OpenAI has opened up the infrastructure that powers Codex to any developer, through a new public API.
The Agents API is now in public beta, giving developers the same harness and infrastructure that powers Codex: context management, tool use, and coordination between multiple sub-agents working together. It also provides environments where an agent can work with real files, run code, and save intermediate results — and keep running reliably over multi-day tasks, not just a single request-response exchange.
Why this matters
Most "AI agent" products today are thin wrappers built by individual companies with their own reliability quirks. This puts a serious, production-grade agent harness directly into developers' hands, which is likely to accelerate how many genuinely autonomous (not just chatbot-style) tools show up in ordinary business software over the next year.
What This Means for Learners
If part of your job involves evaluating or commissioning AI tools, "built on the Agents API" is a phrase worth learning to recognise — it signals a tool built on real long-running agent infrastructure rather than a single prompt call, which changes what you should expect it to reliably do unsupervised.
