AI Without Borders Ep. 18 | The Self-Driving Agentic Infrastructure with Vercel CEO Guillermo Rauch

AI support agent now handles 82% of tickets. When Karpathy tweeted that, my main advice to Vercel is the tools that you thought were the right tools for AI, even just a couple weeks ago, might not be the ones. There is new bets that the company can afford to work on. You have to say, I have this hypothesis for the future, and once upon a time, v0 was one of those bets. Not every bet is gonna pan out. For 1,000 bets that fail, I just need one winner. Is a bitter lesson for agents. If you're constraining your agent too much, you may think that you're too smart, like, well, I'm going to put this agent into this tiny little box and I'm going to create a whole universe around it where I decide other rules and whatnot. My cautionary tale is that there's only two business models left in the world of software.

Hello everyone and welcome to a new episode episode of AI Without Borders. I'm thrilled to have Guillermo Rauch, founder of Vercel. Vercel is a platform that builds frameworks and infrastructure to create and deploy web applications. But most importantly for this podcast, Guillermo is originally from Argentina. Guillermo, welcome.

Thanks for having me.

I wanted to start with something Vercel posted about in the self-driving infrastructure post Vercel paints a future where the platform evolves from a passive layer into a self-healing system. Where do you think AI is already good enough to run parts of that loop end-to-end autonomously? And where are you seeing more failure points that make that autonomy not possible yet?

Yeah, this is a great question because I've always been quite outspoken that companies trust Vercel. As the most important infrastructure of their business. Because every business, even, you know, 100-year-old, 200-year-old businesses, or even governments are becoming more and more so online entities. Just yesterday, the US government announced the new food pyramid and it was delivered as a website built with Next.js, which I'm really proud of. And this is happening across the board. Every company's— every company, whether new or old, is becoming an internet company. And increasingly these companies are becoming AI companies, which is very, very exciting for us. And the goal with Vercel and our self-driving infrastructure is to serve them. But because this is critical infrastructure, I've always been, as you know, super AI-pilled. I'm obsessed with AI. I use it all the time, but when it comes down to dealing with security, privacy, availability, durability, I've always been more analytical. Would I give an agent full access to all of my secrets and all of the customer data and all of that. And so I'll tell you what we've learned so far. One of the things that Vercel does and why we like to call it self-driving is that anytime something breaks in production, you start getting errors, even anomalies. And what I mean by an anomaly is suddenly you have more traffic, suddenly you have more load, perhaps you weren't expecting it. Vercel is automatically detecting that. If you compare that to the world before Vercel, engineers would first write their software, host it somewhere, and then if they were really disciplined, which is, as you know, developers, we are not always super, uh, buttoned up in everything, you had to remember to configure alerting and you had to handpick the alerting threshold. Because the reality is that software, and perhaps the people that don't write production systems for a living don't know this, there's always some amount of noise in the system, a crash here and there that to you, the engineer, you decide, well, that's okay, that's acceptable. Maybe to give you an example that's super relatable, people all over the internet install the weirdest browser extensions and Chrome extensions. And sometimes those Chrome extensions break the software and the engineer gets notified, hey, this random extension to capture online coupons broke mybankingwebsite.com. And so the engineer makes a judgment call there and says, that alert is not interesting to me. We do not support the coupon Chrome extension breaking our website. And so what you want is a very adaptive, intelligent, and dynamic system that surfaces to you the really important insights from production. And so what Vercel Agent does is, for example, if you're using Slack or other systems that we can notify you through, it'll say, hey, these are some errors or some anomalous activity that you should be paying attention to. But it doesn't stop there with just alerting. It does an entire AI investigation. It correlates what just happened with a lot of other data points. And what's fascinating is this is already how we're running Vercel.com. So one of the principles at Vercel is that we use our own product first. So you were asking me, how did you gain conviction that AI was at a point where it could be trusted with such an important thing as monitoring our production systems? And the reason that we gained that trust is because We created AI evaluations, which is sort of, I call it, is the exams that you give the AI in order to determine that it's good for a task. If you compare it to the domain of people, whenever you hire someone, you give them training, and once they're trained, maybe you certify them. And then once they're certified, you can say, okay, you can be on the factory floor performing this really mission-critical duty. So AI companies We are doing that same thing for the agents that we bring to production. And then Vercel itself is sort of the first customer of this capability. And by the way, this is just one of the ways in which Vercel is self-driving, and there are many others, but I wanted to highlight this example because as engineers, we typically enjoy the act of building the thing, but running the thing in production is where it gets not so fun. So for example, AWS had a massive outage recently in US-East-1 and half the world of engineering had to be woken up in the middle of the night to tend to that. And so one of our perspectives is that it's the AI and it's the agents that should hold the pager, not humans.

And are there any types of errors or anomalies where you've seen AI struggle more?

One thing that I've seen recently that I'm really happy about is that AI is getting much better at knowing what it doesn't know. So the default behavior of AI in older models was hallucination. When it gets an error that has complete gibberish, it follows on because it's doing this next token prediction thing. So it goes on as, well, that must be the truth. Whereas recent models are becoming a lot more, quote unquote, introspective. They can refuse things. And this is something that we work into the evaluations as well is when you don't know something, you should say you don't know. By the way, this is advice that I always give entrepreneurs, people, my children, like the best people that you can work with are the people that say, I don't know, honestly, I don't know, I have to check. You don't want that coworker that is super confident all the time. It's just half of the time it's wrong. It's the worst. And so this is something that we pay a lot of attention to. The other thing that's really important is we are operating sub-agent architectures. So I think it's misguided to assume that there's gonna be an agent that you can toss any problem to, at least within the constraints and limitations of the technology today. So the best way to think about the Vercel agent, which is sort of the brain of this self-driving infrastructure, is that it's a group of agents, experts in different things. So we're very carefully tailoring the behavior of the agent for when it's taking on the role of monitoring the production system, of parsing the logs, of understanding the telemetry and the instrumentation. The other thing that's really important to mention is we still support all of the quote unquote traditional mechanisms by which you monitor the systems. We have Vercel drains, which allows you to get the data and put it into the traditional monitoring systems. But increasingly we're seeing that AI outperforms the traditional monitoring systems because they require so much configuration. So this is also what I always encourage people to think about is what is the theory of a system and what is the reality of a system? You always have to compare what AI can do for you today versus what the more human-centric system is doing for you today. And so there is expert drivers that would never crash, but sadly there are a lot of drivers that drive under the influence.. And so when we compare, is Waymo or Tesla self-driving better than a human driver? We have to look at the actual results and the actual data. And so this is one of the things that has given me a ton of conviction in, in this sort of transition into a self-driving infrastructure world. The other benefit that Vercel has is that, you know, I started the company before AI was as big of a thing as it is today, frankly. And we already had a vision. We couldn't quite articulate it in the, in the words self-driving infrastructure. But I wanted the cloud to be automated away. My main frustration that led me to start Vercel was all of the configuration that I had. It was more so in the deployment phase. It was also like, hey, I need to launch my new idea to the cloud that also required insane amounts of configuration. So Vercel has always come from this perspective that the infrastructure has to be automatic. It has to be self-healing. It has to scale, uh, basically when more traffic, it scales more. Less traffic, scales less, those kinds of things.

And do you think as we make that transition from this self-driving infrastructure where it not just alerts in an optimal way, but starts to self-heal and the SRE role becomes a little bit of an approval mode of like what the agent proposes, that the amount of SREs needed will be massively reduced?

Well, I'll tell you what our perspective is. So when these alerts fire, you get the AI investigation and our perspective, this is why we call them investigations. Our perspective is that they can propose policies policy changes to a human operator. So this is also how we think, you know, we will disrupt the world of security, because what we notice is web application firewalls are a cohesion or a collection of human-written rules for what looks like malicious traffic. The entire security industry was built up on this idea of a firewall with an enumeration of rules. And it kind of reminds me of Hilbert's program versus the incompleteness theorem. I would argue that it's impossible to produce a collection of rules of all the bad traffic in the world. What's good traffic to you might be bad traffic to me. And this is once again where AI will vastly outperform human operators because what these anomalies are saying, I, I read one yesterday. I, as a CEO, enjoy reading these investigations because I'm bewildered by what AI can do. And it was this analyst analysis of you're getting all of this traffic from a Tencent network in China. The, the agent is really good at analyzing the, the traffic logs for security reasons, and it was analyzing that the ASN, which is sort of the origination band of the IPs that are accessing your website, who issued these IPs? Well, those IPs came from Tencent in China. We don't want traffic on that particular endpoint, huge amounts of traffic coming from that particular network, it made no sense. And so I'm able to apply policy suggested by the AI in real time. There is a world where, let's say that the human is not available, going back to AI models becoming so good at this sort of like judgment call situations, there might be a world where, look, if I'm not getting a, uh, judgment call made by a human,, it might be so important that I make a decision for that traffic that a further automation is deployed. Right now, we're very much in the world of an investigation produces a policy proposal or a pull request, right? Which is very important as well. So when something is defective, when something is wrong in the code, we can open a pull request and say, we believe that this— you're missing this check. For the engineers that are listening, something that happens a lot is you get a 500 error, which is, you know, a bad error. Like it makes your stomach sink a bit, but sometimes they're harmless. It's because you didn't handle a certain kind of input and someone created a malformed request. And what I mean by this is that imagine you create a form that says first name and last name, and in first name, some people Teenagers do this, but also bots do this. You try to game it and you paste 5,000 emoji into the first name. And so your code didn't expect that and it crashes or it throws an exception and whatever. AIs are becoming really good at saying that 500 error you just got, that is not as concerning. That is just some malformed input. It's someone trying to mess with the system. I can open a pull request and fix that for you, dog. You just accept it. So that's kind of where we see the world evolving, where it's self-healing because the AI has really good judgment. It has a lot of really good context that Vercel is giving it, but it keeps you firmly in the loop in making these important decisions to your systems.

And I, I wanted to ask you a question related to this agentic future. I would say we're starting to see more agents being deployed. For the majority of companies still in the small hundreds, maybe low thousands of agents. But if we do get to a world where we have millions of agents deployed per company, do you feel that the current infrastructure and way the internet was built still holds? Or alternative way of framing for you is if there is one or two things that you would do differently if you were starting to design the infrastructure from zero.

As everything is changing, there's infrastructure that you've built where I feel like we've gotten lucky, or maybe it stays scaled bug, who knows? But for example, when we started Vercel, I was obsessed with making it the easiest to use command line interface on the planet for deploying software. If you have been with us for the long journey, you might remember we initially called our CLI Now because I'm a very impatient person. So I would write some code and I just wanted to press a button called Now and put it in the cloud. Turns out Coding agents are computer use agents. Claude Code, Codex, Open Code, the agents that people are using to write software also know how to run programs. Turns out this is one of the situations being in the right time and the right place. The Vercel command line interface is awesome for agents. Because they— so I mentioned on Twitter the other day, during Christmas week and New Year's week, we had record usage of the platform. And the reason for that is v0, but it's also that every other coding agent in the world, when they— when you write your holiday hack and everyone's been vibe coding, when you tell it to deploy, they like to deploy it over Vercel. It's very easy to use for the agent. And so we're going into a world where agent ergonomics And agent accessibility is a priority for your business. If you're a developer tools and infrastructure company, agent ergonomics means better command line interface. It means more accessible documentation in the places where AI expects to find it. It means that you design the APIs in a certain way. It makes you mindful about error messages, not just for humans, but error messages for agents and things like this. On the infrastructure side, I also think the change is magnificent. One of the things that has been becoming also obvious as of the last couple weeks and months is the rise of the sandbox. So agents can make a lot of mistakes. They can delete files, they could be hacked, they could be coerced, it could be, you know, they can get prompt injection attacks, et cetera. So we needed to develop a capability so that we could sandbox them. To sandbox it means to make them— give them the illusion that they can do anything, but help the developer basically create a security boundary. And so within the sandbox, the agent feels it can do anything, sky's the limit. But then the developer from the outside world can communicate with that sandbox and extract value from it. And so this is one of the new primitives that Vercel has been investing very heavily into, Vercel Sandbox. It works with any agent., and it allows companies that are creating agentic experiences to give the LLMs this ability to use computers. Because what we've learned, you were asking things that I would've done differently. When we designed v0 originally, we didn't give it the full power of a computer. We gave them the power of Next.js with some constraints running within the browser, knowing where the— if I had had the perfect 8-ball, of course, or the perfect magic ball, Agents were headed in a direction where they were getting really good at using all of these operating system tools, these command line interfaces. They can test, build, QA, search the web, et cetera. So you have to be kind of simultaneously fearless and fearful. Fearless in the sense that the more you give your agent, the better results you're gonna get. It's a bitter lesson for agents. If you are constraining your agent too much, you may think that you're too smart, like, well, I'm gonna put this agent into this tiny little box and I'm gonna create a whole universe around it where I decide other rules and whatnot. You might be underutilizing, you might not be getting as much bang for your buck from the model because you're not giving the agent enough tools to work with. Imagine you hired a, a new coworker. And, and by the way, funny enough, in, in management, I've learned this to be a huge frustration that employees can get when employees don't get enough information. When they feel like they're getting siloed, when they feel like they don't get access to the tools that they need to use and they have to like ask IT for permission and access, et cetera, it makes for a very frustrating experience. You feel disenfranchised and disempowered as an employee. Agents can also be similarly disenfranchised. Agents, you only get to emit text like old AI chatbots. And by old, I mean literally like ChatGPT, 3 years ago. If you would give them a universe, if you give them an operating system, then agents can be way more effective, but they can still have a lot of security controls. And so that's the idea of the sandbox. We believe that this is going to be the future of compute. Why? Because the old world of compute was you, Daniel, and I would write a program, we would ask AWS for a computer, We would transfer the program into that computer and it would run, and that process would take us a week. Then Vercel came and we made that process instead of taking a week, take literally like less than a minute. But it's still all human mediated. It's you and I got together, we wrote some code, we pressed enter. The next, and even today, is agents are writing the code, agents are testing the code, and agents are deploying the code., and all of this is happening in cloud computers, which we call sandboxes. And I believe that if you think about how many Daniels and Guillermos existed, and many times they had the ability to say, okay, the software is written, let's ship it. That was not that frequent. Like there was only like tens of millions of us and we needed to write all the code by hand. And then we would decide to deploy it. I think there's going to be billions of agents, all of them using computers, all of them creating new computers, all of them creating new software. I don't think the world has realized yet just how much more demand for compute there will be. And I think this is partly because so much of the attention has gone into the GPUs, but I would make the case that the GPU is actually not that interesting in a sense, because the GPU is streaming all of these tokens. And this is why, you know, Fluid Compute by Vercel made it very, very cost efficient and easy for LLMs to stream tokens into your compute. But now these agents are going to have to use a lot of computers and tools, and these are going to be the CPUs and the more traditional computers, and there's going to be a demand for it, the likes of which we've never seen before.

And you mentioned there in December, you saw a massive activity in people using v0, and I wanted to to ask you, what do you think are some of the second or third order effects if we reach a world where we are currently, where every creative person with an idea can build an application beyond the obvious? And before starting the podcast, I was mentioning one, for example, on web browser agents and how does a world look like where millions of people are using agents to get World Cup tickets and how ticketing platforms might have to adapt and change. And if there are other things like that, that people maybe are not yet thinking about?

Mostly really excited. So when you mentioned that example, when we were getting ready for this podcast, I immediately thought of a coworker. We created this Slack channel called World Cup. I'm from Argentina. We all want to like see as many matches as we can because it's happening here in the US. And one of my colleagues V0'd a real-time dashboard that monitors prices. Oh, just let that sink in. Like, he has a busy job. He was just able to make a couple prompts to make himself more intelligent, to share with his colleagues, to add value to the world. This is the era of personal software. That software program, and this is good to contextualize what I was just saying earlier, that software program that he created could have never been a standalone business. SaaS company. We're the price monitoring company of the World Cup. So there's a lot of commentary that I'm not gonna get into too deeply of like what happens to SaaS in the future? What happens to software as a service? Well, you know, I think it'll still have a place, but think about it. Someone could have come up with the idea of I'm gonna create a price monitoring platform for tickets or something like this and sell it for like $5 a month or whatever. Now we live in a world where my colleague Diego just prompted that software into existence. It's personal. It's the way that he envisioned it. Maybe I would've done it differently. Maybe not. I actually liked what he did. It's agentic because this thing is going out into the ticketing platform and he's querying for the price. And, uh, it actually, I saw the curve of prices is fascinating. So meaning that he's not only orchestrating compute when he built the software, he had the agent get itself a sandbox. He wrote this, the agent wrote the software, the agent then deployed the software, 'cause now it has a URL, it deployed it onto Vercel, but then it's running more software to periodically go and get prices. And this is why I think that the demand for compute and you could imagine like the demands on energy and data centers are just gonna skyrocket because this didn't exist before. This human potential of autonomy when in creativity, when creating software didn't exist before. And so these are the second and third order effects is we yet have not seen what an entire world of people doing this at scale looks like. I think it's still like the pioneers and the people that are in the know. Obviously v0 is democratizing software to an extent where like anyone with an idea can build it, but it's the early days.

No, and hopefully you release that world ticketing platform as we'd love to see it. And yeah, I was building actually during the holidays and I built an app where it's like for collecting some insights from YouTube videos. And it's something that I'm thinking of putting out there. Maybe people would pay like $5. Before it was something that it was not just worth making, even you would charge $5, but now it totally is worth it.

You make it for yourself and then you, you publish the prompt for somebody to make their own. So the v0 community, we're now experimenting with basically like templates that you can purchase with credits that you already have because you're within v0. And so you can, you can, as a maker, share your work in a way that's sustainable to you because so much of the human ingenuity is now being poured into the prompt. It's not so much into the software that you pay a subscription for, but it's like, hey, share the prompt with me, share the recipe. Because I can give that recipe to my agent. And maybe you already had an idea that was similar and you already have a design going on over here and you already have your own platform. So what you really want is for your agent to have access to that and merge it into your idea. And this is the world that I'm really excited about because it's like software is becoming like water. It's very fluid, it's very flexible. And the other thing that popped into my mind is Imagine a world where the collaboration between humans, where in the past used to be that I teach you something that's super technical and I sit next to you and I teach you all the APIs, et cetera, is I can help you prompt your agent. I experienced this just yesterday with one of my colleagues. I built this V0 chess experience of an AI battling another AI in chess, and I was talking to my colleagues like, so to give you some product details, imagine a chessboard that is in 3D and an AI is battling another AI. So in this case, it's Gemini versus Grok and Gemini is winning 1 to 0 right now. And the text is streaming on the right-hand side. What each AI is thinking about the moves is streaming chat on the right-hand side of the screen. And I was talking to my colleague and I was saying, hey, Fernando, have you seen that pattern where The text automatically always goes to the bottom of the scroll. But if I, if I scroll up, I don't want the text to move. Even if the AI is writing new text, I just wanna see that particular move that I was just thinking about. And he and I were going back and forth and like, what is the best way to implement that? And what it, what it ended up being is he just told me a prompt. He's like, hey, tell your agent this., and I copied it and I gave it to my agent and it solved my problem. And so think of this as a new way of humans collaborating, not on the code itself, not on the implementation itself, but say, hey, tell your guy that this is how you should fix that problem. Tell your engineer, tell your designer this. And so the world we're going into is one where we're almost all in a constant chat. With, with agents. When I, when I think about the future of the workforce, when I think about the future of the companies of tomorrow, I think of a company where you're managing agents and it's you and potentially tens, hundreds, thousands of agents working for you with lots, lots of context on your business, lots of context on your preferences, your tastes, and there will be other humans., but I think a lot of the time you're gonna be spending directing these agents.

I absolutely love that concept around that prompt collaboration and not being in software. And, and related to this topic, I was recently hearing the podcast that Lenny did with Jean DeWitt where he shared how the go-to-market team at Vercel is using AI. And it seemed that Vercel, you are using a lot of AI to build internal ways to become more productive. I'm curious, how do you analyze decisions now internally that famous buy versus build for internal software? And you have any frameworks or first principle thinking that you recommend other leaders to adopt when making those decisions?

Yeah, one of my main principles right now, and obviously v0 has been a spear in this movement, is that everybody builds software. If you're part of a tech company, everybody, the expectation is everybody's building software. Why? Because the bar to contributing to software has gotten lower and lower and lower, and it's basically now the quality of your software is a function of the quality of your ideas and thoughts. And we want to hire people that have brilliant ideas and brilliant thoughts and are willing to experiment and are great communicators. That's how I expect the Vercel team to behave. One of the things that I always wished for the world, and I took a lot of inspiration when I started Vercel from Google. Google was very set that everybody had to be an engineer. When they first had their like famous like Google lunch service, they were talking about how like even I think one of the first head chefs was a former engineer. Why? Because they had this mindset of like, if our company is building software and scaling software, we should all know how to build it. And I, So Jean runs our go-to-market function at Vercel, among others, and we always wanted to have the world's most technically sophisticated go-to-market force. Why? Because people come to Vercel to partner with the creators of the most important tools in the web development ecosystem, like Next.js and React and Svelte and Nuxt. Contact our sales team, I expect them to be extremely proficient in our own technologies. And so we started this go-to-market engineering effort. And so go-to-market engineering is all about creating agents that automate sales and marketing at Vercel, that automate the collection of data so that we can be more helpful to our customers. So let's say that you show up, Daniel, and you're like, hey, I'm interested in scaling with Vercel. I work for a very large organization. We're using it right now in one team. We started out with frontend. We wanna do more backend on Vercel. We wanna build agents. Can you help me? I wanna partner you with one, the person that knows most about your business within Vercel. Second, I wanna give that person the most context about you and your business. Imagine I get on a call with you, I get on a Zoom with you, and you have to explain to me how you use Vercel. Why? Like we, we have the data, we are trying to be helpful to you. We should have context on how to be most supportive to you. So all of these processes in the past in go-to-market organizations were human mediated. For example, routing. Routing means like you, someone contacts Vercel, how do we route them to the right specialist? How do we prioritize?? Because someone might be contacting us because they, they're urgently upgrade their software because they're on a legacy software stack that has security problems. So when you give that higher priority, for example, or even better, what if you don't have to contact the human? What if we can route you to the Vercel agent proposed solution? So what's been fascinating and one of our greatest successes with AI that I just keep getting mesmerized about we're going to be sharing the technical notes soon. Our AI support agent now handles 82% of tickets. I actually monitor this. I have an agent that tells me the numbers weekly. And so I know in real time how many, and this number just keeps climbing higher and higher and higher. And what's exciting to me about this is everybody that reaches out to Vercel By the way, some— the engineer that built the Vercel support subagent shared a screenshot with me yesterday. He said, I'm gonna see if I can pull it up. The quote from the customer was, this is the best customer experience I've ever had. I don't know if that particular human knew that they were talking to an AI agent or not, but the way that that person phrased it was beautiful to me, to my ears. I had never had such a good customer support experience so far. It didn't matter that person, if they were talking to an AI or they were talking to a human, goes back to that metaphor of like for self-driving cars, it's all about quality, getting to where you need to go safely and successfully in how they fare against their human counterparts. And so The role of go-to-market engineering at Vercel and Vercel as a whole is you build these agents. We don't go and do the task ourselves. We built the routing agent. We built the agent that sets up the sales rep for success with context. We built the agent that analyzes how well we handled a deal. Could we have done things better? Could we have been more helpful? Could we have communicated an estimation of how much it was going to cost faster? Did we reply to the customer expediently when they reached out to us through our shared Slack channel? Emails. So all of these things used to be humans, and now we're delegating more and more and more to agents. Now, the role of the human continues to be you oversee these agents, you manage them, and you create new ones.

I understand that at Vercel, a lot of the tooling now you build yourselves. Are there any characteristics on the software that you actually do decide it is better for me to purchase from a vendor versus us building internally that you think about?

Yeah, so there's two categories of agents in my mind. One is what I call AS or agents as a service. It's kind of funny, but it's also true. It's an agent that you buy from somebody. ChatGPT is an agent as a service in my mind. I think it's a starting point for most organizations is they get something that's off the shelf. So v0 is an agent as a service. You, you got it and you start using it right away. There's a lot that you can do with platforms like v0 to customize them to your needs. So the next step is typically you add more context and integrations into those agents, but then you also start realizing that using v0 and using ChatGPT dislodges something in your mind. It makes you realize that there is a new way of using computers. It teaches you to prompt. It teaches you about how these agents are, you know how like if you pay attention to these videos, it says thinking, it shows you how agents reason. It tells you how they're using tools. If you dig under the covers, you see how it uses these sandboxes, how it runs code on demand, et cetera. And so that tends to inspire people because then you start thinking like, oh wow, I could create my V0 for this other thing. So one example for this internally at Vercel is we have in Slack a data analyst agent. It's called, as you could expect, we're very creative, D0 is our, is our V0 for data. And we can ask any very hard question of data from anywhere within Vercel. And we build that ourselves. Same with our support agent. One, because the Vercel infrastructure gives us all of these primitives to build agents like these.. It gives us the sandbox, it gives us the workflow. A lot of this is open source. We have the AI SDK, et cetera. And so it's very easy to build an agent today, but also we needed the agent to integrate in a very specific way to all of this private systems. And so it's hard to purchase something like D0 out of the box. In the case of the data analysis, it had to work with our very, very specific data systems. And so it made sense in that case to build an agent. Now what's really cool is that agents can collaborate with other agents through the MCP protocol. So V0 will be able in the future to call D0 if it needs to do data analysis, and that would be internal to the Vercel workspace. ChatGPT, we launched a template on how to deploy your own integrations into ChatGPT. So what we're going into is a world where some agents you purchase out of the box, but I do think companies will be building a lot of them that are custom to them because of that inspiration that they're gonna get.

Now, I wanted to switch gears on the topic of leadership. In the past, you appear in the Acquire podcast and you spoke about the importance of being world-class at one thing. I also had before in the podcast the CEO of HAVI, who said that the most important task for a CEO is deciding what the company should focus on. With AI, I've seen now it's become much easier to do more things. In a world where everything looks like a shiny object, how do you as a leader ensure that your organization keeps focused on the right things and not get distracted by doing too many things in an average way?

100%. So, one is Because Vercel builds infrastructure and platforms, we have the privilege to always be working on Vercel. So even when we experiment on a new idea or a new project, we're always strengthening the basic foundations of the Vercel platform. So on one hand, I do tell a team, hey, please go out and use all of these coding agents. Use v0, use Cloud Code, use Cursor. Why? Because I want you to make Vercel better for agents. One of our values at Vercel is for the web. We, we wanna make the web better, but another thing we've been talking a lot about it is we need to build for the agents. It's the agents that are gonna be building so much of the future. And so the way to think about it is that Vercel has core bets. Our core bets are Vercel is the best infrastructure for agents. Vercel is what agents like to use, but it's also when you have an idea for an agent, I want you to think of Vercel and I want you to think about v0. And so we have those core bets for the company. We have the bet of running so much of the internet infrastructure at every pixel and every token of the internet, Vercel flow through Vercel. And like I mentioned in the beginning of the podcast, you know, that's a huge responsibility to the world. That's also where, you know, vibe coding, you use it, but you have to be careful and we need to be stewards of the web and this important internet infrastructure. On the other hand, there is new bets that the company can afford to work on. You can't have a million of those, but you have to say like, look, I have this hypothesis for the future. Let's place a bet there. And once upon a time, v0 was one of those bets. It was early days of ChatGPT and we said, hey, ChatGPT could build things. Why does it just have to answer questions? Why don't— where are we not even more helpful to our customers? And this is sort of how we, and by the way, even when those, not every bet is gonna pan out. For 1,000 bets that fail, I just need one winner. Now, like I mentioned, the really good thing is that all of those 1,000 bets work on top of Vercel. I'll give you an example. Every agent in every AI interface needs to get tokens from somewhere. Typically people have said, well, I'm gonna get them from OpenAI. Because they were kind of the first mover. But every week there is a new model that will surprise you or be helpful to you or be more performant or more cost efficient. So we placed a small bet and we said, what if we create something that's almost like a CDN for tokens? It's a gateway to tokens. Like you get one key and access to every model in the world so you don't have to configure them and procure them by hand. And the way that we built it is very interesting. We published about this on our blog. We built the Vercel AI Gateway as a Vercel project that you or I could have created as a side hustle, and it's now one of our most successful bets, and it's one of our top priorities for 2026 is that the way that Vercel will help every developer and every agent get tokens is by providing this amazing experience through the Vercel AI Gateway. So for example, when I use Claude Code, I can get my tokens through the Vercel AI Gateway. I don't even need to sign up for Anthropic. If tomorrow I want to use Codex, it can come from the Vercel AI Gateway. If I write a really cool support agent and today I say Gemini is the best, most empathetic, maybe for my language in Spanish, it seems to be the model that performs the best in Spanish. I get it through the Vercel AI Gateway. So that was a nascent bet. Had it not worked out, it would've likely strengthened the Vercel infrastructure. Why? Because now we're stress testing the global compute platform, we're stress testing the network, we're getting feedback about how our developers and their agents think about our tools and infrastructure. And so there's a lot of these bets that tend to be win-win for Vercel.

Those bets in the future was a good way to view it. And now with AI, It's very exciting, but at the same time, everything's moving so fast. Andrej Karpathy tweeted during the holidays that even he feels that he's behind. I'm curious to ask you, even you that are at the forefront of this feel that way. And if you have advice for people, like even like myself, where it's like a mixture of excitement on one end, but at the same time, you always feel like you're behind and everything is going so fast.

Look, I've been in the software world for a while and I have had moments where I felt like that. So when the cloud and mobile were taking off, I had that same sort of like anxiety to move and learn. And Vercel was born out of that trepidation. I remember, so my previous startup was acquired and the company that acquired it was running their own data centers and I was seeing the takeoff of the cloud. Because prior to the acquisition, I was— my startup was running on AWS and I was seeing the takeoff of the cloud and just, I felt like I was falling behind. I was like, I need to make something out of this cloud thing. I need to go and make it easier to use, more accessible. I need to build on top of it. I have to ride the wave. And I felt the same set of feelings. I think they're just becoming more frequent now. With AI. Like it used to be that big shifts in the ecosystem happen every once in a while. The next one that I could remember was servers to serverless. That happened within the cloud wave is, oh, like, look, turns out the software can be a lot more autonomous. I was mentioning that idea of like self-driving before AI. Like I felt that sense of like, I need to, I need to lead in this. This is an opportunity for me to lead. And so you can turn that feeling behind, uh, kind of, uh, set of feelings into opportunities of, well, there's a lot of other people that must be feeling this way. So can I become a thought leader? Can I build a tool that simplifies it? Can I use that tool to my, for my own benefit? So I, if I become really, really good at it, I will be more productive and more prolific. You kind of see this already on X where I see this with our V0 ambassadors, the people that are experts in our community. Like, I'm really impressed with how fast they can produce super high quality things. Clearly they're already V0ing better than me. So you need to stay on your toes. That's my main advice to people. I, when Karpathy tweeted that, my main advice to Vercel that I'm going to reiterate this coming Friday is the tools that you thought were the right tools for AI, even just a couple of weeks ago, might not be the ones. That's kind of the mentality that you have to have on a weekly basis. These assumptions have to be revisited. And then, and the newest one and the most profound one is what's happened with Opus 4.5, which is now the default model on v0. And the way that this model has already changed the way I think about AI is quite profound. And I think that's also what evoked those feelings from Karpathy.

I love that constantly updating mentality. And I also know you're a very active angel investor. With generative AI, is there any one trait that you think now is more important than ever that you look for in the founders that you invest in?

Yeah, so I think AI engineering as a skill is already sort of a reality. Like if you come to me and you say like, I have like 20 years of software engineering experience and I dabble in using agents, I would, you know, I would put a question mark on that. I would be like, I don't know, man, you have to get up to date with the realities of how software is built today. So I look for proficiency in this very uncertain territory, as we were just discussing with about Karpathy tweet, within this emergent world of constantly feeling like you're behind, there are patterns that are emerging. There are tools that are being created. We're creating a lot of really cool tools at Vercel for orchestrating these agents. We're experimenting a lot. And so I look for founders that have already kind of created those kind of tools for even for their own personal use. If you've created an agent or an agent hardness or you have some novel insight into how to scale these models, then I'll pay more attention. Generally speaking, my perhaps exaggerated, but cautionary tale kind of bet is that there's only two business models left in the world of software. You're either building an agent or you're building agent infrastructure. Unclear that there is another thing that you could be building today. And so if you're thinking about disrupting an existing ecosystem or business, you have to think about in terms of like one of those two things.

One topic that you're— I know you're passionate about is generative UI. I was curious to get how do you envision the future of web interfaces is going to be on one end. With generative UI becoming more popular, ideally you have more personalized interfaces for each person. At the same time, I've been thinking if every time, for example, you log in in your CRM, you want it to look differently, or if there's some benefits of some degree of standardization.

So the generative portion is going to continue to eat more and more and more into the world of static software. One of the things that positioned Vercel for success in this AI wave is that we were betting that the world is not going to be static. It was going to be dynamic, and I see generative as the evolution of dynamic. Dynamic was some clever programmer said if else, if else, if else, boom. Dynamic also meant even like product recommendations on the Amazon product listing page and things like that. Generative is that based on input and context, we produce an entirely novel outcome. I would argue that a really good AI chatbot, for example, the way that we want Vercel Agent to serve you is that you can ask any question of the Vercel data. We launched kind of a glimpse of that yesterday with you can now chat with your Vercel integrations. So typically when you install an integration into a platform like Vercel, it's kind of annoying because if I installed the Supabase database, I don't wanna go to like the Supabase UI and log in and figuring out where my resources are. Well, you're in luck. Now you can go and say, hey, where are my Supabase databases? And what is the latest table that I created or my agent created? So the answer to these questions is gonna be generative and it's gonna contain UI components and UI elements that are not used in the response to a support ticket, for example. The other consequence is when you talk to the Vercel agent somewhere else, like Slack, we'll also produce novel UI within the context of that conversation. When you come back to Vercel, we will try to estimate and imagine what are the things that you want to see when you come back to Vercel. When you go to vercel.com, that experience will also proactively be more generative and tailored to you. And again, you need the right— if you're thinking about how this applies to your company or your business, you need to have the right foundations in in place. You have the right frameworks, you have to have streaming infrastructure, for example, like Vercel Fluid Compute. Because if you were relying on CDNs that use caching and things like this, well, now everyone is getting the same page and that will train the customer in the wrong way. If, if every time I go to your website, I get the same thing, then what reason is there for me to come back? So I think people are freaking out a lot about like the demise of the web and like, Will anyone use interfaces? Yes. If there are static interfaces, no one will use them. If there are generative interfaces, people will keep coming back because they'll get more and more value. And now what a lot of people ask me is what is the difference between a UI that gets generated on demand and what do you do with systems like v0 or Claude Code, etc.? Well, I think increasingly what you're going to be using v0 for is for creating agents themselves. I don't think you're just going to be using v0 for saying like, build me a marketing page. That's a static What is the point in that? That is the world we're trying to move away from. We don't want to just create like things that are just completely like inert. And so what you should expect from these systems is that you're gonna be giving birth to systems that themselves are generative. It's very meta. It's like recursive. And I will expect the entirety of the web to become self-improving. There's a platform right now being built in Vercel that I'm really excited about called Flint. Their idea is that when you use their agent, your marketing website will become self-improving. The page itself, the content itself will become self-improving. This is like a dream for every CMO because a marketing leader knows that they have like to constantly update marketing copy on the website and it gets out of date and you get new customers. And you have to update the logos and the CMS. It's like, it is kind of like you dread going to the CMS. Who wants to go to the CMS? Lee from Cursor recently tweeted like, we just want agents to change the website. So we just wanna tell them, but what about if you didn't even have to tell them? The website itself is adaptive and self-improving. We are experimenting with this for the new version of Vercel.com. Where if you're accessing the website from a certain part of the world or you entered from a certain part of the site, we can tailor the entire experience to that session.

Very exciting to see that released. And this is why I love doing the podcast, Guillermo. A lot of golden nuggets in the conversation. I appreciate you, you taking the time.

Thank you.