← Back to live feed1 story
Internal code reviewed by WIRED reveals a new reasoning effort for the company's Codex agent that allows it to autonomously manage tasks without constant user prompts. This "Persistent mode" lets the agent proactively create follow-up tasks, operate across sessions, and message users first based on previous interactions, continuing its work until a user puts the system to sleep.
OpenAI confirmed the feature is in testing but has no immediate plans for a public launch. The capability will be powered by the forthcoming Astra model as the company develops a new class of always on products while addressing safety risks and model misalignment seen in a previous Hugging Face incident.
Sign in to suggest edits
Key sources
- SOURCE@zeffmax“proactively create follow up tasks for itself”x.com
- SUPPORT@zeffmax“no immediate plans to launch it”x.com
- SUPPORT@zeffmax“forthcoming Astra model will enable persistent agents”x.com
- SUPPORT@kimmonismus“continue them across sessions, use past interactions and “knowledge of the user””x.com
- SUPPORT@techmeme“designed to let AI agents 'continue working until put to sleep'”x.com