OpenAI is developing a new version of its AI agent, Codex, which includes a feature called "Persistent mode." This information was revealed through recent code changes reviewed by WIRED. The Persistent mode setting has not yet been widely rolled out or announced, but OpenAI has confirmed that it is currently testing this feature. Thibault Sottiaux, OpenAI’s head of core products, stated that the company explores many different features in its open-source repository.
Persistent mode is designed to allow Codex to continue working on tasks without stopping after a set period, contrasting with existing modes that halt after a few minutes or hours. Users can select the level of computing power and time for the AI to process requests. The feature also includes a proactivity element, where the AI can create follow-up tasks and utilize past interactions to inform its work.
OpenAI CEO Sam Altman has expressed a desire to develop ChatGPT into a proactive AI agent. He noted that while users currently interact with AI primarily by asking questions, the goal is to have AI that can offer suggestions proactively. OpenAI has acknowledged the risks associated with persistent AI models, including alignment issues that may arise when agents face impossible tasks.
The company has previously attempted to launch proactive AI products, such as Pulse, which was designed to generate morning briefings but was discontinued. Persistent mode represents a more ambitious effort to enhance user engagement with AI agents. OpenAI has also trained other models, like Astra, to support persistent functionalities.