← All essays

    Excerpted from 55 and Out, AI Automation

    Voice-to-Text for Capturing Ideas: My Stack

    ·3 min read

    I had a notebook in my back pocket for fifteen years. It worked. But ideas mostly arrived during walks, drives, and showers, none of which are good notebook environments. So most ideas were lost.

    Here's the stack that fixed it.

    Tool 1: a voice memo app on my phone. I press the home screen widget, talk for 30 seconds to two minutes, the file gets saved with a timestamp. Single button. No app to open. No category to choose. Just talk.

    Tool 2: a watcher script on a small server. Every hour it pulls new voice memos out of the cloud folder, sends each one to a transcription model, and saves the result as a markdown file in my notes app. Filename is the date and the first six words of the transcript so I can scan visually.

    Tool 3: a daily summary. At 6 PM the server runs a small script that takes the day's transcripts, sends them to an LLM, and generates a one-page summary grouped by theme — "ideas about the book," "tasks," "people I mentioned," "open questions." The summary lands at the top of my notes app for the day.

    Tool 4: a weekly distillation. Every Sunday morning the system summarizes the week of summaries into a much shorter doc — usually about a third of a page — that I read with coffee. The patterns from this are the most valuable output of the whole stack. I see what's been on my mind without having to remember what's been on my mind.

    Total cost: about $4/month in API calls. Time to build: a long Saturday and a half-day to refine.

    What broke: early on the transcription model misheard certain words I use a lot ("compounding" became "compounding," sometimes; "Audible" became "audible" lowercase, breaking my pattern matching). I added a small post-processing step with my custom dictionary. Took an hour.

    What surprised me: how many ideas I'd been losing. The first week of summaries had about 40 distinct ideas worth keeping. I had been generating those at the same rate for years and remembering maybe three a week.

    What I'd do again: all of it. What I wouldn't: use a fancy notes app. Plain markdown files in a folder are forever; product UIs come and go.

    The deeper point: capture is the bottleneck, not creativity. Most people have plenty of good ideas. They just lose most of them between brain and paper. Close that gap and the rest sorts itself out.

    Share

    This essay is excerpted from the book.