Vercel CEO discusses splitting models from agents and the new AI landscape
In an interview after the ShipNYC conference, Vercel CEO Guillermo Rauch talks about two killer apps for agents, data security challenges, and the shift from single-lab partnerships to multi-model strategies.
Scale and Two Killer Use Cases
Vercel, known for its cloud infrastructure for deploying agents, now handles 6 million deployments daily, half triggered by coding agents, with over 1 trillion tokens flowing through its AI gateway every day. Following the company's ShipNYC conference, CEO Guillermo Rauch shared insights on the current AI moment.
Rauch identifies two killer apps for agents: coding agents and internal corporate agents. While coding agents dominate token usage, internal agents help companies run more efficiently—for example, a sales rep can query account growth without waiting for a dashboard project.
Data Security and New Tools
To address data security and audit challenges, Vercel developed Eve, a framework for defining agent instructions and skills in natural language, and Vercel Sandbox, which enforces policies on data access and outflow. Rauch warns about risks of coding tools training on entire codebases, recalling a conversation with Airbus's president about proprietary C++ code.
Competition with Labs and Open Standards
Rauch notes a shift from clients picking one lab (e.g., OpenAI or Anthropic) to using multiple models, including Gemini, DeepSeek, and GLM-5.2, driven by price/performance. While OpenAI recently released tools to publish directly to the web, Rauch sees it as an opportunity. Vercel aims to become the "AWS of this generation" by advocating for open protocols, separating models from agents.
