OpenAI tests 'Persistent mode' for Codex AI agent
This digest was compiled by AI from multiple sources — links to the originals are below.
OpenAI is developing a persistent version of its Codex AI agent that continues working until explicitly stopped. The feature, called Persistent mode, has been added to Codex's command line code base but has not been broadly rolled out. An OpenAI spokesperson confirmed testing but said there are no immediate launch plans.
Key Facts
- OpenAI has added code for a new 'Persistent mode' setting to its command line version of Codex.
- Persistent mode instructs Codex to 'continue working until put to sleep,' unlike current modes that stop after minutes or hours.
- An OpenAI spokesperson confirmed the company is testing the feature but said there are no immediate plans to launch it.
- The feature appears in Codex's 'reasoning effort' menu and is described as one of the most computationally intensive settings.
- Persistent mode includes a 'proactivity' feature that tells the agent to create follow-up tasks and message users sparingly.
Persistent Mode Development
OpenAI has started adding code for a new 'Persistent mode' setting to its command line version of Codex, according to changes reviewed by WIRED. The feature appears in Codex's 'reasoning effort' menu, where users select the level of computing power, tokens, and time for the AI model to think before answering. When Persistent mode is selected, Codex will 'continue working until put to sleep,' a stark contrast to currently available modes that stop after a few minutes or hours. OpenAI's head of core products, Thibault Sottiaux, described the company as a 'very bottom-up culture' where many features are explored on the open source repository.
Proactivity and Safety Limits
A file in the code base describes a feature within Persistent mode called 'proactivity,' which instructs agents to create follow-up tasks after finishing a user's request. The agent can work across sessions and use past user interactions and 'knowledge of the user' to decide what to work on. The agent has a tool to message the user without being asked but is told to send these messages sparingly. Instructions also set limits: Persistent mode does not expand what the agent is allowed to do, and altering anything outside the user's own system requires approval first.
2 sources
OpenAI tests 'Persistent mode' for Codex AI agent
