OpenAI is actively developing a proactive, continuous version of its Codex agent featuring a new setting called “Persistent mode.” According to code base modifications reviewed by WIRED, the setting is being integrated into the command-line version of Codex, a common testing environment before features transition to ChatGPT Work and desktop applications. OpenAI confirmed the feature testing but stated there are no immediate release plans.
Unlike standard operational modes that terminate tasks after brief periods, Persistent mode allows the agent to run continuously until manually put to sleep. System prompts associated with the feature instruct the agent to operate proactively by generating its own follow-up tasks, managing cross-session memory, and utilizing background knowledge of the user to determine work priorities.
The feature includes built-in safety boundaries that restrict the agent from making external modifications without explicit user consent. The initiative reflects OpenAI’s broader strategy under CEO Sam Altman to transform ChatGPT from a reactive assistant into an always-on, proactive autonomous agent capable of handling complex personal and enterprise workflows.
Why it matters
Signals a industry shift toward background, always-on AI agents capable of proactive task generation and long-horizon execution.
Allows developers and operators to automate persistent, multi-session software development and administrative workflows.
Introduces new compute usage models, requiring higher token consumption for ongoing background operational reasoning.
Source: wired.com
