Vercel CEO on Agents, Sandboxes, and the AI Infrastructure Battle
Original: Vercel CEO Guillermo Rauch on the fight to split off models from agents
Why This Matters
Vercel's scale (1T+ daily tokens) signals its growing role as critical AI deployment infrastructure for enterprises.
Vercel CEO Guillermo Rauch discussed the company's AI infrastructure role at ShipNYC, noting 6 million daily deployments — half triggered by coding agents — and over 1 trillion tokens flowing through Vercel's AI gateway daily.
In a TechCrunch interview following Vercel's ShipNYC conference, CEO Guillermo Rauch outlined how the company has evolved from a frontend deployment platform into a central node in AI infrastructure. Vercel now processes 6 million deployments per day, with roughly half initiated by coding agents, and more than 1 trillion tokens passing through its AI gateway daily.
Rauch identified two 'killer apps' for agents: coding agents, which are driving the majority of global token usage, and internal corporate agents that help companies operate. He noted that the latter raises significant challenges around secure data access, auditability, and access control logs.
To address these, Vercel developed two tools: Eve, a framework for defining an agent's instructions and skills in natural language, and Vercel Sandbox, which isolates agents in a controlled environment where data access and egress can be governed by policy.
Rauch highlighted data leakage as a core risk, citing a conversation with the president of Airbus: decades of proprietary aerospace C++ code could be inadvertently uploaded for model training if a developer installs the wrong IDE. The Sandbox is designed to prevent such scenarios.
He also described a real-world internal agent use case — a sales representative whose bottleneck is not skill but access to fragmented company data — as emblematic of where enterprise AI is heading.