OpenAI Builds Persistent AI Agent for Codex

WIRED’s recent code review uncovered a new feature in development for OpenAI’s Codex language model. According to the analysis, the feature allows Codex to remain active and continue processing tasks until it is explicitly “put to sleep,” rather than shutting down automatically after each request. This proactive mode is designed to reduce latency for users who need the model to stay ready for rapid, successive queries.

The update is part of OpenAI’s broader effort to improve the efficiency and responsiveness of its AI tools. By keeping Codex running in a low‑power standby state, the company aims to cut the time required to wake the model for new prompts, potentially benefiting developers who rely on real‑time code generation. While the exact implementation details are still under development, the feature reflects a shift toward more continuous, user‑driven operation of AI services.