SAN FRANCISCO, Sept. 2 — OpenAI is quietly testing a new “Persistent mode” for its coding agent Codex that would let the AI keep working on its own until told to stop, according to code changes reviewed by WIRED and published on August 27. The feature appears in the command line version of Codex, which is publicly accessible by default. An OpenAI spokesperson confirmed the capability is being tested but said there are no immediate plans to launch it.
For families and everyday users, the shift is subtle but meaningful. Today’s AI agents stop after a few minutes or hours, even when a task is unfinished.
Persistent mode would change that — the code indicates Codex would “continue working until put to sleep.” That kind of always-on helper is exactly what OpenAI CEO Sam Altman has been describing lately. In podcasts, interviews, and private investor meetings, he has talked about turning ChatGPT into a proactive, always-on AI agent.
An agent that plans its own next steps
Persistent mode shows up inside Codex’s “reasoning effort” menu, where users pick how much computing power, tokens, and time the model gets to “think” before answering. But the deeper change is in a separate file describing a feature called “proactivity.” In that mode, the agent is instructed to create follow-up tasks for itself after finishing a user request. It can work across sessions, drawing on past interactions and “knowledge of the user” to decide what to do next.
It even has a tool to message the user unprompted — though it is told to send messages sparingly. There are guardrails.
Persistent mode does not expand what the agent is allowed to do, and any change outside the user’s own system requires approval first. That appears designed to limit potential dangers.
A shared playground
Thibault Sottiaux, OpenAI’s head of core products, told WIRED that “OpenAI is a very bottom-up culture and many different things are explored on the open source repo which is a bit of our shared playground.” The proactivity file sits in the shared core of Codex, which suggests the feature could extend beyond the command line tool to other products, including the Codex desktop app and ChatGPT Work.
OpenAI, along with rivals Anthropic and Meta, is racing to deliver general-purpose AI agents that can automate chores like filing expense reports or scheduling doctor’s appointments. Right now, these agents are used mostly by software engineers.
Silicon Valley sees a much broader customer base ahead. The feature has not been broadly rolled out or announced. Whether it ever ships — and how carefully OpenAI paces it — is the question to watch.
