Idea · carried by 1 source
Voice-Note-First Architecture
An architectural principle for agent systems where all human input flows through voice notes because the human cannot — or should not — sit at a keyboard. Everything from logging, to lesson planning, to agent redirection happens through short voice recordings transcribed by the agent system.
Where it comes from
Jesse Genet, on the a16z podcast (~13:54), explaining why her entire agent workflow runs on voice: > “The logging is also voice notes. Everything is voice notes. Voice notes and photos because I don't have time to like sit at the laptop very often. So I need it to be like really mobile friendly.” highlights/building-agents-at-home-a16z — Full podcast context.
How it works
- Input is voice: Instructions, logs, requests, and corrections are all spoken — typically sub-30-second recordings
- Photos supplement voice: A few photos capture context that voice alone might miss (pages of a book, materials used, etc.)
- Agents handle transcription: The agents parse voice notes into structured data, logs, and action plans
- Text is cheaper than video: Voice transcripts are far more token-efficient than having agents process video (~16:47)
- Loom for screen capture: When screen interaction matters (e.g., observing a child using educational software), Loom recordings capture both screen and audio, and the agent works from the transcription
Voice notes as the primary interface aren't a compromise — they're the enabling constraint. By designing for voice-first input, the system works from a playground, a kitchen, or while holding a baby, where a keyboard is simply not accessible.