Vercel CEO Guillermo Rauch talks about the struggle to separate models from agents

Known for its cloud infrastructure that allows developers to deploy agents without managing servers, So he sends It has quietly become one of the most central companies in artificial intelligence software. The company currently sees 6 million deployments daily, half of which are powered by crypto proxies, and has over a trillion tokens flowing through it. The company’s artificial intelligence portal daily.

Following the company’s ShipNYC conference last week, we sat down with Vercel CEO Guillermo Rauch to learn about this moment in AI, and how platform companies like Vercel end up competing with the big labs. Here’s a slightly modified version.

There seems to be a different energy in the community this year, fewer pilots and more focus on how to make things work well in practice. I’m sure you’ve seen this a lot with clients, but I’m curious what that journey looked like within Vercel.

Last year was all about prototyping. The sky’s the limit, unleash agents, everyone can build, etc. We did that, and we learned a lot because we had hundreds of agents developed and deployed organically within the company, and then I started to learn about the reality of agents in production, and some of the challenges.

The biggest lesson for me was the home-run use cases, the two killer apps for agents. One is the coding agent, of course. This increases the use of tokens in the world, but when you produce a lot of software, you need a place to put it. The second killer app for agents is an in-house agent that helps you manage the company. The challenge here is how do you access the data securely? How can you audit what an agent does? How do you get a track of all the tool calls and access controls the agent had to endure in order to get the job done?

To solve this problem, we came up with a framework called Eve, where you can put customer instructions and skills in natural language. The other tool is the Vercel Sandbox, which places the client in a small cage. They can be free to express their intelligence, but then you can apply policy to what data they can access and what data can leave the sandbox.

What kind of problems does it help you avoid?

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

I’m curious to know more about the second killer use case. We all know about programming agents, but what does an in-house agent look like in practice?

So, there’s a salesperson sitting over there (in Versell’s office). It works on a mount. Its job is to grow existing accounts. The bottleneck for people like her wasn’t her creativity, intelligence, and ability to build relationships, it was data. “I don’t understand which accounts are growing the fastest. Give me the five accounts that have added the most seats in the last two weeks, so I can prioritize my work.” She couldn’t ask this question in the past. She needed to wait until the first quarter project was completed to get a new sales dashboard.

We’ve been in a bottleneck for years at Vercel, and it’s been really frustrating because on the R&D side, we’re the fastest moving company in the world. But in terms of the sales engine, Salesforce architecture (aspect), I was very incompetent. I had never opened Salesforce in my life when I started.

I now feel like I can really impact the entire company, because Eve can be used for our customer-facing agents and can be used to improve productivity. Same technology, just APIs. Agents are forcing businesses to open up, and this will have big implications in the long run. A lot of SaaS giants build their entire kingdoms on trapping your data, and that doesn’t sit well with proxies.

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

Last year, there were a lot of people who chose one lab partner, saying they would build everything on OpenAI or Anthropic. Now they say, I understand how it all works – the model, the tools, the data platform, the sandbox, the gateway – every piece is plugged in and running. You can use OpenAI, you can use Anthropic, or you can use Gemini. We’re seeing a lot of growth in Gemini, even though it’s not in the news as much, because people are improving production now. The truth is that when you optimize production, you start looking at price/performance, and the Gemini models have great price/performance characteristics. You can also bring open models, so Deepseek and GLM-5.2 will take off. Data doesn’t lie.

There are places where you’re in direct competition with labs as well, right? Just last week, OpenAI released a new set of tools that publish directly to the web without having to leave the OpenAI area.

It’s a natural next step for them to host small websites. This is a great opportunity for us, because now people will think of ChatGPT as a website building tool. Then, if they continue to ask the form questions about web hosting, the form will recommend us. But you’re right, as models or platforms add more capabilities, they come into direct competition with existing infrastructure platforms.

I really think that at this point we are deciding whether the model and agent will be paired.

Do you get all your intelligence from one place? Or do you get a module, library, or building block from one provider, and then build on top of it. This is very much like software engineering has always been, and this is really what we bring to the market. We will be the AWS of this generation, so we are clearly fighting for a world of open protocols.