← All essays

    Excerpted from 55 and Out, AI Automation

    What I'd Automate First If I Were Starting Over

    ·3 min read

    If a friend asked me today, "I want to use AI to make my life easier — where do I start?", here's the order I'd actually give them.

    Day one: a transcription pipeline for voice notes. Pick the simplest voice memo app on your phone, drop the recordings into a cloud folder, set up a script (or pay $5/month for a service) that transcribes them and emails you the text. This single thing changes how many ideas you keep. Cost: under $10/month or one Saturday to build.

    Day two: an email triage prompt. Not a full system. Just a saved prompt that you can paste your inbox subjects into to get a quick "what matters today" sort. Use it manually for two weeks before you build any automation. You'll learn what kind of automation you actually need by doing it by hand first.

    Day three: a markdown file of prompts. Start it empty. Every time you write a prompt that works well, paste it in. After a month you'll have ten that are worth keeping. After six months you'll have fifty. That file is more valuable than any AI app you'll buy.

    Day four: a weekly review draft. Have a model generate the first 70% of your weekly review from your calendar and notes. You finish the last 30%. This is where most of the time savings come from for me, and it took me a year to figure out.

    Day five: stop and use the four things above for a month before you build anything else.

    That last point is the most important one. The mistake I made — the mistake almost everyone makes — is to keep building automations because building is fun, before you've actually used the ones you have. Most of the value of any automation comes from the second month of using it, when you've learned its weaknesses. If you're constantly building new ones, you never get to that second month.

    What I would not start with:

    • A custom GPT with a "personality." Cute, low value, distracting.
    • A vector database of all your notes. Beautiful, almost never queried, expensive to maintain.
    • An agent that does multi-step tasks autonomously. Not yet. The reliability isn't there for high-stakes work.
    • Any "AI everything" suite. The best AI tools are small, sharp, and one-purpose.

    What I would invest more in:

    • Voice → text capture (the bottleneck of ideas).
    • Text → structure (the bottleneck of action).
    • Pattern recognition over your own data (the bottleneck of self-knowledge).

    If I had two years back, I'd build less and use what I built more. The marginal value of the next automation is almost always less than the marginal value of using the last one better.

    That's the whole report. Thirty essays, distilled to: capture more, decide better, build less, use what you have for longer than feels right. The tools will keep changing. The discipline doesn't.

    Share

    This essay is excerpted from the book.