OpenAI is testing an execution profile termed "Persistent Mode" within its Codex agent codebase, designed to enable continuous, self-directed task execution without standard step-count timeouts or per-turn pauses.
Code commits surfaced in the public repository of the Codex command-line interface indicate that the agent can proactively generate follow-up tasks, maintain state across development sessions, and continue working autonomously until explicitly halted by the user.
Architecture and Reasoning Budgets
In current implementations, coding agents typically operate in bounded request-response loops or terminate execution after a predefined number of tool calls or minutes of inactivity. Persistent Mode is integrated into the Codex "reasoning effort" configuration tier, which allocates extended compute, context memory, and token allowances for multi-phase engineering tasks.
According to codebase documentation, an agent operating under Persistent Mode will continue executing workflows asynchronously, polling for state changes and queueing subsequent sub-tasks without requiring repeated user prompts.
Permission Sandboxing and Safety Boundaries
To mitigate the failure modes associated with unsupervised agent operation, OpenAI has embedded explicit permission boundaries within Codex's shared core framework. Persistent Mode does not grant escalated tool permissions; any file modification, external API call, or network egress outside the local project environment remains subject to mandatory user confirmation.
Thibault Sottiaux, OpenAI's head of core products, confirmed that the company is actively testing persistent capabilities within its open-source repository while noting that no public production release date has been set.
The architectural shift targets long-standing limitations in autonomous coding tools, where complex refactors and debugging pipelines are frequently cut short by session time limits, while attempting to manage agent drift and security risks in long-horizon reasoning.
