OpenAI is developing a new capability called persistent mode for its Codex AI agent, allowing it to work continuously and proactively rather than simply waiting for user instructions. The feature, currently in testing within Codex's command-line tools, represents a significant step toward OpenAI's goal of transforming AI assistants from reactive tools into autonomous agents that can anticipate user needs and complete tasks across multiple sessions.

What Is Persistent Mode and How Does It Work?

Persistent mode is a new setting within Codex's inference strength menu that fundamentally changes how the AI agent operates. Unlike standard modes that stop running after a few minutes or hours, persistent mode allows Codex to continue working until the user puts it into a dormant state. This means the agent can tackle complex, time-consuming tasks without interruption.

The feature includes a capability called "proactivity," which enables the agent to think beyond simply answering a user's immediate question. Instead, the agent can independently schedule follow-up work for itself, continue performing tasks across sessions, and even send messages to the user without being prompted. However, OpenAI has built in safeguards: the agent cannot expand its original operational permissions, and any modifications to content outside the user's own system require explicit user approval first.

"OpenAI's culture is very bottom-up. People will try out a lot of different things in the open-source codebase. It's a bit like our common testing ground," said Thibault Sotio, OpenAI's core product lead.

Thibault Sotio, Core Product Lead at OpenAI

Why Is OpenAI Building This Feature?

The development of persistent mode reflects OpenAI's broader ambition to create AI agents that operate more like autonomous assistants than passive tools. CEO Sam Altman has publicly stated his goal of transforming ChatGPT into a proactive, continuously operating AI agent. One key motivation is increasing usage of OpenAI's most advanced models; currently, only a small portion of ChatGPT's large user base actually uses these state-of-the-art models.

Persistent mode is also one of OpenAI's most resource-intensive settings, requiring significant computing power and tokens to operate. This reflects the computational complexity of maintaining an agent that can work independently over extended periods while maintaining context and making intelligent decisions about what to do next.

How to Understand Persistent Mode's Current Status and Rollout Timeline

• Testing Phase: Persistent mode is currently in early testing and has not been widely rolled out or officially released to the public yet.
• Code Availability: Changes to Codex's command-line tools are publicly available by default, allowing developers to see new features before they reach mainstream products like the Codex desktop application and ChatGPT Work.
• Future Expansion: The proactivity code resides in Codex's shared core rather than being isolated to the terminal tool, suggesting OpenAI plans to bring this capability to more products in the future.

An OpenAI spokesperson confirmed that the company is testing the feature but has no plans to launch it in the near future. This cautious approach reflects the complexity of deploying autonomous agents responsibly, particularly when those agents can take actions and communicate without explicit user prompts.

What Does This Mean for the Future of AI Assistants?

Persistent mode represents a meaningful shift in how AI assistants could function in the future. Rather than users initiating every interaction, agents could proactively identify tasks, complete them across multiple sessions, and even reach out to users when they've finished work or encountered a decision point. This moves AI from a tool you summon to an assistant that works in the background.

OpenAI has attempted to build proactive AI products before with mixed results. Pulse, launched last year, generated morning briefings while users slept but was discontinued in early summer 2026. The company's renewed focus on persistent mode suggests it has learned from these earlier attempts and is taking a more measured approach to autonomous AI capabilities.

The development of persistent mode also highlights the ongoing challenge of balancing autonomy with control. OpenAI has explicitly designed permission boundaries into the feature, ensuring that agents cannot exceed their original operational scope without user approval. This suggests the company is thinking carefully about how to deploy autonomous agents in ways that maintain user trust and safety.