for coming on. I think what you've built is amazing. Your story is great. You're a big ball of energy, and some of these ideas are really, really good. I'll share more. | |
Guillermo Rauch | So, there's one more thing. There's a free ideas thread coming on X.
| |
Shaan Puri | So, okay, is the sequel going to be like a **shitty movie sequel** where, you know, it just doesn't look...? | |
Guillermo Rauch | Up at first... oh, but I did set a high bar. I don't want to brag, obviously.
| |
Shaan Puri | When that comes out, can we do brainstorm two on here where we riff on it? Because that's what we do here.
So, alright man, thank you so much.
|