Updated
Updated · TestingCatalog · Jul 23
OpenAI Builds Real-Time Voice for Codex, Expanding ChatGPT Into a Multi-Agent Assistant
Updated
Updated · TestingCatalog · Jul 23

OpenAI Builds Real-Time Voice for Codex, Expanding ChatGPT Into a Multi-Agent Assistant

3 articles · Updated · TestingCatalog · Jul 23

Summary

  • Recent ChatGPT bundle additions point to a real-time voice mode for Codex that lets users speak requests while background agents handle tasks and report results back.
  • System prompts describe a broader assistant than a coding tool, with actions including checking Slack, opening documents and calendars, controlling Spotify, browsing, shopping and scanning Uber Eats for dinner.
  • Several details remain unresolved: OpenAI has not confirmed which model would power the conversations, and a Codex real-time voice entry has not appeared in the ChatGPT desktop voice selector.
  • The work fits OpenAI’s recent consolidation of Codex into the ChatGPT desktop app alongside Chat and Work, after Codex Remote reached general availability on paid plans in June.
  • Hints from Codex staff suggest more news could come within 24 hours, with attention on parallel worker agents and faster Cerebras-served GPT-5.6 Sol inference rather than a new chip partnership.

Insights

Could OpenAI’s new voice-driven Codex ultimately replace your operating system by secretly orchestrating all your daily apps in the background?
How will OpenAI prevent cascading errors when autonomous background agents simultaneously manage your code, calendar, and credit cards during a call?
As AI models converge, will the true battleground shift from raw intelligence to who controls your persistent memory and cross-app workflows?