Guillermo Rauch. Vercel.

Guillermo Rauch:

Get in. We're rebuilding the whole freaking Internet. Coding is a misnomer. Everything is code. A slide deck is code. A document is code. Excel spreadsheets is code. If I DM you, it probably meant that my agent couldn't figure it out. No one at Vercel has the right to block things. The spice must flow, the tokens must flow. The products have to be shipped. You don't get to relax. I'm in Ibiza. Send me the email and when I get time to review it, your product goes out. I refuse to have an AI as a friend. Human empathy and emotion and connection is ours. Overrated startup advice. Ambition, to be honest.

Nakul Mandan:

You have five kids, a sixth one on the way. You're very obsessed with fitness. You're at the gym every morning. Where is all this energy coming from?

Guillermo Rauch:

You know, I remember my early days in SF, before I was married. I think the level of discipline that I had my entire life was like so well curated for learning and working. It was Monday through Monday. The number of hours put on top of that, I think I joke like 10,000 hours? 996? That's rookie numbers. What are we talking about?

Nakul Mandan:

For the last decade, if you built on the Internet, you probably touched something Guillermo Rauch made. Next.js is the most popular React framework in the world. And Vercel, his $9.3 billion company with more than $500 million of revenue run rate, powers the front end for OpenAI, Stripe and Nike. A kid from Argentina who became one of the great infrastructure builders of the modern web. Now he's doing something even bigger, giving the power of building to everyone. With v0, you describe what you want in plain English and working software appears. But AI is not just changing who builds software. It's changing how companies get built and how society will function. Today we get into what happens to talent, to companies and to builders, when everyone can build. Knuckle up. Guillermo, welcome to the show.

Guillermo Rauch:

Thanks for having me.

Nakul Mandan:

So in September 2024, after Paul Graham's Founder Mode essay, you put out your own idea, which is Recursive Founder Mode. And your main point, if I understood correctly, was that Founder Mode doesn't scale. Recursive Founder Mode is where it needs to be. What do you mean by Recursive Founder Mode?

Guillermo Rauch:

Yeah, Vercel has been in this journey towards becoming a multi-product organization. So we started out hosting frontends. Mostly front-end sites and web applications. And then we started evolving into hosting backends, hosting agents, helping people with the whole software development lifecycle, or we could call it like the agentic software development lifecycle. And so in this journey, I've had to expand the product portfolio, grow the team, and I totally identified with the idea of founder mode because you need someone that has this pioneering spirit, like break down tape, push through, be the voice of the customer all the time and just feel extreme pride over everything, every detail, like make sure everything is perfect. But it clearly doesn't scale, especially as you think about an organization that's working on multiple products, multiple services, millions of customers. And looking back on things that have worked really well for us, And by this, I mean people come to me and they're like, man, how is this thing so good? Like yesterday someone asked me, why is the AI SDK so cool? How do they communicate so well on X? Why does everything seem to be so nicely polished? Immediately my mind went to the founder of this project, Lars Graml, who for me embodies that because he's an entrepreneur at heart. He joined Vercel and he was a seasoned engineer. And he came in with a very strong point of view of the world. He'd been working on this developer experience for making, uh, basically model agnostic applications in TypeScript. And so when he joined Vercel, I got a founder, a visionary, and someone that if given the right company setting can like, again, blast through walls and cut down tape and whatnot. The things that you would want a founder CEO to do is he's almost like the CEO of his own product. And over the years, we've been very fortunate to identify a handful of these people that what I've seen is that these are the people you want as you're scaling your company. You know, the founder, CEO can only go so far. You kind of want to build the machine that builds the machine. You know, software factories are in vogue these days. And I think of it as like, how do I set up this recursive software factory? I need this, you know, managers of these factories within the company. We've done it for Fluid Compute. We acquired a one-man company, a guy that had this, like, he was like totally full stack. His name is Tom Leonard and he would take care of the marketing, the comms, the product design, and he would write this low-level Rust-based runtime. It's like, oh, this could be a great person to add to our Fluid Compute offering.

Nakul Mandan:

I think the thing that resonated a lot with that essay, I mean, it went, it exploded on Twitter that day over the weekend and in the weeks to come. Was that the founder has to go deep in what matters. The founder has to be in the weeds of everything. How do you now ensure in this model that you're talking about, like the unit of ownership is at this founder of that product level? What do they own? Where do you go deep versus where do you force yourself to stay out so that somebody else feels that they have the flexibility?

Guillermo Rauch:

Yeah, I think what I can contribute is I set up the overall organizational framework. The values of the company. Obviously my executive team and I decide on like, you know, what are the KPIs, targets, north stars, but you need to create these pools of autonomy within the company because the best people in the world, they could be starting companies. They could be vibe coding 10 million ARR businesses and deploying them on Vercel And so you really have to create these pockets of autonomy. And so these people, I think, have control over, or input into and influence into the messaging, the marketing, and obviously using a platform under them. Namely the Vercel platform. A lot of people actually want to come to build products at Vercel because they, they know that they get to use Vercel. So funny enough, a lot of our new products are completely bootstrapped on top of the Vercel infrastructure. And so it's a little bit like, you know, we, we have millions of companies building on Vercel and now we also have like companies within Vercel.

Nakul Mandan:

Does that autonomy extend to them deciding who to hire or even ship against your judgment?

Guillermo Rauch:

So this is a good example is that, you know, I, I give enough rope to see like, you know, people pleasantly surprise me all the time with their innovations, discoveries, ways to communicate their customer obsession, right? Like I expect people to go out and talk to customers. So you're asking what is a defining characteristic of this, of this founder mode thing? At the end of the day, we're here to ship. We're here to drive urgency. We're here to make customers happy. So I, for me, like my love language is seeing someone within the company take full ownership of the end-to-end business. And I think AI is gonna force people into this because if you thought your job, your job was like a very narrow thing, like I write code in Rust. Well, models can also do that very well now. And so you have to, you have to expand your scope.

Nakul Mandan:

In this world, are you as CEO the Chief Quality Officer or Chief Taste Officer?

Guillermo Rauch:

I joke internally a lot, like I do a lot of QA and I put a lot of input into the autonomous systems of quality assurance. So my focus has really been into, and I think this is what should be going through everyone's minds, is like, how do I agentify myself? How do I agentify my skills? No pun intended. You know, one of the big ecosystem contributions that Vercel has made recently is skills.sh, which is this massive repository of skills. We invented it because we invented Next.js and obviously we have like a lot of the know-how of the world and how to build React applications. Those people work at Vercel. And my challenge was how can I grab any customer in the world, no matter how tech native, Maybe they, they're a healthcare company or financial services. How can I even the playing field? How can I make sure that their Next.js code is as good as my Next.js code? Enter AI Skills. So we created this repository where we said, by the way, it was so funny when we put that out, people were like, aren't you just open sourcing all of your know-how? Is that allowed? Is that like, are you just, are you outta your mind? And this is one of the moments where like founder mode is so important. Because my conviction is our value is the entire Vercel platform and the trust that we get from customers when we give them, when we open source and give away things like this. So skills is where, you know, I'm thinking a lot, okay, how can I, if I have a quality assurance skill today, I was chatting with one of our content engineers. I've spent a lot of time trying to perfect the way that I communicate succinctly with people. The reason that I love X, like Naval has said many times, is it's wonderful that it limits you to 280 characters. He and I actually were pretty outspoken at the time, like, why did you even go to 280? Why didn't you stay at 140? Right? Because it's this amazing, you know, compression and creative constraint that helps you. And so I put a lot of my work over the years in like, how can I communicate effectively with customers? And I coach a lot of folks when they join Vercel, you're not here to just ship technology into the void. You're here to help with storytelling. You're not here to just launch. You also need to land. People need to understand the thing that you're working on. And so nowadays I'm thinking, all right, I've learned all of these things. How can I transfer that to the internal Vercel agent that helps this factory continue to produce high-quality products, delight, and accelerate the business.

Nakul Mandan:

How do you then recruit for that talent? Like, let's talk about that. What are you looking for? What is the DNA you're looking for? Founder is a broad word. Like, are there skills you're looking for in interviews?

Guillermo Rauch:

Honestly, the requirements are changing weekly, right? And so what I look for a lot these days is, do you have this mindset, first of all? Some of the world still debates whether AI is here to stay or not. Here there's levels. Some people are still stuck in like, well, agents can only do a few things. And some people are like, no, I'm actually using agents to superpower my work and to ship great things. I still have this bias of like, no matter what the technology is, the implementation detail, I look for people that can ship great things. There is an element of proof of work that I look for when I recruit or when I reach out to people, but there's also this future-based mindset. Like, are you able to extrapolate and position yourself in the future? Where are things headed 6 months from now, 12 months from now? What are the requirements going to be? Because we are in the middle of an exponential. So I can't have people that, you know, show me like, oh yeah, the You know, I built this awesome developer platform at, I don't know, Uber, Lyft, or whatever. And then it's like, okay, can you actually show me how that becomes self-actualizing? What would you do today? What are you going to ship next week? That kind of thing.

Nakul Mandan:

Has your own interviewing changed in the world of AI?

Guillermo Rauch:

Tremendously.

Nakul Mandan:

Yeah, give us examples because this is the multi-billion dollar question for all of us.

Guillermo Rauch:

I ask people to build, first of all. You know, you can't be of two minds. I either go on X and say like, everybody can build and use v0 and like Vercel plugin and whatever. And like you can ship immediately. And I also have to eat my own dog food. So I actually have that expectation. Like I ask people to build stuff.

Nakul Mandan:

During the interview process.

Guillermo Rauch:

During the interview process. And everybody, by the way.

Nakul Mandan:

They can choose it or you give them something? Yeah.

Guillermo Rauch:

I, I get, some people just don't have a strong opinion on what to use. So they, they take our recommendation. You know, funny enough, we call this, work prompts. It's always like, we're, we're not hiring agents per se, but our expectation is like, hey, here's a work prompt. What can you, what can you do? Depending on the role, I think, you know, it might be something that we, we have the expectation that you just vibe code something and like you produce an output and artifact and how you talk through it. And I also always look for what is your perception of the details and what interesting side quests did you take? I've asked people to build certain things and they come back with really clever ideas for the platform itself. They can tell me what features I'm missing or where they found problems or where they got stuck. I had an awesome experience yesterday because someone delivered this beautiful output, but she told me, well, at this point I wasted an hour on this thing. That's an opportunity for us to fix. So she kind of came in with a point of view on what she wants to join to improve about the company and the platform. So I think the level of transparency and the availability of information today is unprecedented.

Nakul Mandan:

Do you go through this with even non-technical hires? So marketing, sales, customers, somebody who's the Chief Customer Officer, all of that.

Guillermo Rauch:

My marketing team knows that the future is them building agents. That produce the marketing artifacts. A common problem between in marketing organizations is can I beg, borrow, and steal engineering resources or agency resources so that I can ship an idea into the world? That problem doesn't exist at Vercel because you have the platform, you have the agents, and because you're expected to create that factory. You're not here to like file in Jira tickets. Hey, can someone help me with this like poster or this? I need to host a dinner with prospects. Can you help me with the creatives? No, like you solve that problem. And that's, I think that's where the bias towards founder mode comes is I'm hiring people that can solve problems, solve problems. You know, I have, I have 99 problems. Like if you're coming to me to like add to the problems list, Like I'm

Nakul Mandan:

So…but how have you scaled this in recruiting? So how many people are you today now?

Guillermo Rauch:

Nearly 800.

Nakul Mandan:

800. So I presume there's recruiting happening at Vercel now that doesn't touch you, meaning you are not part of the interview process.

Guillermo Rauch:

Oh, I'm heavily involved in every hire.

Nakul Mandan:

Every hire?

Guillermo Rauch:

Every hire.

Nakul Mandan:

You meet every hire?

Guillermo Rauch:

We have agents. I don't meet every hire, but we have agents that can be posted on the progress of all of the interview loops. We've put in a lot of my requirements. Into the different prompts that summarize the candidates. I spend a lot of time reviewing their work, reviewing who they are, going through their profile, the things that they've created. I ask questions. Sometimes I ask to have almost an asynchronous conversation. We care a lot about candidate experience. Even if we don't hire you, we think of you as a customer. So we want to balance moving really quickly, but also asking some questions or asking for revisions on some of the work that gets presented. I take these decisions like very seriously because hopefully people understand that this is a long-term project. This kind of work takes generations. So I take each hire so seriously because I'm gonna invest in you and because you expect, I expect you to stick around for a decade.

Nakul Mandan:

In our earlier conversation, we just said You're kind of the Chief Quality Officer. It sounds like you're also the Chief Talent Officer at the company.

Guillermo Rauch:

It's all quality, right? We want the highest quality people we can find.

Nakul Mandan:

But it sounds like you're very involved in every single hire, even to this date.

Guillermo Rauch:

And that's also a point of leverage, right? So I can, you know, the way that I think about it is I can give a lot of consistent product feedback. And in fact, I embrace the fact that we have one of our values is iterate to greatness.

Nakul Mandan:

ITG.

Guillermo Rauch:

No one at Vercel has the right to veto, sorry, to block things. I've been joking a lot about this Dune reference. The spice must flow, the tokens must flow, the products have to be shipped, customers need help. No one is here to say hold the production lines. Only security can do that. And the occasional executive saying like, but that's a veto. Most of the time it's a veto. Which means that the onus is on you to veto. You don't get to relax. I, I'm in Ibiza, send me the email, and when I get time to like review it, your product goes out. And so we have very, very few and very carefully designed choke points. And so that's where we introduce friction. But, um, you know, Malte, our CTO, and I spend a lot of time reviewing this machinery. Because at the end of the day, by the way, it's all engineering. One of my banger tweets recently was everything is code. A slide deck is code, a document is code. The reason that coding agents are driving most— we have this data from the AI Gateway. Most of the token volume in the world is coming from coding agents. Coding is a misnomer. Everything you do, Excel spreadsheets is code. It's just that it was so deceptive because we used to drag and drop columns and do the math and perhaps do all of this. And we didn't think it was code, but it's actually a series of tool calls and a little Python script. Everything's code, right? And so you can think of the streamlining of your organization and its operations and the agents that you design. You can apply that engineering mindset to it. And so Malte and I were always obsessed with where are the choke points? Where is it? Where, what is our throughput? We study, we ask our agent. Let's study the changelog rate, meaning how frequently are we communicating product improvements to customers. And so this is kind of how we keep that pace and keep the quality up.

Nakul Mandan:

So, you know, everything you're saying there then relies on exceptional people, founder mode type people themselves. But in a world where OpenAI and Anthropic are throwing ridiculous packages, seed funding is everywhere. I'm saying that as a VC. Why, if a founder can build things that they want to see in the world over a weekend and then distribute, why should they stay at Vercel? Why not go out and do their own thing?

Guillermo Rauch:

Founders are also people that have long-term commitments and long-term visions. I see this like revolving door between like Google and OpenAI and Anthropic, and I'm like, you know, that doesn't look good for anybody. So you're saying that the guy has spent 9 months, left, and now he's over at this other thing and I think the world is beginning to actualize, as Dario was saying, I'm a little worried that people are just in for money. It's like, yeah, and that's the problem, right? And I think one of the antidotes to this is to invest in people that are not necessarily like this, like they have like this big fame and Twitter followers and whatever. You can get there. You know, people have joined Vercel and have built careers like that. But, um, it's also about investing in the people that have high slope. And I spend a lot of my time identifying and investing in people that are willing to, obviously it just, it takes an insane amount of work. And again, the founder mentality comes there too. Like you have to be ready to like work really freaking hard. And this is something I'm, you know, over the years I've become more and more explicit to people. I, you know, in our interview process, I'm very explicit now. This is going to take at least a decade and we're going to work insane hours and we're going to have a lot of fun. So there's also the grind, slog culture and whatever. And, and It's fun. You have to calibrate for some people like myself find this really fun.

Nakul Mandan:

Let's say somebody is a high slope person You've invested a lot in them. 3 years in, 4 years in, so not 9 months, but 4 years in, they come to you and say, Guillermo, I need to leave. I'm compelled by this idea. Like I want to start a company. What happens?

Guillermo Rauch:

Yeah, I'm supportive of people starting companies. I try to suss it out also in the beginning. Like sometimes I can tell like this is a person's like anxious, like I was never a good employee in the sense that I actually would, I deliver a lot of value for the company that I worked for very quickly, but I'm a person that is just not meant to work well in sort of a hierarchical structure and whatever. There are people that actually satisfy that itch going back to like the recursive founder mode. They satisfy that itch within Vercel and I get this feedback from people. I got someone DMing me the other day saying like, this person was an entrepreneur in the past life. And he was like, I just love what you guys have been able to create for me. Uh, which is like, I almost designed my perfect life within Vercel. The reason that I'm selective and the reason that I spend a lot of time in the interviewing process is that you want, you don't want jitter in your organization.

Nakul Mandan:

Yeah, but partly my question was that can recursive founder mode survive in an era where shipping is fast and seed money is everywhere?

Guillermo Rauch:

It could arguably. Do better and explain why. So one of our quote unquote founders or product leads, the product that he's running within Vercel, he kind of tried to build it before and it didn't work unless it was part of a broader platform. And this has been true for a few of the acquisitions that we've made that have been really successful where you say, I have this awesome idea and maybe it doesn't do quite as well unless it gets presented as part of a broader platform. And this is overall a strong thesis that I, that I have around the future of software. I don't subscribe to the idea that there's a complete SaaS-pocalypse coming and whatnot, but what is true is that let's say that you just close a deal with a very large marketing agency. What they're moving from is they used to buy software for bespoke needs that they have. Now they've moved into a world where their IP becomes the agents and internal tools that they're building. And this is a kind of transition from SaaS to build your own. And it is quite competitive because the rigid static software off the shelf is nowhere as near as good as the one that they're coding and deploying on Vercel In a past life, they could have never thought about like, what? Like I'm gonna create a product that's competitive with an Adobe product just for my needs is kind of a crazy thought, right? And yet I'm hearing about it every single day. And what it's doing is it's making them more efficient, is adding to their IP, and it's simplifying their life because they went from procuring 10 or 20 pieces of software to just procuring one platform. And I do think that platforms are going to do extraordinarily well in this world. Maybe some of the bespoke pieces of software that you could put into that platform do not survive outside of it. And that's more of a reason for someone to come and say, hey, Another example, our feature flags offering, the entrepreneur that built that outside of Vercel, he was having some decent traction, but can you really come in? Like people expect this to be part of a broader offering. So now he's, he has one of our fastest growing products at Vercel and he's having a fantastic time because he's found a delivery. He's found a channel, he's found distribution, which is to your point, something that will be a bit of a choke point as software becomes easier to make.

Nakul Mandan:

One theme that I consistently read as I was preparing for this was that there's two things that pull at each other, I think. One is that this culture of iterate to greatness it seems like there's a lot of urgency at Vercel. On the other side, you're obsessed with taste. Do you see them pulling at each other or you think no, it's fine?

Guillermo Rauch:

In a naive way, yes, because are you gonna ship slop and then you iterate to greatness? I will say part of the hiring process, part of the culture, part of the platform that we give you sets the floor really high. So let me give you an another example of a fantastic quote unquote founder within Vercel. ShadCN this gentleman created the world's most popular UI component framework. When you ask an AI, build me an application, 9 times outta 10, he uses shadcn which incidentally is also his, his online nickname. What shadcn did is he invented a component system that raised the floor of literally the entire internet. This sidebar that he gives you, beautiful and extensible. The button, the dropdown, even the AI components, the reasoning bar and the shimmering text and all of this stuff. Off the shelf, beautiful. Within Vercel, we extended this component system into the Vercel Geist component system. So it's kind of hard to come into Vercel, be competent, be a high agency individual, have access to all of these primitives, and then like ship a POS into the world. And so I think the platform itself, Vercel is like a recursive self-improvement machine, right? The platform itself is constantly raising the floor. The agents, our design engineering team right now, and extremely excited about this, they were like, design engineering always weighs in on polish, the delight, the surprising details in the customer journey. But they know that to scale themselves, we can't just keep hiring a billion design engineers. And so now, right now they're working on the agent that gives you that review. That incorporates a lot of like, to your point about taste, it incorporates taste. I think people sometimes create this false dichotomy between AI and taste. You know what's tasteless is grabbing the model and having it output as it is. That's literally how you end up with purple gradients and highly separated uppercase text and a little green dot that says online out of the blue. Meta announced their new model today. And you know what else they did? They announced the skills. So one of the skills is, I think they call it like the de-slopifying or the anti-slop skill. It literally tells the model to like not be itself. It's like, don't do this, don't do this, don't do this. I think every company needs to take this very, very seriously because you can take a lot of that taste that you have, institutional best practices, et cetera, and turning them into skills and agents that'll give you massive massive upside.

Nakul Mandan:

Let's shift towards the heart of what's happening in the world today with AI. Uh, let's start with engineering. How should engineers think of what's happening to their profession? And it probably is a big identity shift to go from, I'm coding 24/7, I'm obsessed with it, I love this craft of building, to CodeGen.

Guillermo Rauch:

I have a lot of thoughts. Look, it's happened several times in my career. The gatekeeping of what is coding has happened many times. The first time I felt that was I knew that JavaScript was a faster way to ship and build software. Why? Because it sat on top of the web platform. And so shipping a website with JavaScript was infinitely faster than shipping a desktop application. The gatekeeping/coping at the time was like, pfft, JavaScript's a toy language. That's a joke. I have my Windows, GTK Qt, C++ code base for desktop applications. What are you doing with that? Get off my lawn. Then JavaScript started powering some of the best experiences on the planet. Gmail was made possible because it was a JavaScript-heavy application that moved desktop into the web browser. And so there's been this disruptions to the industry nonstop where you had to self-actualize. If you were the grumpy graybeard gatekeeping, I only do desktop software, you missed the web, you missed mobile. This happened with data centers. Oh, I'm the data center guy. I know how to rack hardware and what is this AWS thing and get off my lawn. And so you missed cloud. And so for me, I've taught my, this skill. My dad also has this mindset. So if like, if it's new, we're, we're kind of into it. Doesn't mean like we go all in. We are open-minded. I bought Bitcoin. I mined Bitcoin as soon as it came out, and I try things constantly. So I, I think this has helped me in my life keep up with the changes that you can take a fork in the road. It's like the fork in the road meme is like you either have this identity crisis or you, you leap into the next platform and the next big opportunity in your life. And I think happened with AI autocompletion and Copilot. I was like, my God, this is the second coming of Jesus. And people were like, no, no, no, it's junk. And then it happened with ChatGPT and then it happened with agents and it's gonna keep happening.

Nakul Mandan:

But what's the new engineering skill then in today's world?

Guillermo Rauch:

So there's a meta game here. My best engineers have been the ones that adopted these tools the fastest. They've developed an understanding of them. That's deeper than the average. So there's a parallel here. If I ask a model, build me a site, I get purple. And most people that are in the know can tell right away, like this, it is like the em dash in the ChatGPT speak, right? The same happens with engineering. Like if you just say like, fix it, make no mistakes, you can only go so far. And so there's a meta game and a meta skill of how good are you deploying this? The ultimate meta skill that I think is the next evolution of software companies is that you're building the factory that produces software. And so I think what's— I can objectively look at our data and I've seen the acceleration of delivery correlated with the acceleration of revenue. And why is that happening is because we're leaning into this shift of I work on the output to I work on the machine that produces the output.

Nakul Mandan:

It's a recursive founder mode taken to agents.

Guillermo Rauch:

It's the technologification of recursive founder mode. Yeah, 100%. In fact, funny enough, this is hilarious. One of my, uh, poster child of, uh, founder mode is literally turning the AI SDK repository for the world to see into a software factory, meaning you're gonna be able to report an issue and 99% of the work is gonna be solved automatically. Or so a lot of people said with AI SDK, how are you gonna keep up with the model changes? It's like, well, first of all, if AI is super intelligent, I can have AI reflect on what is the new capability and then add it automatically to AI SDK. That was kind of a dream a year ago. Now it's reality. Any customer report or any advancement in the underlying models can automatically get typified and then released.

Nakul Mandan:

But are you finding then accordingly, you know, I actually have discussed this with many people at AI companies that are you finding that product people are becoming more capable now because it's about what to build or engineers who can be, yeah, where is this going?

Guillermo Rauch:

Yeah. Lee Robinson wrote a great post. I think he called it like the rise of the product engineer. And I think that's spot on. You know, I see it a lot in Silicon Valley, perhaps it's not yet as widespread, but it's a duality of knowing, you know, product management skills and thinking about your P&L and thinking about your customers and thinking about their case studies and thinking about expansion and thinking about retention and also thinking about the engineering of it. And you know, when I say stuff like this, it does sound a bit like science fiction or alchemy, like, oh wait, you have, so you have an autonomous software factory produced by someone that is this unicorn product designer engineer, but we're kind of getting there, right? And I think again, that goes back to like why The future is going to be very founder-centric, flatter organizations, fast information flows, higher ownership, and, um, and transparency.

Nakul Mandan:

You know, when you were 17, being able to build was rare, and you showed your exceptionality to the world by building on open source, distributing it. Today you can build with AI, you can distribute on XR, Hacker News So how does a new How does a 17-year-old today show that they're exceptional to the world?

Guillermo Rauch:

I think at least for me, I've spiked in not just the ability to write the software, but try and go a few steps further. Anyone can build software, but some people are able to communicate the value of that software better. Some people are able to get customers and then develop from there. So I can obviously post and say, here's my technology. So you have to think about what is the delta between me and all of my competitors. There's a handful of people that I follow that have even run small businesses, but I follow them because of— I'm attracted to their storytelling ability. I recently invited— every Friday we do a fireside chat with a company, and I invited an indie maker. His name is Marc Lou, and I actually started asking him like, walk us through your process. Tell us how you design a landing page, how you think about your call to actions, how you think about correlating visitors and revenue. So his mind is this sort of like full-stack AI-assisted engineer and business builder. And so I think that represents a lot of the ethos that we all have to have, especially in this world, to your point. And so I think it's not gonna be enough Maybe I had some alpha back then because I was writing the code and I could publish it and then I could get some GitHub stars. That today feels much more like a baseline. I would have to go further. In fact, I'll tell you something that was really cool. So when I launched Socket.IO, one of my open source projects, Microsoft started using it for a very cool product, Microsoft 365. When it was born, they took the Socket.IO protocol and Socket.IO client and they embedded it into their tool. And looking back. That could have been an amazing story to tell. I just kind of, I don't know, I kind of let it slide. I took it for granted, maybe because I was having so much success with just the code and the product alone. Today I would be spending like a lot more time with my customers talking about them. And this is the other realization that I had. Open source is starting blurring the lines with businesses and startups. Like it used to be that you just presented your project. You didn't even care that much about the name. You didn't care that it had a logo, a website. And so the bar just kept getting higher and higher and higher. And I would expect today, like I talk to 15, 17-year-olds who are building businesses, creating revenue breakthroughs. I still think the reality is models can only do this much. If you have a very compelling thing that you've been able to do with the models, that it didn't occur to other people, or you've pushed them. Sometimes there is actually meta alpha where you educating the world on how different models are better or how they disagree or how combined they produce better results. That's what's attracting a lot of people to our AI Gateway. It's no longer a game of just one model, and it might be a game of efficiency as well. It might be that you are able to produce at much lower cost. So I still think there's alpha everywhere, but again, the game is fundamentally different.

Nakul Mandan:

There's a theoretical possibility already today that everybody's a builder and everybody can do more to automate and 10x themselves. But when I talk to CEOs, they are seeing a huge spectrum even in their own organizations around like, is every marketer 10xing themselves? Is every finance person 10xing themselves? No, they're finding that Of course, engineering is deeper, but like there's top 5% people in every organization, even in AI organizations that are really 10xing themselves and 95% are still lagging. Two questions there. One is, are you finding that at Vercel? Maybe not because of the technical nature, but I'm curious. And second is, how much of that is just a cultural upheaval that every company has to go through? Because the 5%, the delta between the top 5% and the next 95% has multiplied.

Guillermo Rauch:

No, this is real.

Nakul Mandan:

Even at Vercel.

Guillermo Rauch:

Well, so here's how we're solving this. So what we noticed is most organizations said, in order to superpower my organization with AI, I'll give everybody a Claude subscription. And then the divergence started. You have this prompt library and you follow Raoul Pal on X and you have skills and whatever. And that other person just like prompts. Clocks in, clocks out. And so how do I raise the floor? How do I make everyone build on a platform? So maybe we didn't have this problem as drastically as some of the anecdotes that I've heard. Literally, I heard from one of our customers that they invite AI CEOs to do fireside chats to inspire the organization on how to use agents because people don't even use them. And so they want to hear the success stories and things like that. Maybe they're skeptical, I don't know. But what we did is we created an internal agent that is in a self-improvement loop.

Nakul Mandan:

This is the one you announced 4 days ago.

Guillermo Rauch:

Yeah. So I told the story because what we announced is Eve. We announced the framework of how you can build an agent to run your organization. And our own internal agent is called V. And so instead of having people all over the place, go to ChatGPT over here and go to Claude over there, they go to V. And this is to us a culmination on a lot of like scars, lessons learned, trial and error. At the end of the day, for an agent to be successful, it has to be like the iPhone. There's just one place to go. I can call, I can text, I can play video games, I can do anything with this thing, but I just know I take it outta my pocket, one interface. That's the future of agents. We internally, we used to call this the Omni-Agent project. We wanted to have one front door so that everyone in the company, when you don't know something, when you want a job to be automated, when you have a question about the business, when you have a question about a customer, or even when you want something automated, you can go to V. And under the hood, V has a collection of agents, applications, capabilities, crucially skills. and IT and a team that maintains it.

Nakul Mandan:

What you just described, I believe in it, but at the same time, you earlier said you don't believe it's the end of SaaS. Like, but why will somebody log into a CRM ever? Why will somebody log into a Zendesk? I'll always talk to V

Guillermo Rauch:

Okay, so great questions. So for example, I literally just approved some spend for a sales and marketing platform today that plugs into V And yesterday I gave feedback to one of my colleagues that the one of our content solutions It takes too long because it was not designed for agent interoperability. So I'm literally reconfiguring all of my priors. I used to think your software is good or bad based on, I don't know, the website is fast, it has a lot of features, I can click around and it works reliably. Now that's all table stakes, of course. But the thing that is making some platforms spike is their ability to interoperate with the Vs of the world. And if you're not good with V, I actually like, I have a few vendors on the chopping block, like literally because of their inability to, and by the way, it's so, it's subtle things sometimes because maybe they do have an MCP, but their MCP takes, I don't know, 2,000 tool calls and it fails a lot. And they just, they didn't put, you know that they check the box, you know that they're either late to market. And they're trying to play catch up, or you know that their infrastructure doesn't hold up. So this one's fascinating, right? Like literally today there's a major GitHub outage, and I have a lot of love for what that product has done and what that platform has done for the world, and I have a lot of friends there. But the reality is that their infrastructure has not been able to hold up with the agentic demand, and that has become a bottleneck for us. And so I, I see this pattern a lot because what has happened is a lot of software maybe did present the right API, maybe did present the right interface, and now you have a billion agents querying it at 100 requests per second. And so it's airing a lot of dirty laundry. You know, we used to say, come to Vercel because you're gonna need this elastic serverless compute capability. And so people would say, no, you know what? I, I don't think I need that. Well, look at this. From week to week, agents can 100x your load. Our own load has doubled in a couple months. And so I'm sometimes, obviously not every piece of infrastructure is perfect, but some of the key bets that we placed, I'm really thankful for because no one could have predicted this massive acceleration. And what's probably happening at a lot of companies is that someone has an agent, and is deploying to Vercel and is deploying to the homegrown legacy platform and the agent is just going and going and going. And the Vercel one is delightful and the legacy one literally gets replaced. So a lot of companies are finding this with their, I mentioned that marketing company. Can your platform where you deploy software absorb the creative demands that all of these coding agents have unleashed? Most of the time the answer is no. It's like I gave you a Ferrari. I was like, here's Cursor, it can do anything. All right, deploy. Oh, I just went into a 2-week queue, or I can't deploy an internal tool or an agent. And so there was that impedance mismatch that obviously has fortunately really favored us.

Nakul Mandan:

Let's shift towards how AI has changed Vercel and how it operates and stuff. Starting with you, how has your life as a CEO changed with AI? What are decisions you are personally making now you've outsourced to AI? Yeah, walk us through some real examples.

Guillermo Rauch:

It's been amazing. You know, one thing I respect about, super random thought, but like Gary, Y Combinator, he did something that's so unusual. He went back to the arena and he started building with AI. I would say to every CEO, you have to do that. I recently mentioned this. It should be mandatory that your C-suite and your top engineers need to come to SF and spend some time here because sometimes it's hard to believe what these agents can do and how fast companies are upgrading their operational rhythm, their best practices, their internal platforms. And so, and the world is moving so fast. And so, you know, when, when Gary starts publishing G-Skills and whatever, yeah. That's fantastic because how can he select or pick or invest in the next generation of companies if he doesn't understand the technology? And the fact that it's somewhat controversial or funny or a meme is kind of, uh, so ironic to me cuz like it's so obvious and that's what I try to do every day. I try to gain a grasp on is your product, is your idea, is my customer's UI good? In this agent world, and that requires you to use agents. You can't outsource that. It's such a dramatic platform shift. It's a little bit like mobile in the sense that if you, if you were a CEO and you were like, you know what, I'm not even gonna get a mobile phone, how could you possibly understand that you need to shift your roadmap and priorities to be a part of that platform? Right? And so I think it's the same with agents. So I spend a lot of time dogfooding our products from an agentic perspective. I spend a lot of time looking at the landscape, looking at how others are doing things, building things.

Nakul Mandan:

But are there decisions that you're spending less time on because AI is able to give you a recommendation much faster?

Guillermo Rauch:

I do a lot of research with AI. I just published, and I'm actually quite chuffed about this. So I created a research folder. I just heard about this. Yeah, I'm a CLI guy. So like I, when I started the company, the big idea was a CLI that made it one command to deploy your application. 10 years later, what do agents need? One command to deploy their applications. And I've done the same with AI. So I try to use this AI, like I call it the inexhaustible IQ machine. I try to unleash it in the lowest level substrate I can find, meaning the most access possible to my computer, my data, et cetera. And so CLI I really like, and I fire off a lot of asynchronous research tasks. The ones that are personal, I launch on my CLI. The ones that relate to the business, I just have a DM thread with V. So I go to Slack and I talk to V, and in, in that context, V can access the company's know-how. In my personal context, it can access my personal information, my my Gmail, my experiments, because I also try, and maybe this is, I don't know how unique this is, but I try to put myself in the shoes of someone that doesn't know anything about Vercel. Sometimes I even ask the agent, you know, I have the thought, and this is the beautiful thing about agents, like the ability to expand your mind and your throughput is insane because as they become better doing longer horizon tasks, they don't need a whole lot of babysitting. I can say, hey, go and research how easy it is to sign up for the Vercel AI Gateway and get an API token. I had that thought. It's like, I'm curious, how easy is that? And in a past life, that would've been like a product review session. It would've been a meeting. It would've been some slideshow, someone like taking screenshots of the onboarding funnel or whatever. It would've been a conversation. I would've been pestering something on Slack. Hey, Can you check on that, please? No, go to the agent. So when I go to a human, I'm really, you know, respecting their time and their attention because I kind of, if I DM you, it probably meant that my agent couldn't figure it out.

Nakul Mandan:

Actually, so yeah, let's dive into that specific thing. What is the limitation as of today? And maybe where that limitation will be in a year. Why do you still need to go to a human today at Vercel?

Guillermo Rauch:

Yeah. So agents can many times inform. So one pattern that is fascinating and that I think Jack from Square is arriving to this conclusion as well, is that it's not gonna be an or anymore. It's always gonna be you, me, and the agent. I was talking to the CEO of Slack yesterday. I was telling him like, I have a very specific product requirement. I needed to be in a DM thread and obviously with respect of privacy and security, both parties need to agree. But imagine that we always agree that it's fine for an agent to be there to assist us or to proactively contribute to the conversation. So it's no longer like, oh, I go to you or I go to the agent. Most of the time, and it's fascinating because in this process, I also teach people. I think this is very interesting. We have V, but that doesn't mean that everyone This is, uh, it's, it's hilarious to me because, you know, yesterday I said we now have intelligence that is as accessible as running water. And I really think that our brains might not even be able to comprehend what that means. And I do have a small example of this. It's that I have to put myself in a conversation with an agent and then I, I ask the agent in front of you because then that person knows just how capable this thing is. And that might be our limit right now is that we tend to, you know, we all follow the inertia of the technologies that we grew up with, how we've been doing business for the past few decades, and we do need to actualize. And by the way, to your point about what models can do, I have to revisit these priors pretty often.

Nakul Mandan:

Every other month, every month.

Guillermo Rauch:

We've been investing a lot at Vercel in benchmarking because it literally has become almost every other day. And so we, we released a DeepsecBench because the talk of town right now is cybersecurity. And the question that I get a lot from CIOs and CISOs is, how do I protect myself? How worried should I be about the threats from Mythos or OpenAI models? And I myself was finding myself asking our CTO who created this tool DeepSec, which allows you to run Basically, it's really interesting. So agents, much like human beings, can take on roles. You know, when you go home and you're with your kids, I'm not the intense CEO, like hopefully for them, right? Like I'm like playful and whatever. And when I went to the World Cup final, I was a soccer fan rooting for Argentina, and now I'm the CEO. And the same happens with models. So DeepSec, tells every agent and every model, go and be extremely paranoid and pay more attention to auth and security and privacy and try to find vulnerabilities. And you would be shocked at how good the models can be at analyzing themselves. It's a little bit like psychopath therapy because they wrote the code and then they can tell you how the code is wrong. And what was happening is Every model release was changing the economic equation. We noticed, for example, that Sol is incredible at finding vulnerabilities, but it was costing for our codebases thousands and thousands and thousands of dollars. And then Kimi came out and we realized, oh, it's not quite as good as Sol, but it's 7 times cheaper. And we need to adjust these priors every day. Because there's a new OpenAI release. There's been 3 or 4 open-weight American models in just the last couple weeks. And so we've sort of transferred our, you know, instead of me having to vibe out what is the right agent or model, we're trying to formalize that. And then we'll try to hopefully soon when you ask V, hey, can you do a security review of this? It's automatically calling DeepSec. And DeepSec itself is gonna say, all right, how much do I invest? Because the question about intelligence on tap is that yes, it's like tap water, but you imagine that you have several faucets and it's also liquid gold. And are, are you gonna have just like a casual, like liquid gold sprawl for this task? And so, so much of this, and this is why, you know, for people that are thinking about the future of their companies, so much of this is what's gonna inform your P&L. Tokens are gonna become a huge percentage of your operating budget. And so you're gonna have to think about, well, in this, for this area of the business, we spend this much with this models. For this part of the business, we spend this much. It's gonna vary tremendously.

Nakul Mandan:

But the way you're describing, why do you even need execs anymore, right? Because the way you're describing is you go for research of any kind, even internally, first to V. It gives you full visibility into what's happening. You only go to a human when there's a very specific thing that there's a problem that needs to be solved. Do you need execs or super ICs then outside of you?

Guillermo Rauch:

Or maybe they're approaching the same thing, right? And again, goes back to like why we're so horizontal and transparent. You know, our COO came into with this incredible perspective when she joined that she was going to create the most technically competent and most scalable go-to-market organization on the planet. And what she did is two things. One, create the agents that have sales skills. So lead routing and going back to what informed eventually the creation of the Eve framework and V was that she built like a dozen go-to-market agents. I don't even know, I, I don't personally don't even know them because they're in the weeds of the operational rituals of my go-to-market team. but if I had a question about go-to-market, where would I go? It's like I have to learn about like these 12 agents. And so I go to V So she started sort of rethinking go-to-market from an agent-first perspective. Our data organization did the same thing. Data is actually maybe one of my, my proudest agentic accomplishments because we're able to bend the headcount curve. We created the organization that just works on the agent. They work on the agent that can answer data analytics. They work crucially in the security model because we need to have policy for who can access what information, but that has to be transferred into the brain of the agent. And, and so what they're working on, going back to the factory, they're working on the machinery for data science at scale. And it's fascinating because my expectation a few years ago was, I told Abby, our VP of data, I want every person in the company to be able to file a ticket and within an SLA, get an answer from an analyst or data scientist. Nowadays, the SLA is something that we measure in our observability platform, is how fast is time to first token. And instead of going into a ticket queue, it just goes to V. So you ask, for example, our CTO asked the other day, can you give me ideas for banger tweets? And V said, we recently reached 2 billion deployments, we should tweet it. I mean, if that's not an AGI moment, because it had to piggyback on the data agent and it had to come up with hypothesis. What could we, what could be interesting to tweet about? Could be progress in the business. It could be something that is not obvious. And it came up with this observation about a milestone. And so this is kind of how we run the company now. Those executives are agent builders.

Nakul Mandan:

If you were to rebuild Vercel today, let's say you were starting Vercel today, how would the org design look like for you?

Guillermo Rauch:

This was not, and maybe this was coincidental, but we called Eve, our agent framework that's open source and people can create their own agents with it. It sounds to me like the genesis of a company. So when you start a company, you need to lay down the foundation. And the foundation used to be, at least from my sort of engineering brain, it used to be the website or the cloud infrastructure. There's something more primordial than a.com and perhaps even a Delaware incorporation, which is the agent. That's gonna help you run the business. My son's favorite movie is Bicentennial Man. Have you watched that?

Nakul Mandan:

I have not yet.

Guillermo Rauch:

It's an Isaac Asimov book turned movie. It takes place in San Francisco. That's why we loved it. We love it. I think as a family, we probably watched it 50 times. I'm getting tired of it. There's a scene when the robot, who's the main character, goes and finds the factory that created him. The factory is run by the human that invented the robot and the robot that wasn't robots. He has this sidekick that follows him everywhere and when he does quote unquote surgery on the robots, and this robot knows everything about the business and is funny and charming. And this is kind of how I envision the company of the future is you and this V-style agent that you're gonna keep investing into. Even by the way, things like personality. So one of our interns, Madison, she works on the V team and she realized V doesn't speak like a Brazilian, boring, long. It's just like the vibe is off. And so she worked on improving the personality and it makes for a better employee experience and it's more like us. And I'm not gonna anthropomorphize too much here. Like these are still like, these are machines, these are infrastructure components, right? But they are part of the, you know, you start a company and you start your agents.

Nakul Mandan:

You are 800 people today. If you were starting Vercel today, do you think at scale you would actually ever get to 800 people or it would be 100 people and insane number of agents?

Guillermo Rauch:

It's hard to say because in the case of Vercel, the size of the ambition is every pixel of the internet and every token of the internet. And so it's a pretty bold ask, right? Like all of the websites and all of the agents. And so that requires quite a bit of firepower. But every day I'm surprised by the scale that we get. And again, this is why I'm telling you this sort of like real-time anecdotes because I'm learning. Design engineering is the one that I heard from our engineering leader today. He was like, yeah, John thinks that this agent is gonna change his headcount requirements. He thought he was gonna have to hire all these people. His creation is outpacing and outperforming his expectations. This is what I was telling you about. You do have to get in the weeds of experimenting because sometimes you don't know what's possible.

Nakul Mandan:

Okay. So let's talk about the product portfolio itself because so AI-generated code was already a great tailwind for Vercel. You were the application cloud for the world. Why even come up with v0? Was that a coincidence that it happened? Because v0 is now a big part of the product portfolio.

Guillermo Rauch:

I think a lot about, maybe this already came across from my answers, I I think a lot about time to aha. I think about both for agents and people. Like I think a lot about like, what is the friction on getting the first output on our platform? And what we learned with v0 was, okay, it's not just about getting something built, is how can you get something that is of the quality bar of Vercel? That was the first inspiration. Can I create a Next.js project? With all of the know-how of Vercel. Imagine if Vercel could package all of the skills of what it means to create a great website or web application, and then all you need to add is the prompt, at least to get started, right? And so I do think that every single day you have to keep investing in the skills of your own and how far you're gonna take this project and whatnot. But what if we could raise the floor? And so v0 became that thing for us where it's so much easier now to like even try Vercel. A lot of people that interview with us increasingly tell me that, okay, like I could go to Vercel and I could like, you know, get the account and then clone a Git repo and do all of these things. Even our own, what we were known for, which is like this like super awesome developer experience, we wanted to challenge it and it was like, is it possible to get started with Vercel even faster? And so v0 became that for us.

Nakul Mandan:

And then on the application cloud itself, how does the cloud need to, or the infrastructure offering need to evolve where the user is now an agent versus a human?

Guillermo Rauch:

Two primordial dimensions. So one is the services that are needed for the agent age are fundamentally different. The inventor of the cloud, let's call it, it was Amazon Web Services. They created a bunch of services. That made sense for the web. One of them, for example, is the CDN. And so a CDN was necessary because origins could go down or they couldn't scale, and you need to cache data and you need to globally distribute static content. The disruption that Vercel brought to the web world was we realized that content was no longer static, it was dynamic. And so we needed to build a modern CDN. For how the web was evolving, more personalized, obviously today more generative, more dynamic. But now we're not just building dynamic web applications, we're also building agents. And so the metaphor has, you know, a striking similarity because just like we needed a CDN for pages, now we need a CDN for tokens. And so this is a new service, AI Gateway, that was invented for the agent age. And so that's one dimension is What are the net new building blocks that you need to build agents? So AI Gateway, Sandbox, these have metaphors to the cloud computing age of, of yesteryear. And the other thing that you need is that, you know, it used to be that you as a human being need to study. Okay, so for an agent, I need an AI Gateway. Okay, got it. You go to a workshop. Oh, and the agent will run code. So I needed to run it in a secure sandbox, right? Sandbox. That's not gonna happen anymore.

Nakul Mandan:

Yeah. The agent chooses everything.

Guillermo Rauch:

The human says, I wanna build an agent. And the, and it says, okay, I'm gonna use Eve. I'm gonna use Agero. I'm gonna do this, I'm gonna do that. And so what we're thinking about is, okay, what are people wanting to build and how can we be there for their agents? And so it's a combination of having the new services. but it's so many, you know, the, the founder of Notion called this, it's, it's the design has changed dramatically. Actually, he had quite such a profound statement. There's companies in the Valley that have been known for their glorious, beautiful UIs and illustrations and iconography, fonts, et cetera. And he says something that was striking. He said designers should now be designing CLIs. And APIs. It's like, think about that. You were using a design tool, illustrating whatever, and now you have to think about an interface that was created in the 1970s for the VT100 terminal. You know, my wife was asking me the other day, what is TTY? And I think it stands for teletypewriter. And it's this like ancient technology. Like it's like you have to go to like a museum to learn about these things. It turns out agents need this new, more fundamental, more close to the metal interfaces. They need access to the foundations of computing. They don't need the interfaces that were created for human beings. And so it's really important that companies are able to very quickly iterate to that. And for example, we just launched a framework that helps you very easily deploy MCP 2.0 backends on Vercel. So if a company wants to get there and present themselves to these agents, they're gonna have to deploy MCP servers. I used to be like, okay, like webpages and, and, and fast and core web vitals. And now I'm telling you MCP 2.0 servers.

Nakul Mandan:

Because from the outside it does feel like v0 has gotten a lot more attention, but every— the way you are describing it, correct me if I'm wrong, Vercel still is massively the infrastructure company that will power the building of software, whether by agents or not. v0 might be the shiny object that can enable—

Guillermo Rauch:

It gets you to the platform, right? So what I care so much about is that whether you're using Claude Code or v0, can you start deploying software? Can you start deploying agents?

Nakul Mandan:

That still remains the soul of the company.

Guillermo Rauch:

It's the soul of the company, will be forever the soul of the company.

Nakul Mandan:

So let's talk about finally just the the AI impact on building. One question that comes up often, product used to be a multi-year moat. Are there moats anymore if everybody can build fast? Like, what's your thought on how is all of this going to change building of companies, building of defensibility?

Guillermo Rauch:

At the end of the day, I think customers will always go to the highest quality products across every dimension, right? I'll give you an example, infrastructure. The reason that infrastructure has such a strong moat is that you can't just vibe code infrastructure. You literally have to have operational rigor and best practices around security. And we're benchmarking all of these agents to make sure that like our infrastructure is the most secure and performant in the world and it's globally distributed. And so I think platforms and infrastructure products, I think have done really, really well in this age and they have a tremendous demand. I think bespoke products. The bar for what is a venture-grade investment will recalibrate for sure, because VCs are always talking about like, is that a lifestyle business or is that a venture-backed opportunity? And so there's probably a world, and folks have been calling these solopreneurs, and I hear a lot of anecdotes of folks that are doing like tens of millions in revenue a year, single-person companies. I think there's going to be reinvigoration of small businesses, ironically, because think about it, right? Like you can do what took a team of 100 people. We used to call it a SaaS company and required tons and tons and tons of investment. You're gonna have the time of your life cuz there's probably still a lot of demand for software, right? Like think about it, like the idea that my cousin is just gonna vibe code an app for herself and then maintain it. What I, There's a little bit of science fiction to that, right? So I think software is just fine. The question is if it's a bespoke piece of software, how big is the opportunity there?

Nakul Mandan:

Is there anything that you believe machines will not be able to do that humans do today?

Guillermo Rauch:

Yes, I do think it's kind of a— maybe it's a recursive answer, but I think the human empathy and emotion and connection is ours and will be forever ours. And I actually think it's also under our control, meaning I refuse to have an AI as a friend, and I care a lot about the connection to the people that work with me. For example, I actually think the social networks are quite Lindy in the sense that connecting to other people in the age of AI actually has a premium. It's amazing. And not only social networks, but even like, you know, like spending time with people in physical space. I think engineering itself, I think the hierarchy of like the capital allocation, what needs to be built with what values and what taste that needs to remain under our control. And I think will remain under our control.

Nakul Mandan:

So it's an allocation.

Guillermo Rauch:

It's capital allocation at the end of the day, right? And like, who do we give our tokens to, to take us into that next big, bold next step? Because I think with enough tokens, I actually tweeted about this at one point, there was a phrase that got really popular in the early days of Linux and Unix, with enough eyeballs, all bugs are shallow, meaning the power of open source is that I can go into a mailing list and I can say the Linux kernel is crashing when this happens. And this gentleman, I believe it was Eric Raymond or someone like that, mentioned he noticed this pattern of like, you put it out there and people come and go and they see this bug and eventually it gets fixed. With enough tokens, all bugs are shallow. All performance problems are shallow. All security problems are shallow. There's still a hierarchy of like, what is it that we're going to build? Where are we going to devote our tokens to? And there's going to be a market force in play because yes, I do defer purchasing decisions to my agent, but I also oversee them. And I own the P&L of like, all right, agent, what do you go and spend money on? Right. And so we're still in control. We need to remain in control.

Nakul Mandan:

So it makes sense. Capital allocation will remain a human thing. The founder energy of what to build will remain a human thing. But not everybody in the world is like that. And so my question is really, how do you see society going forward? Like Dario obviously has a very doomsday type prediction around all jobs are going to go away and stuff, but like maybe 98% of people are not that high agency that they want to make that. They just want to go to a job and have a good work-life balance.

Guillermo Rauch:

I have a completely, I have a completely different take. I'm so optimistic because I think we're going to go into like a, did you watch Ready Player One?

Nakul Mandan:

Yeah, of course. Yeah.

Guillermo Rauch:

It's like this. Infinitely whimsical world. You see this a lot because people are building their own video games. I predicted this a while ago and people loved it that the web was gonna become a lot more whimsical and it's happening. Just yesterday, something went viral that I shared with my entire team. Someone vibecoded a 3D examination of the human body. And I remember when I tweeted the whimsical thing, I was like, when I was a kid, we got Encarta. For our, uh, home computer. It was this encyclopedia that because of the distribution model of I build content, I put into a CD and I get a chance to do it once a year. So much work had gone into like the shiny pages. The page about the human body had this really cool illustration and interaction, and I could click on like the stomach and it would blow up. And like, this was on Windows 95. Fast forward, we went into what's scarce is access and what's the alpha is in massive distribution in the update rate of information. So we got bland Wikipedia and it's literally like a white page with text. And so now, so my prediction was, no, we're going to go back to like really cool immersive experiences and worlds and starting to happen. Someone shared, I think it's like, uh, uh, I don't know, like humanbody.vercel.app or something like this. And I can guarantee you that that was all prompt-based. The creativity was infused in like the presentation layer. It's like a piece of art. And so I think artists will make a comeback, which have been somewhat suppressed by this like, again, industrial-scale content machine, um, small businesses, because it's, it's a little bit like I actually prefer to get like the handmade chocolate or the olive oil. Uh, my, my, my family produces olive oil, and so I, I love our olive oil. It's different from big oil. And so, you know, much like Shopify, who we're partnering with so that we can enable more businesses to get started, you can now v0 a shop that gets this incredible backend of Shopify. Now I think those platforms are just gonna do so well because people are now gonna have the ideas and instead of being like, ah, there's no freaking way I can do it, one day you're just gonna prompt it. And so I think that we're completely eliminating that friction.

Nakul Mandan:

Yeah, it's like going back to the Renaissance. Like it's almost like a new world of creativity.

Guillermo Rauch:

Get in, we're rebuilding the whole freaking internet.

Nakul Mandan:

Yeah, yeah, yeah.

Guillermo Rauch:

It's wild.

Nakul Mandan:

Yeah. Okay. Let's shift gears towards the inner game. You know, you, you were a legendary coder. Few have built what you've built, and then you became a CEO, a legendary CEO. Now, were there parts of the journey where, where it was tough to become a leader from an individual, like 100x coder? Like, can you walk us through how that journey from a legendary coder to a legendary CEO has been?

Guillermo Rauch:

It's tough, right? Um, first of all, In the early days, you have to do everything by yourself. You have to write the code, you have to build the infrastructure, you have to design, you have to market, you have to do payroll, you have to do like all of these things and, and you are not world-class at all of them. I can guarantee you that, you know, there's a few people that can do a handful of things at a world-class level, but I tell, I tell them, let's figure out, go and negotiate a good deal with this company or this vendor. Like they just like, they have no freaking clue. And so I think what the challenge of today is going to be for every leader is that they will be able to do more. If, if they stick to what they know, they're gonna, they're gonna be left behind because there has been an amplification. Your bandwidth as a human being has been amplified. Is what we were talking about earlier. If you just stayed in your lane of like, I'm the CSS expert, I'm the JavaScript expert, I have bad news for you. So I think the way that I think about it is each leader has to like become more full stack. There's been times in my life, you know, I, you were asking me about burnout before the interview. I think the number one job has always been, I think Ben Horowitz talks about this, right? Like maintaining your own psychology. And going back to the metaphor of like a company's job is to— the new product is your factory. There's a quote by Elon that said what he realized at Tesla is that the product was not the car. Of course, the car. The product is the factory that can produce those cars at scale. And so it's the same for software now, right? And, but you are responsible for your own personal factory. And so I maintain this factory to the best of my ability. It didn't always come naturally. I was, I was not always, you know, Monday through Monday, you know, religious about my routine and, uh, you know, taking care of my body and my diet. I think I do a pretty good job now. I'm always trying to optimize more.

Nakul Mandan:

Did those changes come because you hit the ceiling somewhere?

Guillermo Rauch:

I did.

Nakul Mandan:

So what forced the changes?

Guillermo Rauch:

Yeah, I, I think the lesson always becomes clear when the student is ready. So I think there's probably mistakes I'm making now that are like super obvious and like I'm gonna be there in 2 quarters or whatnot. And so I always have this learning mindset. I've never arrived. No one has ever arrived. I think ultimately, ultimately it's all about the journey. And, uh, and I'm very almost like spiritual about this. It's, it is for me a spiritual journey. And so there's this like whole like self-reflection is bad or whatever thing. I'm the opposite. I'm finding myself and I'm discovering. So I had two things that motivate me. One is like understanding myself more and then understanding the positive impact that I can have on others. If I work out a lot and I treat people really well and I raise their ceiling and I help them do cool things, it comes back. And that's when I learn, say, oh, that was really good. I'm probably also doing things that are bad. And so just constantly trying to be adaptive, but no, I like, I have low lows. I did have a transcendental moment for myself, which was, so the problem with startups, by the way, is that it's never a straight line. It is never like a perfect curve up into the right. And so sometimes you are doing really well and then you regress. So sadly, life sometimes gives us the A/B test. I'll give you an example. I started the company So focused on simplicity, and I mentioned is, is what's made us so successful. Like we brought to the world this radical idea of like a single command deployment. And then I did get distracted and started throwing shit at the wall at times. And so life gave me that A/B test. See what happens when you do all the other shit and see what happens when you stay focused and go deep. And so that's why sometimes it's really painful.

Nakul Mandan:

Did you end up also having to level up as a manager or manager of managers, or was all of that path more natural?

Guillermo Rauch:

It's constant, right? And this is why I encourage everyone to have kids because at the end of the day, I mean, the impact that we can have on other human beings is, we sometimes I think underestimate, right? And I'm not just talking about like my ability to manage and like get really cool outputs and really good code and whatever, but like The impact that we can have on people's lives. And we spend so much time at work. Think about it, like for me and my colleagues, this is our life's mission. And so we spend a lot of time together. So I try to spend every second of my life thinking about how can I be a better manager, peer, student, reward the right things, build the best culture. Maybe this is something that I also started paying more attention. I literally invoked the value of Vercel today, which is iterate to greatness. And so working on the dissemination and promotion of our values is what's actually given me scale because I think there is a perspective that management is one-on-ones. And luckily we have Jensen Huang who has shattered that illusion. Management is not about what, I don't know, Harvard Business Review told me that I have to do a one-on-one. No, there's, at the end of the day, the best manager will scale himself. So Jensen says, well, I have a meeting with 10 people at the same time. You know what also gives you scale? Leading by example, values and adherence to values. You know, I religiously do customer support rotations to the point where it's kind of become a meme that people tell me like, oh my God, the CEO giving me support. It's like, Yeah, like of course there's other CEOs that are not giving support. Great, great for me and my competitors, right? And so I try to lead by example and I don't want to be performative. I really try to be in the weeds and feel the pain and do the best that I can for my colleagues.

Nakul Mandan:

Is there a CEO mistake you made that you can share that you still think about?

Guillermo Rauch:

I hate making mistakes. I just freaking hate it. So someone said, uh, I don't know who was it, but like, show me a good loser and I'll show you a loser. And so when Argentina lost, I always say I'm Argentinian, blah, blah. And people were saying like, Argentinians are such sore losers. No, they're champions. That's why your country doesn't have as many World Cup championships. And yeah, I, I'm, I don't. Advocate for beat someone's leg, like take down someone or whatever, like be physically violent or whatever. But like, I don't like to lose and I don't like to make mistakes. And so for example, if I've lost focus and I had seen before the right path and I have, you have that fork on the road and then you take the wrong one. Now, luckily for me, and this is what gets me so, I say so motivated. I never fall into despair is because the vast majority of mistakes are, uh, two-way doors. And so I think sometimes we think of a mistake as catastrophic, and the reality is you breathe and this too shall pass, and you've overcome that sometimes within 15 minutes, sometimes within 2 weeks, sometimes within a month. And so you just have to grind it out. You just have to grind it out.

Nakul Mandan:

You have 5 kids, a 6th one on the way. Congratulations. Uh, you're very obsessed with fitness. You're at the gym every morning before, like, where is all— and then you're responding to people's tweets. You're responding to customer support tickets. Where is all this energy coming from?

Guillermo Rauch:

Yeah, I mean, a couple of things. My, uh, the feedback that I would get from my teachers in kindergarten was like, What do you guys give this kid? He's a freaking ball of energy, right? And so like, I'm glad that I found my place in the world where I can catalyze an insane amount of energy into the world. Like, uh, I don't know, like a, like a low friction conductor material, but you have to tune that and you have to make that sustained over time. I invest in the rituals that give me that. And so obviously cardio is Cardio, I mean, it's fascinating, right? Like you have to spend energy to increase your energy budget. Has anyone studied this? Like, this is why fundamentally, like, I, you know, people are really excited about GLP-1s and things like that, but like, it's never been about the weight. Like what you get out of grinding it out at the gym is that you've expanded your energy capacity. And so for someone so obsessed about optimization and engineering and stuff like that, like what can I do to increase my energy budget? Because then I can be a better father, husband, like the LinkedIn bio thing, but husband, child, father, whatever, like advisor. Like it is a byproduct of how much energy I have, I think. So part of that is like, how much do you do of that baseline maintenance work? Part of that I think is also just like the personality, like how I am and stuff like that. And. This is like a fun game. And like, if you love what you do, you don't feel like it's work. You don't feel like, oh my God, I'm like, and by the way, that also means like you have to tune. This is what's great about you ask what is the value of an executive leadership team is that some of my execs get that energy capacity from other tasks that maybe I don't love as much. You know what I mean?

Nakul Mandan:

Have you ever had burnout in your career? Have you ever felt down in your career?

Guillermo Rauch:

So I mentioned it's never a straight line into the top. And so I think I've had lulls. I think by labeling it a burnout, I think people maybe like try to now say like, oh, it's a condition or it's, it's a, and I, this would be highly controversial to some people, but like, I try to like, you know, look at that as another subject of observation. It's a little bit like, you know, that you're, you have a certain level of hunger. And you have a certain level of energy and you have a, you know, I, I, something funny that I do is I measure how many keystrokes I have per day. I vibe coded my own app that counts my keystrokes. And so yeah, I think you have to accept that just like everything else, like not everything will stay at the peak of performance every single day of your life. But the important thing is the attitude is like, I don't fall into despair if it's If it's lower, it's, it's like yet another cool thing to look at. I, I'll give you another example. So some people love to look at their heart rate at rest. I think that right now I'm probably near my like best over the last couple years. I think last year I had a really good year. A couple years ago I didn't, uh, like maybe like cardiovascular, like VO2 max, whatever. And maybe a few years before I did. Could I call that burnout? Not really. And it's just another thing that I, you know, you go through and, and you get better at. And again, I think you really have to find the thing you love. Like it's so easy for me to get energized because there's so many cool things that we can do with Vercel or we're doing every day, or I'm reading from my colleagues and it gets you, gives you energy.

Nakul Mandan:

Next up is our quickfire round. So you ready? Yeah. Most overrated startup advice.

Guillermo Rauch:

I'm such a bad person for like quickfires because I give everything too much thought. Overrated startup advice, probably I think Ambition, to be honest, because most startups will die because they were so freaking ambitious that they shipped a bunch of slop into the world. Whereas I think you have to like moderate your ambition. It's almost like it's your, it's your secret. It's your, my secret sauce is that after I get to this milestone, I'm gonna go 100x.

Nakul Mandan:

So one product that is not yours that you wish you had built?

Guillermo Rauch:

Nowadays I'm building so many cool things just for myself that I'm, uh, but is there a product that gives you delight? So Google Chrome, it's a marvel of engineering. You know, there was an essay recently about we use an AI to recreate a web browser. No, you didn't.

Nakul Mandan:

No, you didn't. You just copied Google Chrome.

Guillermo Rauch:

Not even. Like even copying it is hard. Just the CI pipeline is an engineering marvel. Like I could go on for hours. This is what I do for my fun time. Like I go deep into like research of like I learned yesterday of how they apply different optimization profiles for binary size in certain parts of the architecture and for performance in certain other parts of the architecture because just hitting a certain size of download at that level of gigantic software scale is so difficult. And just like the security work that they do and like when, you know, remember when they were gonna sell it? I was like, I was telling my CFO, okay, We have to buy it. Like I'm pro-regulation now.

Nakul Mandan:

Most overhyped thing in AI right now?

Guillermo Rauch:

Most overhyped. I think voice, controlling your computer with voice, North Star, love it. The way that I see it, and I don't know if it's just my old age typing habits, 'cause I love typing, whatever. Like I just don't see it. I think there's moments where it makes a ton of sense. But we're not, we're not there yet.

Nakul Mandan:

The last thing you built with your own hands, no agents, last product.

Guillermo Rauch:

That's a great one. Oh, so one of my biggest pet peeves in life, you know what burns me out? When my flight doesn't have Starlink. And so there was a flight recently with no Starlink and I had to open Neovim again and write code like a stone age caveman I think I was trying out something in Next.js. I can't even remember what I was building, but it was quite the experience, by the way. And like, you know, as much as like coding is outdated now, I do think it's really useful to like every once in a while, just go back to like the code. And I'm thankful for that experience. I'm not thankful for the terrible Gogo Inflight whatever bullshit they gave me.

Nakul Mandan:

What's your favorite thing that has come out of Demo Fridays?

Guillermo Rauch:

Oh, that has shipped in the product or—

Nakul Mandan:

Or that you've seen, maybe just a demo also, like what's the favorite one that you remember?

Guillermo Rauch:

Oh yeah. So I mean, going back to like the heart of the company is enabling everybody and every agent to ship. We're making deployments so fast that it's hilarious. Like every day I'm hearing about, we shaved 7 seconds, we shaved 1 second. We just shipped a new version of Next.js that cuts down build time sometimes by 50%. And like, I really get a kick out of, I'll tell you, my, when I was a kid, the experience that maybe blew my mind the most was I could FTP. Remember FTP? You drag, you drag and drop a file and now you have a live application on the internet. And some FTP clients allowed you to edit the file and you'd save it. And now your application is edited live. And that's what I've been chasing. Modern software engineering has multi-hour deployments at some enterprises. I heard from a company the other day, 50 minutes to ship a change. And so at Vercel, we're aiming for single-digit seconds for everybody in the world at any application size scale.

Nakul Mandan:

So this came out of a—

Guillermo Rauch:

And so on Demo Fridays, I just saw another massive improvement the other day that for many, many classes of applications, we're already in the single-digit seconds and like it's It's gonna become real time and this is gonna be crucial for the agents 'cause they need that fast feedback loop.

Nakul Mandan:

One responsibility you'll never give to a machine.

Guillermo Rauch:

Anything that deals with the emotional side of like being there for my family or everything, I guess like the horror movie, like love and all those things, like even writing, by the way. So someone asked me yesterday, does AI write your tweets? I said, hell no. AI reviews my tweets, gives me information. I mentioned the research stuff. So the way I think about it is next token prediction is amazing as a mental model. What comes out of my mouth is the next token. The system prompt is everything, is the room that we're in, is everything that happened today. It's my DNA, it's my upbringing, it's Argentina, it's the USA. It's 3 World Cups and just kidding, it's all of these awesome things. And so there's no substitute for, you can't codify that into a system prompt. There's just way too much information. And I think the best connection I'm gonna be able to make with you is if you get my legitimate, authentic farm-to-table next token from me, not from a machine. So writing is the thing that we have an alpha on.

Nakul Mandan:

Final question. Somebody watching this, somebody who's an immigrant or somebody who's in Argentina or India watching this wants to life you've built, what would be the one thing you'd tell them that this journey actually requires?

Guillermo Rauch:

You know, I remember my early days in SF, uh, before I was married, and I don't know, I can't compare myself to other people because I don't know what their habits were, but I think the level of discipline that I had in terms of like my entire life was like so well curated for learning and working. And, uh, it obviously still is, but like It was Monday through Monday. Someone, someone tweeted the other day, like, are people really working out here? And it was kind of a troll, like in, in like Soma and whatever, like our startups, because there's so much grind slop material. I'm really proud of the dedication that I've put into it because I have had some gifts. Like I was good in math at school and I won the like local math competition at school and whatever. But the number of hours put on top of that, I think like, I joke like 10,000 hours, 996, that's rookie numbers. Like, what are we talking about? Like you have to put in the time.

Nakul Mandan:

Yeah. Yeah. Great. Guillermo, this was so amazing. Thank you so much.

Guillermo Rauch:

That was fun.

Nakul Mandan:

Yeah. Yeah. Amazing.