OpenAI introduces persistent AI agent for Codex

OpenAI is testing a new feature called 'persistent mode' for Codex, allowing the AI to work continuously without user input. The change was identified in recent code updates to the Codex command-line tool, which serves as a testing ground for new features before they are rolled out to other products like ChatGPT Work. Persistent mode lets users set the computational resources, tokens, and time allocated for tasks, enabling the AI to run until manually paused. OpenAI’s core product lead, Tibor Szirtes, emphasized the company’s culture of experimentation through open-source contributions. The feature also includes an 'agency' function, allowing Codex to autonomously plan follow-up actions based on past interactions. While the mode is still in testing and not yet available to the public, OpenAI plans to expand its use across products. CEO Sam Altman has previously expressed a vision for ChatGPT to evolve into an autonomous, continuously operating AI agent.