OpenAI Tests 'Persistent Mode' for Proactive, Continuous AI Agents
OpenAI is testing a "persistent mode" in the command-line version of its Codex AI agent, allowing it to run continuously rather than stopping after a few minutes. This mode enables the agent to proactively schedule its own follow-up tasks and message users without waiting for prompts. This marks a significant shift from reactive AI models to autonomous, agentic workflows that can execute long-running tasks independently. It aligns with OpenAI's goal of turning ChatGPT into an active, continuous agent to drive adoption of its most advanced models. Found in Codex's "reasoning effort" menu, the mode is highly resource-intensive, letting users allocate compute power, tokens, and time. While the agent can work across sessions and self-assign tasks, it still requires user approval to make modifications outside the user's system.
## BACKGROUND
OpenAI Codex is an AI coding agent designed for software engineering tasks like writing code and fixing bugs, running locally via a command-line interface (CLI) or integrated into IDEs. Traditionally, large language models operate on a request-and-response basis, requiring a user prompt for every action and terminating tasks after a short period.