OpenAI is reportedly testing a new "Persistent mode" for its Codex AI agent. This feature would allow Codex to continue working on tasks until explicitly stopped by the user. Unlike current modes that time out, Persistent mode is designed to operate across multiple sessions. A key aspect of this mode is its "proactivity" feature, which enables the AI to generate its own follow-up tasks. Codex will utilize past interactions and user knowledge to determine what to work on next. The agent can also communicate with the user without a direct prompt, though sparingly. Despite its extended capabilities, Persistent mode includes limitations to prevent the AI from performing unauthorized actions. Users will need to grant explicit approval for any modifications outside their system. The feature's placement in the shared core of Codex suggests it's intended for broader applications beyond just the command line. This new mode represents a significant shift towards more autonomous and long-term AI task execution.