OpenAI Developers outlined several workflows for Voice in Codex on August 5, presenting the feature as a way to discuss ideas, launch Codex sessions, and monitor ongoing work without staying at the keyboard.
The examples move beyond dictation. OpenAI described a workflow in which a brainstorming conversation becomes Linear tickets that Codex assigns through a heartbeat-based delegation process. Other examples involve discussing feature ideas, test plans, validation, implementation, and pull requests by voice.
The posts also present Voice in Codex as a coordination layer for multiple sessions. A quoted user, James Eversole, characterized one chat as a place to develop an idea before handing the finished direction to a new session, while the voice conversation checks on other threads. Another user, Gerry, claimed that speaking while playing a game caused Codex to create separate threads with worktrees and later report which features were ready.
OpenAI’s examples extend beyond software development. One post describes organizing thoughts while traveling to class and beginning a new experiment. Another outlines a voice-directed content workflow involving keyword research, competitor analysis, content outlines, and distribution. A separate quoted post describes a similar process for SEO research, with voice commands used to approve or reject steps, redirect work, or stop a task.
Some users welcomed the session handoff and the prospect of managing work as a planner rather than writing code directly. One quoted user claimed to have drafted and revised a 14-session database experiment while walking to class, although the post provides no independent verification of that result.
The replies also highlight unresolved usability questions. Users requested Voice support with API keys, Linux availability, push-to-talk, per-session muting, stronger mobile support, access in every chat, and clearer controls for selecting the model and effort level.
Source: OpenAI Developers on X
