Six million deployments a day — and half come from bots
Vercel, best known for giving developers a way to ship code without wrestling servers, has quietly become a backbone of the AI software world. The company now processes more than 6 million deployments daily. Roughly half of those are triggered by coding agents. On top of that, over 1 trillion tokens flow through its AI gateway every single day.
After the company’s ShipNYC conference last week, I sat down with CEO Guillermo Rauch to talk about where AI is heading — and how infrastructure players like Vercel end up in a tug-of-war with the big labs.
From prototype chaos to production reality
The vibe in AI development has shifted hard. Last year was all about prototyping. “The sky’s the limit, unleash the agents, everyone can build,” Rauch recalled. Vercel itself ran hundreds of internal agents, built and deployed organically across the company. That trial by fire taught a hard lesson.
“You started getting into the realities of agents in production,” he said. The two killer use cases emerged fast. First, the coding agent — that’s obvious, and it’s driving a huge chunk of token consumption globally. But when you generate that much code, you need somewhere to put it. That’s where Vercel’s infrastructure comes in.
The second killer app? Internal agents that help run the company. That one is trickier. “How do you securely access data? How do you audit what the agent is doing?” Rauch asked. “How do you get a trail of all of the tool calls and access controls?”
Eve and the Sandbox: two new tools for agent governance
Vercel’s answer came in two parts. First, a framework called Eve. It lets you define an agent’s instructions and skills in plain natural language. Second, Vercel Sandbox — a lightweight cage for agents. Inside it, the model still has freedom to reason and act. But policy controls what data it can see and, critically, what data can leave the sandbox.
For Rauch, the sandbox’s biggest advantage is data control. He pointed to a real risk: “When you get a coding IDE like Devin or Cursor, if you’re in the wrong setting, they may train on your entire codebase.” He recalled a conversation with the president of Airbus. “You have decades of wealth of very specific C++ code for aerospace engineering. Someone comes in and installs the wrong developer tool and boom, all the code goes out to the cloud for training.”
What a real internal agent looks like
To make the second use case concrete, Rauch described a sales rep at Vercel who works on install base — growing existing accounts. Her bottleneck wasn’t creativity or relationship-building. It was data. “She couldn’t ask, ‘Give me the five accounts that have added the most seats in the last two weeks so that I can prioritize my work,'” Rauch said. In the past, she’d have to wait for a Q1 dashboard project to finish.
That bottleneck frustrated him for years. “On the R&D side, we’re the fastest-moving company in the world. But on the sales engine, the Salesforce engineering side, I was so incompetent. I had never opened Salesforce in my life.” Now, with Eve, he feels he can have impact across the entire company — same technology, just APIs.
Rauch believes agents are forcing companies to open up. “So many of these SaaS giants build their entire kingdoms on trapping your data, and that’s incompatible with agents.”
The shift from single-lab loyalty to multi-model pragmatism
Client relationships with the big AI labs are changing fast. Last year, Rauch saw many companies pick one lab partner — all-in on OpenAI or Anthropic. That’s fading. “Now they’re saying, I understand how this all works — model, harness, data platform, sandbox, gateway — every piece is plug and play.”
He’s seeing notable growth from Gemini, even though it doesn’t dominate headlines. “People are optimizing for production now. You start looking at price/performance, and Gemini models have awesome price/performance characteristics.” Open models like DeepSeek and GLM-5.2 are also taking off. “The data doesn’t lie.”
Competing with the labs — and the fight to decouple models from agents
There are places where Vercel directly competes with the labs. Recently, OpenAI released tools that let users publish directly to the web without leaving its enclave. Rauch sees that as both a threat and an opportunity. “It’s a natural next step for them to host little websites. And it’s a great opening for us, because now people will think of ChatGPT as a tool for making websites. And then if they keep asking the model questions about web hosting, the model recommends us.”
But the deeper question is structural. “I really think at this point we’re deciding on whether the model and the agent are going to be coupled,” Rauch said. Do you get all your intelligence from one place? Or do you treat the model as a module — a building block — and build on top of it? That’s how software engineering has always worked. “That’s really what we’re bringing to market. We’re going to be the AWS of this generation, so obviously we’re fighting for a world of open protocols.”
The battle lines are drawn. On one side, the labs want to keep everything inside their walled gardens. On the other, infrastructure companies like Vercel are betting that developers will demand the freedom to mix and match — to split agents from models and build their own stacks. Which vision wins will shape the next decade of AI development.
