OpenAI Tests Persistent Mode for Codex, AI Agent Works Autonomously

According to Wired, OpenAI is developing a "persistent mode" for its coding agent Codex, aiming to let the AI work autonomously without waiting for user commands. The feature is in testing and not widely available; an OpenAI spokesperson confirmed testing but no immediate release.

The mode appears in Codex's "reasoning effort" menu, where users can set compute investment. Once enabled, Codex runs until put to sleep, unlike other modes that stop after minutes or hours. OpenAI also designed an "initiative" capability, where the agent plans follow-up tasks after answering, works across sessions, and may message users proactively, though sparingly.

Permissions remain unchanged; modifying content outside the user's system requires approval. The code sits in Codex's shared core, suggesting broader deployment. CEO Sam Altman wants ChatGPT to become a proactive AI agent, boosting usage of advanced models. Previous attempts like Pulse were discontinued.