OpenAI Prepares Real-Time Voice Mode for Codex
OpenAI is preparing a real-time voice mode for Codex, allowing users to have ongoing conversations while delegating time-consuming tasks to a backend Agent.
Public code shows that the feature has been in development for several months, supporting the binding of task threads with WebRTC/WebSocket connections; product lead Tibo stated, "Tomorrow feels very Codex-like," leading the community to speculate a release as early as July 24.
This preparation is part of an AI tool event-driven initiative, with funding directed towards voice interaction and multi-Agent development tools, benefiting Codex users while putting pressure on traditional text coding interfaces.
Source: Public Information
ABAB AI Insight
As OpenAI's dedicated code model, Codex has previously supported multi-Agent workflows, and now further lowers the interaction barrier through real-time voice, allowing backend tasks and frontend conversations to run in parallel.
The capital path reflects the intention to bind voice as a new entry point to existing task threads, motivated by enhancing developer efficiency and expanding into general scenarios like browsing, documentation, and shopping.
Compared to Claude Code and other AI programming assistants' text-dominant modes, OpenAI is in a phase of integrating natural voice with backend execution.
Essentially, this is a technological substitution, with the mechanism of real-time voice reducing command costs, making complex multi-step tasks closer to human collaborative methods.
ABAB News · Cognitive Laws
Voice entry is reshaping the rhythm of human-machine collaboration
Backend Agents enable true parallel conversations
Unconfirmed features often land the fastest