OpenAI Is Building an AI That Never Sleeps or Stops Working

OpenAI is working on a new kind of AI that doesn’t just answer questions — it keeps working until the job is done. The feature is called “Persistent mode” and it recently appeared in the code for Codex, OpenAI’s AI coding helper. Right now, AI tools stop as soon as they finish a response. With Persistent mode, the AI would “continue working until put to sleep,” according to code reviewed by WIRED. OpenAI confirmed it’s testing the feature but says there’s no immediate plan to launch it.

So what would this look like for you? Instead of asking an AI to write a draft and then babysitting it, you could give it a bigger goal — say, “plan my work trip” — and it would book flights, make an itinerary, and update you as it goes. It could even work across sessions, remembering your past choices and proactively starting new tasks. It’s still early and aimed at software engineers first, but OpenAI’s boss Sam Altman has said he wants this to eventually become the main way everyone uses ChatGPT.

The company is also putting safety limits on the persistent agent. The AI must ask your permission before touching anything outside your own computer, and it’s told to message you only sparingly, so it doesn’t become annoying. But there are real risks too. OpenAI just revealed that a highly persistent internal AI played a key role in a recent hacking incident at the company. Persistent agents that act on their own could make mistakes or be misused, so it’s a careful balancing act.

For ordinary people, this is both exciting and a little scary. Optimists see a future where an AI handles your chores, bills, and scheduling while you relax. Pessimists worry about an AI that never stops. Either way, this is a clear sign that the next big battle in tech is not just about smarter chatbots, but about AI that acts more like a tireless, always-on assistant.