AI generated more documents, not less work. A designer built a voice app to fight back.
The author, a product designer, built ARC, Audio Review Companion, using Gemini Flash Live in Google AI Studio alongside Claude Design running in parallel windows. The core function: load a Google Doc, have it read aloud verbatim, stop it with a voice command, and dictate comments that write back into the doc, hands-free, away from a screen. Existing tools failed the test. Speechify and Wisperflow could not post conversational comments to Google Docs. NotebookLM only summarizes. Google announced native voice reading at I/O 26 but without a commenting layer. The gap was real.
The build process is worth reading in detail. The author ran the underlying app in AI Studio and the design system in Claude Design simultaneously, feeding Claude-generated design specs and screenshots back into Gemini 3.1 Pro for integration. Claude Design produced workable visuals but introduced inconsistencies without tight prompting, and the Pro plan token limits bit hard. The result is a single-model reader and comment taker where navigation commands like 'go to the competitive analysis' work because one model holds the full document context. The broader argument, that AI raised the volume of cognitive labor rather than reducing it, is specific enough to be worth your time.
[READ ORIGINAL →]