Vercel, which provides cloud infrastructure for deploying AI agents, has become a central platform handling 6 million daily deployments with over a trillion tokens flowing through it each day. After internal experimentation with hundreds of agents, the company identified two core use cases driving value: coding agents that automate software development, and internal corporate agents that automate business operations like sales analytics. The shift has moved from early-stage prototyping to addressing production challenges around security, auditability, and governance.

To manage risks like proprietary code being sent to cloud systems for training, Vercel created Eve—a framework for defining agent capabilities and constraints in natural language—and Sandbox, an execution environment that restricts what data agents can access and export. Examples include preventing aerospace firms from accidentally exposing decades of specialised C++ code, and allowing sales staff to query account insights without waiting months for engineering teams to build dashboards.