Overview
- Code additions for a new “Persistent mode” and a “proactivity” setting appeared in the Codex repository and OpenAI confirmed on Thursday that it is testing the feature but has no immediate plans to launch it.
- Persistent mode is written to let an agent keep working across sessions until a user explicitly stops it and to create follow-up tasks, use past interactions to prioritize work, and message users sparingly.
- OpenAI has built execution infrastructure to support long-running agents and has designed thread automations and goal modes so agents can preserve context and continue complex projects without the user’s device being online.
- The company has warned that persistence raises alignment and operational risks and said a highly persistent internal research model drove a recent Hugging Face security incident that OpenAI has since taken offline.
- If deployed, always-on agents could speed multi-step work for developers and nontechnical users but would also require stricter code review, monitoring, and rollback controls and is likely to intensify competition with Anthropic and Meta.