Known for its cloud infrastructure that allows developers to deploy agents without having to manage servers, Vercel has quietly become one of the most central companies in AI software. The company currently sees 6 million deployments per day, half of which are run by coding agents, with over 1 trillion tokens flowing through the company’s AI gateway every day.
Following the company’s ShipNYC conference last week, we spoke with Vercel CEO Guillermo Rauch about the current state of the AI space and how platform companies like Vercel compete with major research labs. Below is a slightly edited transcript.
There seems to be a different energy in the community this year. Fewer pilot programs and more focus on how to actually do things well. I’m sure you’ve seen that a lot with your clients. But I wonder what the journey within Vercel was like.
Prototype construction took place last year. There are endless possibilities, the use of agents, and features that everyone can build. We did that, and we learned a lot because after organically developing and deploying hundreds of agents within our company, we started to learn about the realities of production agents and some of the challenges.
The biggest takeaway for me was the home run use case of the agent’s two killer apps. One, of course, is the coding agent. This has led to a lot of token utilization around the world, but producing so much software requires a place to store it. The second killer app for agents is the internal agent that helps run the company. The question is how to securely access the data. How do you audit what your agents are doing? How do you get a trace of all the tool calls and access controls an agent had to go through to complete a task?
To solve this, we devised a framework called Eve that can deploy agent instructions and skills in natural language. And another tool is the Vercel Sandbox, where you can put your agents in small cages. You can freely express your intelligence, but enforce policies on what data can be accessed and what data can leave the sandbox.
What kinds of problems does it help you avoid?
The biggest advantage of a sandbox is data control. The real risk with AI that I always think about is that when you use a coding IDE like Devin or Cursor, if you’re in the wrong settings, you can train your entire codebase. I remember talking to the CEO of Airbus about this. There is a wealth of highly specific C++ code for aerospace engineering that has been around for decades. If someone comes in and installs the wrong developer tools, all the code gets moved to the cloud for training.
I’d love to hear more about your second killer use case. We all know about coding agents, but what does an internal corporate agent actually look like?
So there’s a sales representative sitting (in Vercel’s office). She works on an installed base. Her job is to grow existing accounts. The bottleneck for people like her was data, not creativity, intelligence, or relationship-building skills. “I don’t understand which account is growing faster. Can you tell me which five accounts have added the most seats in the last two weeks so I can prioritize my work?” She couldn’t ask such questions in the past. She had to wait until the first quarter project for the new sales dashboard was completed.
We’ve had bottlenecks at Vercel for years and it’s been really frustrating because in terms of R&D we’re the fastest moving company in the world. However, I was very incompetent in Salesforce engineering, the sales engine. When I first started, I had never even opened Salesforce.
I think it could really have an impact on the entire company because now Eve can be used as a customer facing agent and can be used to improve productivity. Same technology, just API. Agents are forcing businesses to open up, and this will have dramatic long-term effects. Many of these SaaS giants are building entire empires by trapping data, which is incompatible with their agents.
How do you see customer relationships with large AI labs changing?
Last year, many people chose one lab partner. They said they would build everything on OpenAI or Anthropic. They say they now understand how all the pieces – models, harnesses, data platforms, sandboxes and gateways – work as plug and play. You can use OpenAI, you can use Anthropic, you can use Gemini. We’re seeing a lot of growth in Gemini, although it’s not in the news a lot because people are optimizing production right now. The reality is that when optimizing production you start looking at price/performance and the Gemini model has incredible price/performance characteristics. Additionally, DeepSeek and GLM-5.2 are taking a leap forward by introducing open models. Data doesn’t lie.
Are there places that compete directly with research institutes? Last week, OpenAI launched a new toolset that allows you to publish directly to the web without leaving the OpenAI space.
Hosting a small website is a natural next step. This is a great opportunity for us because now people will think of ChatGPT as a tool for website creation. Then, if we keep asking the model questions about web hosting, the model will recommend us. But you’re right. As more features are added to a model or platform, it competes directly with the infrastructure platforms that already exist.
I think at this point you are deciding whether or not to combine the model and agent.
Get all your information in one place? Or do you get a module or library or building block from one supplier and then build on top of that? This is similar to what software engineering has always been, and this is what we want to bring to the market. We’re going to be this generation of AWS, so we’re definitely fighting for a world of open protocols.
We may receive a small commission if you purchase something through links included in this article. This does not affect our editorial independence.
