OpenAI’s Responses API now runs agents in a computer environment

OpenAI is shifting from single-purpose models to full-blown agents that can handle complex workflows. The key upgrade? The Responses API now comes with a computer environment—a container workspace with a shell tool, filesystem, structured storage like SQLite, and restricted network access. This means agents can do more than just answer questions: they can run services, fetch API data, generate spreadsheets or reports, and handle multi-step tasks without you building a custom execution system. Instead of forcing developers to manage intermediate files, paste huge tables, or deal with timeouts and retries on their own, OpenAI handles the infrastructure. The model proposes actions like reading files or sending curl requests, and the platform executes them in an isolated loop. This turns agents from clever talkers into actual doers. Why it matters: By embedding agents in a real execution environment, OpenAI is making it practical to automate production workflows—like deploying code, querying databases, or generating reports—without the usual security and scaling headaches.