Central Players in AI Software: Vercel’s Quiet Rise

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

A Shift in the AI Community

After the company’s ShipNYC conference last week, we sat down with Vercel CEO Guillermo Rauch to discuss the current state of AI and how platform companies like Vercel compete with major labs. Here’s a lightly edited transcript.

Last year, Vercel focused on prototyping, unleashing agents, and allowing everyone to build. This led to hundreds of agents being developed and deployed within the company. However, as the company transitioned from prototyping to production, it encountered several challenges.

The biggest lesson learned was the importance of ‘home-run use cases,’ or the two killer apps of agents. One is the coding agent, which drives a significant portion of token utilization in the world. The second killer app is the internal agent that helps run the company. The challenge lies in securely accessing data, auditing what the agent is doing, and getting a trail of all the tool calls and access controls the agent incurred to get the job done.

To address these challenges, Vercel developed Eve, a framework that allows users to lay out an agent’s instructions and skills in natural language. Another tool is Vercel Sandbox, which puts the agent in a ‘cage’ and allows it to express its intelligence while applying policy on what data it can access and what data can leave the sandbox.

Internal Corporate Agents: A New Paradigm

For [the] sandbox, the biggest advantage is data control. A real risk of AI is when a coding IDE like Devin or Cursor trains on the wrong codebase. This can be catastrophic, as seen in the case of decades of wealth of very specific C++ code for aerospace engineering being compromised.

The second killer use case is the internal corporate agent. A sales rep, for instance, can use an internal agent to grow existing accounts by getting the five accounts that have added the most seats in the last two weeks. This would normally require a Q1 project for a new sales dashboard, but with the internal agent, it’s possible to get the data in real-time.

Vercel’s CEO, Guillermo Rauch, emphasizes that agents are forcing companies to open up, which will have dramatic long-term implications. Many SaaS giants build their entire kingdoms on trapping users’ data, and this is incompatible with agents.

Client Relationships with Big AI Labs: A Shift in Paradigm

Last year, people were choosing one lab partner and building everything on OpenAI or Anthropic. However, now they’re saying they understand how it all works – model, harness, data platform, sandbox, gateway – and are choosing between different options. Vercel is seeing a lot of growth in Gemini, even though it’s not as well-known, because people are optimizing for production now. They’re looking at price/performance characteristics, and Gemini models have awesome price/performance characteristics.

There are places where Vercel is in direct competition with the labs, such as when OpenAI released a new set of tools that publish directly to the web without having to leave the OpenAI enclave. However, this is also a great opening for Vercel, as people will think of ChatGPT as a tool for making websites, and then recommend Vercel if they ask the model questions about web hosting.

Vercel is fighting for a world of open protocols, where you get a module or a library or a building block from one provider and then build on top of it. This is more like software engineering has always been, and that’s what Vercel is bringing to market. They aim to be the AWS of this generation.

Decoupling Models and Agents

Vercel’s CEO, Guillermo Rauch, believes that the choice is between whether the model and the agent are going to be coupled or not. If they’re coupled, you get all your intelligence from one place. However, if they’re decoupled, you get a module or a library or a building block from one provider and then build on top of it. This is a more traditional software engineering approach.

Vercel is fighting for a world of open protocols, where you can use OpenAI, Anthropic, or Gemini, and bring in open models like DeepSeek and GLM-5.2. The data doesn’t lie, and people are starting to see the benefits of this approach.