I direct, I don't type
I shipped a feature I never typed — what changes when the thinking moves upstream from syntax to intent, and the judgment that won't hand off
Yesterday I shipped a feature I never typed. I described it, corrected it twice, and signed off. The keyboard was almost idle.
For twenty-five years the work lived in my hands. I thought in syntax. The shape of a solution arrived as a sequence of keystrokes, and the thinking and the typing were the same motion. Now they've come apart. I think in outcomes and hand the keystrokes off.
What surprised me is how much harder the thinking got, not easier. When I wrote every line, the code was the argument — if I could express it, I understood it. Now I have to know exactly what I want before a single character exists, and say it precisely enough that a fast, literal collaborator can't misread me. Vague intent produces confident nonsense. The agent will build whatever I describe, beautifully, including the wrong thing.
So the job moved upstream. I spend my hours deciding, specifying, drawing the boundary of the problem before anyone touches it. The leverage is real — a week of work compresses into an afternoon. But the afternoon is denser. There's no idle stretch where my fingers move and my mind wanders into the solution. It's all decision now.
And here's the thing that keeps surfacing. The agent builds fast and clean, but it builds from nothing each time — no memory of why we did it the other way last spring, no sense of the conventions that aren't written down anywhere. It guesses at my context and guesses well. Well enough that I have to stay sharp to notice it's still guessing.
Holding the intent is the work. Knowing whether the thing in front of me deserves my trust — that's the part I can't hand off, and I'm not sure I'd want to.