Codex adds faster steering for running tasks

OpenAI announced faster steering for Codex in the ChatGPT desktop application in its October 8 release notes. The improvement is intended to make follow-up instructions influence ongoing coding work more quickly. Users may correct an approach, supply missing information or redirect a running task without waiting as long for the agent to react.

OpenAI documents a concrete setting under Settings → General → Follow-up behavior. Users can choose whether a follow-up should steer the current run or wait until the next run. The distinction matters for long development tasks, where a minor requirement change may affect subsequent steps or tool calls.

The release note explicitly covers the ChatGPT desktop app and describes a rollout. It does not prove simultaneous availability in the Codex API, Codex Cloud, terminal clients or every managed workspace. Availability needs to be checked against the selected client and account configuration.

Faster steering improves control of the user workflow, but it is not an independent code-quality benchmark or evidence for unrelated trust controls such as RBAC, audit export or sandbox isolation. User ability to send corrections is also different from a policy-mandated human approval before a tool action.

AgentenCode therefore documents the change in the correct Codex product context and avoids treating a client-specific interaction update as a platform-wide security guarantee.