In those 10 years, a billion deployments — it was mostly people writing code. We made it really easy to ship. That's why we're calling this event Vercel Ship.
But now in the 10 months since, we added another 1.5 billion deployments. And now we're adding a billion deployments every four months. Why is that? Coding agents.
Hello everyone and welcome back to MTS. We are here live at Vercel Ship. This is so fun. So excited to get to talk to you again, Guillermo. And for anyone who's not familiar, which you should be by now, Guillermo is the CEO of Vercel. And we're going to talk about all of the fun things around how agents are changing the internet today. So, welcome to MTS.
Good to be here. I don't know if you knew, we power MTS online.
So, you guys are also using our agentic infrastructure.
Hey, looks like we're your number one customer. This is fun. I'm glad that we get a chance to talk again. A little bit of lore for the audience. But Guillermo, you were my first ever interview on MTS. My first day was Vercel Ship London, I think.
Yeah, Vercel Ship London. Now we're here, Ship SF, San Francisco. It's the best.
It just keeps getting better and better and better. We're ascending every time. Next time we chat, we'll be on the moon. We'll have our next colony on Mars and we'll do Vercel Ship Mars, which will be pretty cool. Um, you had a keynote earlier today and you were sharing a lot of big ideas about where we are in the future and where we are now. I think a great place to start is if you can just establish when it comes to agentic coding, what was the first phase, what was that second phase, and then what's today's phase like? Where are we?
Yeah, you know, it was a little bit of a trip back in time. Not too long ago, we were copy-pasting from GPT into our text editors. Back in the day, we were writing code in the stone ages. Yeah. And it's been awesome to reflect on the acceleration because after that we started getting more into loops, right? Later came the copilots, later came Opus, and now we're at the point where like we're more hands-off and we're letting the agents rip, right? And so Vercel has been kind of, you know, at the forefront of this. I mentioned we had a billion deployments in the first 10 years of Vercel.
Fun fact, we're here at the Palace of Fine Arts. Upstairs there used to be a co-working space.
It was my first office. This was my first office.
So, you're back home.
10 years later, I had investor interviews upstairs. And so in those 10 years, a billion deployments. It was mostly people writing code. We made it really easy to ship. That's why we're calling this event Vercel Ship.
But now in the 10 months since, we added another 1.5 billion deployments. And now we're adding a billion deployments every four months. Why is that? Coding agents. Yeah, they're ripping. And we love to have them. I mean, this conference is for humans. We're human, but behind the scenes, we have agents doing recursive self-improvement, writing code, patching code, reacting to events, and shipping code, and making products better.
Yeah. Let's just recap that stat one more time. So, in 10 years, you hit your first 1 billion. And then the last like 8 months, you hit your next — or like now you have like a billion every four months. Now we're adding a billion every four months. The acceleration continues.
I mean, for a lot of people, enterprises — I mean, this is technically an enterprise event, right? Like we're hosting some of the greatest brands in the world across every vertical. For example, retailers. I was just speaking to 10 large retailers outside. We have Tobi, Shopify, fireside chat later today. That's going to be a banger. A lot of that world, they're still catching up to coding agents. So I mentioned that in my keynote as well, right? Like being a developer, you're taking it all in. You're moving really fast. So, it feels good to be at the forefront of all of this. But just think about the acceleration in years to come, right? As agents are also really good now, right? Like a lot of the, you know, hearsay was in the past, well, they can't do this, they can't do that, they're kind of shitty at this, they're kind of shitty at that. But look at them today, right? Like it's not true anymore.
You're hard-pressed to find a specific vertical where an agent is not outperforming you.
Now, as an engineer that puppeteers agents, I still believe in ourselves. It's like we still have a lot of value to add, but you know, we just have to let them cook.
Yeah. I want to talk about that puppeteering agents aspect because I think that's the newest thing that's been happening in the last few months that maybe wasn't totally true a year ago. You say we first started off with chatbots that you take text, you copy and paste it. You now have coding agents that run all this code for you. And this new era, it's almost like these software factories. It's like self-improvement.
Self-improvement happening now. The way that I see us adding significant amount of value to our organizations is we're going to be working on the infrastructure that begets the coding agents to be productive.
So let's call that the software factory, right? Like you're working on setting up the organization for success. So what does it look like in practice, right? Like getting the signals hooked up into the coding agents. I mentioned something that's very unique about Vercel is we have the production context. We see every request, we see every crash log, we see every exception. And the last thing an engineer wants to know is my website is down, my website is slow. So if we can automatically trigger the coding agents to sort of quote-unquote hold the pager, act on your behalf, improve the software, optimize it, even optimizing your bill.
Right. People have all kinds of different levels of expertise in how much they've tuned their software and cloud infrastructure environments. The fact that Vercel can start optimizing on your behalf, not just for performance, but for cost efficiency, is part of this software factory chapter we're going into now.
It's almost like RSI, but actually just for the agentic platforms themselves. So they can fully — you get the logs about maybe a bug someone hit or potentially an issue that's going on, and then it can self-heal. You can see, okay, in order to perform this task, my agent had to go through these five hops. When I loaded the website, I went through these five paints. If we can give that context to the agents — the traces, the OpenTelemetry data — now they're actually much more intelligent.
You know, you could argue that context is all you need, right? The intelligence is already here, not evenly distributed, because we have to inject it into the brain of the agent.
Are you seeing enterprises already doing this? Are we still in early stages?
We mentioned a few customer stories, a company like Real where they've improved the throughput per engineer by three times by starting to abstract themselves out, invest in the software factory layer. The other trend that I see, you've heard about companies like Stripe and Ramp, they're investing in building their own software factories because the coding agent on its own, without enterprise data and context, not that useful, right? But the challenge that enterprises have is how to securely access this data, how to govern it, how to keep cost under control.
Right. If you have 10,000 engineers and you're like, 'All right, guys, let's just let it rip, token max.'
Yeah. Not a good experience.
So part of our enterprise apps and agent platforms is giving every project and every agent a token budget. So you can say, all right, little bro, you get $500 to spend on this task, right? Also, even for engineers, right? Like the pushback on use more of the intelligence that's available at all levels of scale through the AI Gateway. The choice of models begets also an optimization loop in its own right, and this helps enterprise adoption because they get the guardrails needed from a data governance point of view, but also making costs scale fail.
I want to shout out our sponsor, Lovable. You know the app you've been meaning to build — the internal tool, side project, or product you'd otherwise lose a weekend to. With Lovable, you get software you can ship today. That includes an editable codebase you can inspect and change, plus two-way GitHub sync. Lovable also handles the backend and infrastructure, including managed Postgres, object storage, hosting, payments, and more, so you don't have to wire it all up yourself. And through Lovable's MCP server, you can create and deploy your project directly from the agents you already use. Turn ideas into software people love with Lovable. Lovable.dev. Now, onto the episode.
What's the hardest part about building all of this?
Well, the hardest part about building an agent, believe it or not, is not actually writing the code for the agent. So, the harness, the tool loop — there are so many different ways of shipping an agent today. I think that getting them the access, the identity, the wallet, these are the harder parts, and we're solving them with products like Vercel Connect. We announced we now have a thousand integrations, meaning you deploy your agent and now you can access your data securely.
The other thing is durability. So durability is, okay, I can launch my agent, but it doesn't crash. It doesn't time out. It retries if it fails. And so the compute of the traditional cloud infrastructure was not really designed for this agent era. It was short-lived compute. Most of the time your database was working fine. And so we launched a series of improvements: backends on Vercel, fluid compute, Vercel daemons.
To me, I was asking around and I was asking people, like, what was your favorite Ship today? I got two answers. Vercel daemons, which is this primitive for backend compute, so that something can run for as long as it wants and respond to events, and, you know, you just let it run.
It doesn't shut down when you close your laptop.
That's right. It's like an agent that lives forever. A good daemon, if you will. And the other one is Drives, because as these agents evolve, as you deploy them, as they pick up new work, they need a workspace. They need like this table where we can keep files around, and as other agents come in, we can share a workspace, we can share data state. And so we launched Vercel Drives, which is sort of — think of it as like the Dropbox for agents. They can keep files around across multiple instances of computation. And so this was a big missing primitive of the cloud, right? Of course storage always existed, but this is storage that's purposefully built for agents, for sandboxes, and for the compute they need to do in the cloud.
You know what's really fascinating about Vercel, and I was talking to Gene about this earlier, is you guys are actually dogfooding your own tools. And so it's part of your own process because you're an enterprise. You just mentioned all of these other enterprises who are using your tools within their company, but you're also a company, so you get to see in real time what's working.
We throw away so much, right? Like we get to try, fail, try, fail. And then when we get something that we have a lot of conviction in — for example, Malte, our CTO, was mentioning we have started to replace large amounts of Kubernetes clusters with running those backends on fluid compute on Vercel. And think about moving a backend the size of Vercel that's doing billions of API calls, that is absorbing billions of deployments — to do that at our scale is no small feat, right? And so I gain conviction that when I tell you, put your backends on Vercel, this infrastructure has been battle-tested. The same goes for agents, right? All of our frameworks, AI SDK, we tested them out by building agents that are, one, they're like governed, they're compliant, they have audit trails. We're not YOLOing agents into the cloud, right? So we had to think about so many corner cases of how to truly bring an agent to production that by the time we tell you, hey, here's the agent stack, we feel pretty good about it.
Yeah. One of the other big trends that's happening is we always talk about agents when it comes to coding tasks and how it's going to change the developer, but you guys are a perfect example, as you're dogfooding it, that people who are deploying agents aren't the traditional developers you might see in the past. And so tell me a bit more about how that's trending. Who are the new types of people who are building and deploying agents?
I do think we're at the forefront of a hyperconvergence of job types, roles, tasks to be done, where everything is computer. Everyone is an agentic engineer. And so when an enterprise comes to work with Vercel, we notice that there's all these different personas. You know, like you have people in marketing building software and deploying it. You have people in finance, you have the engineering team. And so what we're giving them is an end-to-end platform that is secure. They can plug in all their coding agents — Claude Code, Codex, OpenCode — and then you need this sort of internet, this secure place to deploy and absorb all this software. And when I say software, it's interesting, right? It might just be that your agent cooks up a report. It's like an HTML artifact and you need an internal place to deploy it to. It might be a micro-app that is sort of ephemeral. It might be personal software. It might be something you build just for yourself. Our design team, when you saw all these beautiful slides, it's amazing because they're building their own design tooling that they deploy internally to Vercel. So they themselves are cooking up on like the design factory, if you will. All of this beautiful pixels that we are giving to the world, they have internal apps that we might throw away.
Wow. It's a little bit like you see these high-production events and tomorrow the Palace of Fine Arts looks completely different.
It all goes back in the box, right? Or even like we had with our Mercedes partnership. It's amazing. I've been to a few of the races. You have this beautiful Formula 1 jacket and I wonder where I got it. Looking great. So, you go to these races. I went to the race in Vegas and like they just put it all up for a weekend, turn it all down.
