Would you trust an AI agent to do work on your behalf while you get some rest? OpenAI is apparently betting that you will, because the company is reportedly working on a persistent, always-on version of its coding agent Codex.
Wired first spotted the move to add the 24/7 bot operation in OpenAI’s GitHub for Codex, where code for the so-called “Persistent mode” first started showing up in the command-line version of the company’s coding agent. Per Wired, features that crop up in the command-line tool tend to make their way to other features over time, suggesting that this new mode could land in the desktop version of Codex in the near future.
Gizmodo reached out to OpenAI regarding the apparent discovery of Persistent mode but did not hear back by the time of publication.
If it goes live in the form that it currently exists, it appears that Codex’s always-on feature will work until a user actively puts it to sleep. It’ll present users with a “reasoning effort” option, where a person can select the level of computing power they’ll allow the AI model to use while contemplating and completing the task—likely as a way to avoid burning an unspeakable number of tokens overnight. Additional code seems to show a feature called “Proactivity,” which would make the AI agent create follow-up tasks for itself without requiring a person’s prompting. That would keep the agent working even after the initial prompt is completed, though how useful that feature is is heavily dependent on just how good the agent is at predicting what should come next.
Regardless, Persistent Mode and anything that comes with it would represent a significant jump for Codex to move to this endless operation model, considering users complain that it shuts off on its own at times—often before actually finishing its tasks.
It’d also be a necessary one, as OpenAI is starting to fall behind on this trend. Ever since OpenClaw got everyone worked up about the possibility of the always-on assistant, it’s become the feature the big players have been chasing (despite some obvious issues like, say, the agent deleting a person’s entire inbox overnight). Microsoft announced its own always-on agent earlier this year, and Meta is reportedly working on its own option called Hatch.
Seems we’re headed toward the age of the sleepless agent. Not sure if that will make anyone rest easy.