--- schema: 1 type: video title: 'Voice Agents Can Just Do Things — Charlie Guo, OpenAI' description: >- The Moviefone hotline was a speech to speech system. The GPS unit in your parents' car was an event to speech system. Charlie Guo works on developer experience at OpenAI, and he opens by pointing out that the three modes of voice interaction are all decades old, which makes the question not what is newly possible but what we stopped noticing. His actual argument is aimed at an assumption almost every builder makes unexamined: that a voice agent has to answer in speech. It does not. Software spent decades learning to reply by highlighting a field, changing a button, raising a notification, or moving a ghost cursor across the screen. A model that hears you can use any of those, and often should. The mode he thinks is most neglected is speech to action, where you talk and the model reaches for tools. He nearly titled the talk after it, calling voice the next capability overhang. Filling in a government form by talking for five minutes instead of typing for an hour. Steering creative software when your taste outruns your ability to operate it. And eventually the general case: if a model can drive a computer as well as a person, why are you talking to one app instead of the machine. For developers he offers a concrete on ramp, which is that your application already exposes its verbs as endpoints and handlers, and each of those converts fairly directly into a tool a voice model can call. He closes on accessibility, and on a better opening question than what kind of voice agent to build: what role does voice play here at all? Speaker info: - https://x.com/charlierguo - https://www.linkedin.com/in/charlierguo - https://www.ignorance.ai/ Timestamps: 0:00 - The misconception about talking back 1:57 - Three modes, none of them new 2:48 - Speech to speech: coaching, support, translation 4:29 - Speech to action, the capability overhang 6:58 - Turning your existing app into tools 7:48 - Event to speech, and hands free contexts 9:26 - What this unlocks for accessibility 12:45 - Reasoning, tool calls, and preambles 14:23 - Ask what role voice plays --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/OpY6MmZFeHo-voice-agents-just-do-things Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- canonical: >- https://liquidgold.cc/channel/aidotengineer/explore/video/voice-agents-can-just-do-things-charlie-guo-openai-668f52177802 markdown: >- https://liquidgold.cc/channel/aidotengineer/video/voice-agents-can-just-do-things-charlie-guo-openai-668f52177802.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T03:42:39.384Z' language: null children: [] videoId: OpY6MmZFeHo durationSeconds: 950 publishedAt: '2026-09-15T13:30:00.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 24 endSeconds: 125 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=24' heldReason: rights - ordinal: 1 text: null startSeconds: 24 endSeconds: 125 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=24' heldReason: rights - ordinal: 2 text: null startSeconds: 125 endSeconds: 137 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=125' heldReason: rights - ordinal: 3 text: null startSeconds: 137 endSeconds: 173 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=137' heldReason: rights - ordinal: 4 text: null startSeconds: 137 endSeconds: 173 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=137' heldReason: rights - ordinal: 5 text: null startSeconds: 173 endSeconds: 184 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=173' heldReason: rights - ordinal: 6 text: null startSeconds: 184 endSeconds: 274 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=184' heldReason: rights - ordinal: 7 text: null startSeconds: 184 endSeconds: 274 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=184' heldReason: rights - ordinal: 8 text: null startSeconds: 285 endSeconds: 418 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=285' heldReason: rights - ordinal: 9 text: null startSeconds: 285 endSeconds: 418 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=285' heldReason: rights - ordinal: 10 text: null startSeconds: 285 endSeconds: 418 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=285' heldReason: rights - ordinal: 11 text: null startSeconds: 418 endSeconds: 428 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=418' heldReason: rights - ordinal: 12 text: null startSeconds: 428 endSeconds: 471 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=428' heldReason: rights - ordinal: 13 text: null startSeconds: 482 endSeconds: 571 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=482' heldReason: rights - ordinal: 14 text: null startSeconds: 482 endSeconds: 571 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=482' heldReason: rights - ordinal: 15 text: null startSeconds: 571 endSeconds: 582 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=571' heldReason: rights - ordinal: 16 text: null startSeconds: 582 endSeconds: 765 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=582' heldReason: rights - ordinal: 17 text: null startSeconds: 582 endSeconds: 765 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=582' heldReason: rights - ordinal: 18 text: null startSeconds: 582 endSeconds: 765 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=582' heldReason: rights - ordinal: 19 text: null startSeconds: 582 endSeconds: 765 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=582' heldReason: rights - ordinal: 20 text: null startSeconds: 582 endSeconds: 765 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=582' heldReason: rights - ordinal: 21 text: null startSeconds: 765 endSeconds: 776 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=765' heldReason: rights - ordinal: 22 text: null startSeconds: 776 endSeconds: 863 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=776' heldReason: rights - ordinal: 23 text: null startSeconds: 776 endSeconds: 863 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=776' heldReason: rights - ordinal: 24 text: null startSeconds: 776 endSeconds: 863 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=776' heldReason: rights - ordinal: 25 text: null startSeconds: 863 endSeconds: 875 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=863' heldReason: rights - ordinal: 26 text: null startSeconds: 875 endSeconds: 950 watch: 'https://liquidgold.cc/watch/OpY6MmZFeHo?t=875' heldReason: rights --- # Voice Agents Can Just Do Things — Charlie Guo, OpenAI The Moviefone hotline was a speech to speech system. The GPS unit in your parents' car was an event to speech system. Charlie Guo works on developer experience at OpenAI, and he opens by pointing out that the three modes of voice interaction are all decades old, which makes the question not what is newly possible but what we stopped noticing. His actual argument is aimed at an assumption almost every builder makes unexamined: that a voice agent has to answer in speech. It does not. Software spent decades learning to reply by highlighting a field, changing a button, raising a notification, or moving a ghost cursor across the screen. A model that hears you can use any of those, and often should. The mode he thinks is most neglected is speech to action, where you talk and the model reaches for tools. He nearly titled the talk after it, calling voice the next capability overhang. Filling in a government form by talking for five minutes instead of typing for an hour. Steering creative software when your taste outruns your ability to operate it. And eventually the general case: if a model can drive a computer as well as a person, why are you talking to one app instead of the machine. For developers he offers a concrete on ramp, which is that your application already exposes its verbs as endpoints and handlers, and each of those converts fairly directly into a tool a voice model can call. He closes on accessibility, and on a better opening question than what kind of voice agent to build: what role does voice play here at all? Speaker info: - https://x.com/charlierguo - https://www.linkedin.com/in/charlierguo - https://www.ignorance.ai/ Timestamps: 0:00 - The misconception about talking back 1:57 - Three modes, none of them new 2:48 - Speech to speech: coaching, support, translation 4:29 - Speech to action, the capability overhang 6:58 - Turning your existing app into tools 7:48 - Event to speech, and hands free contexts 9:26 - What this unlocks for accessibility 12:45 - Reasoning, tool calls, and preambles 14:23 - Ask what role voice plays --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/OpY6MmZFeHo-voice-agents-just-do-things Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/voice-agents-can-just-do-things-charlie-guo-openai-668f52177802 · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks > Missing: no stored document chunks are available.