On September 10, Vercel revealed a new integration that enables developers to construct and launch applications using the OpenAI Agents API. In this setup, OpenAI retains control of the agent's reasoning loop and session data, while Vercel routes each session to its Sandbox service, which provides a separate execution context and stores workspace files.

The design replaces the traditional model of continuously running workers, containers, and GPUs with a request-driven flow that spins up a sandbox only when needed and scales back to zero afterward. This approach is aimed at workloads that experience spikes or operate asynchronously, offering freelancers and small teams a cost-effective way to incorporate autonomous agents into full-stack projects.