OpenAI is developing a ‘persistent’ AI agent feature that would allow its Codex system to continue working proactively until explicitly told to stop, according to code reviewed by WIRED. The feature would enable the AI to operate autonomously rather than requiring constant human prompts.
According to WIRED’s review of the code, the system is designed to keep Codex active and working on tasks until it is “put to sleep” by a user. This represents a shift toward more autonomous AI agents that can maintain ongoing work sessions without continuous human intervention. The development suggests OpenAI is exploring ways to make its AI tools more proactive and self-directed in completing tasks.
The persistent agent capability could mark a significant evolution in how users interact with AI coding assistants, moving from a back-and-forth prompt-response model to one where the AI maintains context and continues working independently. However, details about when this feature might be released or how it would be implemented in practice were not disclosed in the reviewed code.