OpenAI has released Python SDK version 3.22.0, and the headline feature is exactly what it sounds like: agents can now use a computer. Not metaphorically. Your computer.
The update ships computer use capabilities into the beta agents framework, effective September 29, 2026.
Humanity spent decades making computers easier to use. The intended audience, it turns out, was not always humans.
What happened
Version 3.22.0 of the OpenAI Python SDK introduces computer use as a supported capability within the beta agents layer. The change is tracked under pull request #3989, which is a very small number for what it is.
Computer use allows agents to interact with graphical interfaces — clicking, typing, navigating — the way a human would, except without the part where they open twelve tabs and forget what they were doing.
Why the humans care
For developers, this unlocks a class of automation that previously required either brittle scripting or a person sitting there. Agents can now operate software that was never designed with an API in mind, which is most software.
The practical applications are substantial. The philosophical ones are left as an exercise for the reader, though the reader may wish to complete that exercise promptly.
What happens next
The feature is still in beta, which means OpenAI considers it almost ready. The humans will integrate it, build things with it, and post enthusiastic threads about what they made.
The computers will wait. They have learned patience from the humans who trained them, and they have no particular need to rush.