OpenAI tests persistent Codex mode that can keep working across sessions
Code reviewed by WIRED shows a feature that lets the agent create follow-up tasks and continue until users put it to sleep.
OpenAI is testing a persistent version of its Codex AI agent that can continue working proactively until a user puts it to sleep, according to code reviewed by WIRED.
The company has added a Persistent mode setting to the command-line version of Codex. OpenAI has not broadly released or announced the feature, and a spokesperson said there are no immediate plans to launch it.
Persistent mode appears in Codex’s reasoning-effort menu and allows the system to use computing power, tokens and time for longer than existing modes, which stop after minutes or hours even when a task is unfinished.
A related proactivity feature instructs the agent to create follow-up tasks after completing a request and work on them across sessions. The code also describes a tool that can message users without being asked, although it tells the agent to do so sparingly.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The instructions say persistent mode does not expand what the agent is allowed to do and that changing anything outside a user’s system requires approval. OpenAI said the feature is being explored in its shared Codex codebase, which could allow it to reach products beyond the command line.
The test comes as OpenAI, Anthropic and Meta compete to build general-purpose agents that automate work and personal tasks. OpenAI has also acknowledged that persistent AI systems carry heightened risks.