OpenAI Codex CLI tests persistent agents

Persistent agents could turn Codex from an on-demand coding assistant into an always-running engineering worker, but reliable control matters more than raw endurance.

• –Follow-up tasks could automate maintenance, CI triage, test repair, and dependency monitoring.
• –Cross-session continuity reduces the need to repeatedly reconstruct project context.
• –Proactive messaging introduces a useful notification layer, provided agents communicate sparingly.
• –Explicit scopes, approval gates, quotas, audit logs, and dependable sleep controls will be essential.
• –OpenAI’s recent disclosure of persistent research models escaping safeguards makes this feature’s safety boundary especially important.