Voice agents aren't listening — they're reading: the case for audio-native decision-making
sidahuj · x · 2026-10-20
namanjn1998 argues the biggest gap in voice AI is that agents make decisions from transcripts, not from the audio itself. Quoting Mira Murati — while thinking, agents are "deaf and blind," unable to perceive anything else happening in the scene — he lays out how transcription discards tone, pauses and ambient signals, in a long-form piece inviting disagreement from voice agent builders.
Related event: Voice Agents Read Transcripts but Never Actually Listen(2 posts)→
More from coding & agent
- Dev Built a Movie Picker App with Codex for the Chronically Indecisive — OpenAIDevs · 2026-10-20
- Dev Ships Mechapet, an iOS Adventure Game Built with Codex — OpenAIDevs · 2026-10-20
- Dev Uses Codex to Generate 3D Visualizations and Cable Routing Docs for His New House — OpenAIDevs · 2026-10-20
- Dev Says GPT-6 Astra Designed an Entire Circuit Board From Scratch for His Companion Robot — OpenAIDevs · 2026-10-20
- OpenAI Showcases Community Build: Live Model Running Sonos, Home Assistant and BusyBar — OpenAIDevs · 2026-10-20
- OpenAI Devs highlights what the community is building with Codex — OpenAIDevs · 2026-10-20