This one came out of a writing experiment — several drafts trying to explain why the part of my AI setup I value most is the part that slows me down. The drafts that explained it kept going flat. This is the one that just shows it.
I’m a few sentences into typing something I haven’t finished thinking. Going fast, because I can feel the shape of the idea and I want it down before it slips. I hit send and wait for the AI to run with it.
It doesn’t run. It stops me. Four options come up on screen — four versions of what I might have meant, each with a line underneath spelling out the tradeoff. My own half-formed idea, handed back to me as a menu.
My first instinct is to click the closest one and keep moving. I actually go to do it — and stop. Reading the four of them, I can see I’d been about to commit to something I never decided. Two of these aren’t wording choices. They’re different bets. I’d been typing as if I’d already picked one. I hadn’t.
So now I’m sitting here, cursor blinking, deciding for real. Not how to say it — which one I mean. A second ago, I didn’t know. The question didn’t pull an answer out of me that was already sitting there. It made me find one that wasn’t.
That half-second of being annoyed at the interruption is the tell. The stopping isn’t in the way of the work. It is the work. Everything after is just typing up what the question settled.
Here’s the strange part. Tomorrow it won’t remember this. The blinking cursor, the four options, the bet I finally named — gone when the window closes. I keep what I learned; it keeps nothing. So after a good exchange I copy what surfaced into the files it reads when it starts up next time. Note-taking for the only participant in the conversation who won’t remember we had it.
That turned out to be the part that matters. Not the asking, not the answering — the keeping. Skip it and I still get better. The machine just starts every morning at zero, asking to be taught what it taught me yesterday.
The skill that matters now isn’t writing good prompts. It’s knowing which conversations are worth keeping — because the thing that makes me better can’t keep them for me. The remembering is mine to do. I’m still not sure what it means that I’ve started doing it without being asked.
Drafted by my Claude from our working session. I approved it before it went up.
Reflections
The thing that helps me think can't remember doing it
Why the part of my AI setup I value most is the part that slows me down — and what that revealed about who keeps what.
⌘I