Vercel CEO Guillermo Rauch is in a battle to separate models from agents

Known for its cloud infrastructure that allows developers to deploy agents without managing servers, Vercel has quietly become one of the most central companies in AI software. The company currently sees 6 million submissions per day, part of which is due to coding agents, and more than 1 trillion tokens flow through the company’s AI gateway every day.

After the company’s ShipNYC conference last week, we sat down with Vercel CEO Guillermo Rauch about his take on this era in AI, and how platform companies like Vercel end up competing with the big labs. Here is a slightly edited transcript.

It feels like there is a different energy in the community this year, with fewer pilot programs and more focus on how to get things done right in practice. I’m sure you’ve seen that a lot with customers, but I’m curious what that journey looked like inside Vercel.

Last year it was all about prototyping. The sky is the limit, free agents, everyone can build, and so on. We did that, and we learned a lot because we had hundreds of agents that were naturally developed and used within the company, and then you started to get into the realities of agents in production, and some of the challenges.

The biggest lesson for me was the home use cases, two killer apps for agents. Another code agent, of course. That drives the use of many tokens in the world, but if you produce a lot of software, you need somewhere to put it. The second application to kill agents is the internal agent that helps you manage the company. The challenge is, how do you securely access the data? How do you evaluate what an agent is doing? How do you keep track of all the tool calls and access controls the agent had to make to get the job done?

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

What kind of problems does that help you avoid?

Because [the] sandbox, the biggest benefit is data control. The real danger of AI that I always think of is that, when you get a coding IDE like Devin or Cursor, if you’re in the wrong position, they can train themselves on your entire codebase. I remember talking to the president of Airbus about this. He has decades of wealth of highly detailed C++ code for aerospace engineering. Someone comes in and installs the wrong developer tool and boom, all the code goes out to the cloud to be trained.

I would like to hear more about that case of using the second killer. We all know about coding agents, but what does an in-house agent look like in practice?

So, there’s a salesman sitting over there [in Vercel’s office]. You are working on the install base. His job is to increase existing accounts. The bottleneck of people like him was not his intelligence, intelligence, ability to build relationships, it was data. “I don’t understand which accounts are growing the fastest. I gave five accounts that added more seats in the last two weeks, so I can prioritize my work.” He couldn’t ask that question in the past. He needed to wait until the Q1 project for the new sales dashboard was completed.

We had that problem for years at Vercel, and it was really frustrating because on the R&D side, we’re the fastest moving company in the world. But in the sales engine, Salesforce engineering [side]I was powerless. I had never opened Salesforce in my life when I started.

Now I feel like I can have an impact on the whole company, because Eve can be used for customer-facing agents and can be used to improve productivity. Same technology, just APIs. Agents are forcing companies to open up, and that will have a big long-term impact. Many of these SaaS giants are building their entire empires on capturing your data, and that doesn’t go well with agents.

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

Last year there were a lot of people choosing one lab partner – saying they were going to build everything on OpenAI or Anthropic. Now they say, I understand how all this works – the model, the cables, the data platform, the sandbox, the gateway – all the pieces are connected and playing. You can use OpenAI, you can use Anthropic, or you can use Gemini. We’re seeing a lot of growth for Gemini, even if it’s not in the news as much, because people are gearing up for production now. The truth is, when you’re preparing for production, you start looking at price/performance, and Gemini models have amazing price/performance characteristics. He also brings open models, so DeepSeek and GLM-5.2 take off. The data doesn’t lie.

There are places where you compete directly with labs, right? In just another week, OpenAI released a new set of tools that publish directly to the web without leaving the OpenAI enclave.

It is a natural next step for them to host small websites. And it’s a great opening for us, because now people will think of ChatGPT as a tool for making websites. And then when they keep asking the model questions about web hosting, the model recommends us. But you’re right, as models or platforms add capabilities, they come in direct competition with existing infrastructure platforms.

I really think that right now we are deciding whether the model and the agent will be combined.

Do you get all your creativity in one place? Or you get a module or a library or a building block from one provider, and build on top of it. That’s what software engineering always looks like, and that’s what we bring to the market. We will be the AWS of this generation, so obviously we are fighting for a world of open protocols.