OpenAI Tests Codex 'Persistent Mode' to Enable AI Agents to Proactively Execute Tasks Continuously

OpenAI is testing a new feature for its Codex AI agent called "Persistent Mode." Currently in the testing phase, this mode is not yet widely available to the public. Unlike the traditional mode—which automatically halts after several minutes or hours—Persistent Mode allows Codex to run continuously without being suspended. This mode endows Codex with "proactive" capabilities: instead of terminating tasks after answering a user's query, it can autonomously plan subsequent steps and continue executing tasks across sessions, leveraging past interactions and its understanding of the user to determine next actions. The system also permits the agent to send messages without an explicit user request, though such use is intended to be minimal. OpenAI has established clear permission boundaries for this mode; the agent will not gain operational permissions beyond its original scope, and any action requiring modifications outside the user’s system must receive explicit user approval beforehand. OpenAI CEO Sam Altman has previously expressed his desire to transform ChatGPT into an AI agent capable of proactive behavior and continuous operation. Earlier, OpenAI’s proactive AI product, Pulse, ceased operations in early summer 2024.