Codex can keep coding while it waits for an answer

The New Stack reported on 19 August 2026 that OpenAI merged a send_user_message_async tool into the public Codex repository. After Codex sends a question or update, the tool returns an accepted result immediately, and the agent can keep inspecting files or calling other tools in the same turn. Any reply arrives later as a new user message.

Codex already had a blocking request_user_input tool. A second merged pull request removed the experimental send_async_message flag the same day, so availability is supposed to follow whether the selected model lists the async tool. Subagents do not get the tool. The public code, as reported, does not name which model or Codex surface gets it first. OpenAI did not respond to The New Stack before publication.

The write-up also flags a timing gap: there is no checkpoint or automatic rollback if Codex keeps implementing one choice while the developer later answers with another.

Source: Codex can now keep coding while it waits for your answer, The New Stack, 19 August 2026.