Excerpted from 55 and Out, AI Automation
The Three Automations That Gave Me Back 6 Hours a Week
· 3 min read
I tracked my time for two weeks before I built any of these. That's the part most people skip. You can't automate what you haven't measured, because you'll automate the thing that feels annoying instead of the thing that's actually expensive.
Here are the three that paid off.
One: inbox triage. A small script runs every morning at 6 AM. It pulls the last 24 hours of email, sends each subject + first 200 characters to Claude with a prompt that classifies it into one of five buckets — reply needed, FYI, receipt, newsletter, trash. The output goes into a single markdown file I open with coffee. I read the FYI bucket in two minutes, reply to the reply-needed bucket from a clean state, and never see the rest. Time saved: about 45 minutes a day.
Two: meeting notes. I record every call (with consent), drop the audio file into a folder, and a watcher script transcribes it via Whisper, then sends the transcript to an LLM with a prompt that produces three things: a one-paragraph summary, a list of action items with owners, and any decisions made. The output lands in my notes app within ten minutes of the call ending. I used to spend 15 minutes after every meeting writing this up by hand. I have four meetings most days. That's an hour back.
Three: weekly review draft. Every Friday at 3 PM, a script gathers the week — calendar events, completed tasks, journal entries, and the inbox classifications above — and asks an LLM to draft my weekly review. It's never quite right. But it's 70% there, and editing a 70% draft takes ten minutes instead of the hour it took to write from scratch.
What broke: the inbox classifier missed a flight cancellation once because the airline buried it in marketing copy. I added a rule: anything from a known travel domain bypasses the classifier and goes straight to reply needed. Lesson — automations need allow-lists for the things you can't afford to miss.
What I'd do again: all three. What I wouldn't: the calendar-blocking automation I built and killed within a week. It was solving a problem I didn't actually have, which is the most expensive kind of automation there is.