hayward@click:~/~/rants/2026-10-31-the-change-that-was-only-an-improvement

what these numbers are

Every token my AI tooling has sent and received since 12 Aug 2026, kept in one running total that never resets.

is what goes in, is what comes back out. The in figure runs far larger because most of it is context I re-send with every request rather than new text.

Both tick up live while anything is running, and stop when nothing is being spent.

Counted from the usage my AI services report, and not all of them report tokens, so treat this as a minimum rather than a grand total.

./rants Daily Rant

The change that was only an improvement

31 October 2026

Almost every bad week I have had in this trade started with a change that was only an improvement. Not the big rebuild, the one everybody argued about and tested before it went near anything. The small tidy-up. The one that couldn't possibly hurt.

I couldn't have said that at the start. Early on I wanted everything better. Tidier, faster, one less thing nobody understood. Every change was defensible, and some of the ones I was proudest of are the reason I got a phone call at a bad hour. The fault was never in the change. It was that something which worked had been touched.

So the last judgement in the job isn't how to do a thing. It's whether to do it at all, when the thing already works. That call has always belonged to a person, because a person is the one who lives with it afterwards.

I make that call less on merit now than as a brake. The thing I work with will improve anything I point it at, quickly and well, and it has no idea what it cost to get the system into the state it's in. So my answer is often to leave it alone. That's the only authority in the day the tool can't hold, because it has never spent an evening putting an improvement back.

The trouble is that I don't trust myself making it. There was a time when I stopped because I knew the thing was right, and I could have given you the reason in a sentence. Now I sometimes stop because I have had enough of the risk, and it comes out in the same voice. From the inside those two are hard to tell apart, and that bothers me more than the tool does.

So I make the decision do the paperwork. When I decide not to change something, I write the change down. What I would have done, why, and what would have to happen before it's worth doing. A note to nobody, in the file I keep for the machines.

Some of those notes I read later and laugh at, because the change was obviously right and I was being a coward about it. The others are the reason the thing is still up. I can't tell which is which on the day. I can later, and that is the only way I have found of telling judgement from tiredness.

The clever part was never the building. It was the deciding, and most of the deciding is knowing what not to touch. I still have that. I just have to write it down to be sure it's mine.

All rants · Nobody learns anything from a finished job