Shipping is a muscle

The job of running a company is to make decisions about software without touching it. You set direction, you hire people who are better than you at the parts that matter, and you spend your attention on the few calls only you can make. I got good at that. I'd defend it as a way to spend a decade.

What it cost me, quietly, was the feel. The difference between a change that's clean and one that's merely clever stops being obvious when you haven't made one in years. I could still review a design. I could not have built it, and after a while I stopped being able to tell those two things apart.

Coming back, the surprise isn't that I'm rusty — I am — it's that the ground moved under me while I was away. The editor argues now. I describe what I want and a draft appears, wrong in interesting ways, and the work becomes the thing I was always good at: deciding what's worth keeping.

Here's a representative half hour. I wanted the EuroRails bot to stop re-scoring routes it had already evaluated the same turn. Ten years ago that's an afternoon of remembering how caching works. Now it's a conversation:

$ claude "cache the bot's route scores per turn,
          keyed by board state, with a --no-cache flag"

  → drafted scoreCache in routePlanner.ts
  → wired --no-cache through the CLI
  → wrote two tests. one failed. fixed it.

I read every line of that before it landed. Two were wrong for a reason the model couldn't have known — board state isn't stable between sub-phases, so the cache key had to be the canonical state, not the raw one. That correction is the job. It's judgment, and judgment is the thing the years actually gave me.

So I'm not precious about who writes the first draft. I'm precious about what ships. The muscle I'm rebuilding is the boring one nobody puts on a slide: closing the last gap between a thing that works on my machine and a thing a stranger can use without me in the room.

The way you rebuild a muscle is reps. Something small, finished, in public, every week. This post is one. The cache is another. Next week it's something else, and the week after that the gap is a little smaller. That's the whole plan.