← All essays

    Excerpted from 55 and Out, AI Automation

    Letting Claude Plan My Week

    ·3 min read

    The Sunday ritual:

    1. Export the upcoming week's calendar to a text format.
    2. Paste in my three monthly goals and the projects they break into.
    3. Paste in last week's actual time use (from a small tracking app I run on my laptop).
    4. Send it all to Claude with a prompt that's been refined over six months.
    5. Get back a draft week — proposed deep-work blocks, suggested meeting groupings, recommended exercise slots, an honest read on where last week went wrong.

    I keep about 70% of what comes back. The other 30% I overrule for reasons the model couldn't know — my wife's schedule, an upcoming travel day I haven't told it about, a meeting I'm dreading and need to put on Tuesday morning to get it over with.

    What it's good at:

    • Spotting that I've scheduled three back-to-back meetings on Wednesday with no break
    • Noticing I haven't blocked time for the most important project of the month
    • Suggesting that the deep-work block I keep putting on Friday afternoon (when I'm tired) would be better on Tuesday morning
    • Catching that I said I'd exercise four times this week and only blocked two

    What it's bad at:

    • Knowing how I actually feel — sometimes Friday afternoon deep work is exactly what I need because I have momentum
    • Personality dynamics — it doesn't know that Bob is easier to deal with before lunch
    • Energy budgeting — it treats every hour as equivalent. They aren't.

    What surprised me: how often the model called out the same blind spots my wife would have called out. They're patterns. I keep falling into them. Having an outside observer (even a stochastic one) point at them is useful.

    What broke: an early version of the prompt was too prescriptive. The output read like a drill sergeant. I rewrote it to use language like "you may want to consider" rather than "you should." Same recommendations, much easier to read on a Sunday afternoon.

    What I'd do again: this, weekly, indefinitely. What I wouldn't: let the model schedule for me automatically. The act of reviewing the draft and saying "no, not that" is where the value is. If I just accepted the output, I'd lose the discernment muscle entirely.

    Total time per week: about 20 minutes. Net effect: roughly 4–6 hours saved during the week from fewer reschedules, less drift, and better-grouped tasks. Pretty good ROI for a Sunday afternoon ritual.

    Share

    This essay is excerpted from the book.