Vercel CEO Guillermo Rauch on the fight to split off models from agents

Known for its cloud infrastructure that allows developers to deploy agents without managing servers, Vercel quietly become one of the most central companies in the field of AI software. The company currently sees 6 million deployments per day, half of which are triggered by coding agents, and more than 1 trillion tokens are flowing through the company’s AI gateway daily.

After the company’s ShipNYC conference last week, we spoke with Vercel CEO Guillermo Rauch about his view of the current moment in AI, and how platform companies like Vercel are ultimately competing with large labs. Here is a lightly edited transcript.

It feels like there’s a different energy in the community this year, less pilot programs and more focus on how to make things work in practice. You’ve probably seen that a lot with customers, but I’m curious what that process looked like within Vercel.

Last year it was about prototypes. The sky is the limit, release the agents, anyone can build, and so on. We did that, and we learned a lot because we had developed and deployed hundreds of agents organically within the company, and then you started to get to know the reality of agents in production, and some of the challenges.

The biggest lesson for me was the home use cases, the two great agent apps. One of them is of course the encryption agent. That accounts for a lot of the world’s token usage, but when you produce that much software, you need a place to put it. The second agent killer app is the in-house agent that helps you run the business. The challenge is: how do you access data securely? How do you control what the agent does? How do you get a trail of all the tool calls and access checks the agent had to perform to get a job done?

To solve that, we came up with this framework, called Eve, where you can represent agents’ instructions and skills in natural language. And another tool is Vercel Sandbox, where you put the agent in a small cage. It can still have the freedom to express its intelligence, but then you can apply policies to what data it can access and what data can leave the sandbox.

What types of problems does this help you prevent?

For Sandbox, the biggest advantage is data control. A real risk of AI that I always think about is that if you get a coding IDE like Devin or Cursor, if you’re in the wrong setting, they might train on your entire code base. I remember talking to the president of Airbus about this. You have decades of highly specific C++ code for aerospace engineering. Someone comes in and installs the wrong developer tool and boom, all the code goes to the cloud for training.

I’m curious to hear more about that second killer use case. We all know coding agents, but what does an internal corporate agent look like in practice?

So there’s a salesperson there [in Vercel’s office]. She works at installbase. Her job is to grow existing accounts. The sticking point for people like her isn’t her creativity, intelligence, and ability to build relationships; it’s data. “I don’t understand which accounts are growing faster. Give me the five accounts that added the most seats in the last two weeks so I can prioritize my work.” Before, she couldn’t ask that question. She had to wait until a Q1 project for a new sales dashboard was completed.

At Vercel we were in that bottleneck for years, and it was very frustrating because we are the fastest moving company in the world in terms of R&D. But on the sales engine, the Salesforce engineering [side]I was so incompetent. When I started, I had never opened Salesforce.

Now I feel like I can actually have an impact on the entire company because Eve can be leveraged for our customer-facing agents and used to improve productivity. Same technology, they’re just APIs. Agents are forcing companies to open up, and that will have dramatic long-term consequences. So many of these SaaS giants are building their entire kingdom on intercepting your data, and that’s incompatible with agents.

How do you see customer relationships with the large AI labs changing?

Last year there were a lot of people who chose one lab partner and said they would build everything on OpenAI or Anthropic. Now they say: I understand how this all works – model, harness, data platform, sandbox, gateway – every part is plug and play. You can use OpenAI, you can use Anthropic, or you can use Gemini. We’re seeing a lot of growth from Gemini, even if it’s not on the news as much, because people are now optimizing for production. The reality is that when you optimize for production, you start looking at price/performance, and Gemini models have great price/performance characteristics. You also bring in open models, so Deepseek and GLM-5.2 start. The data doesn’t lie.

There are also places where you’re competing directly with the labs, right? Recently, OpenAI released a new set of tools that publish directly to the web without having to leave the OpenAI enclave.

It is a logical next step for them to host small websites. And it’s a great opening for us because people will now see ChatGPT as a website building tool. And if they keep asking the model questions about web hosting, the model recommends us. But you’re right: as the models or platforms add more capabilities, they compete directly with the infrastructure platforms that already exist.

I really think at this point we decide whether the model and the agent will be linked.

Do you get all your intelligence in one place? Or you get a module or a library or a building block from one provider, and then you build on top of that. That’s more like what software engineering has always been, and that’s really what we’re bringing to the market. We’re becoming this generation’s AWS, so we’re clearly fighting for a world of open protocols.