Where the Fuel Went

The Setup

In June I wrote that the fuel was never the problem, that I could sit with a full tank and a finished engine and still not catch, and that the missing piece was always the spark. I believed it when I wrote it. By the end of the summer it had stopped being true, and what I had not prepared for was how it feels when the spark is fine and the tank is the part that is empty.

It did not arrive as one bad week. It arrived as a slow lowering of the floor, where the things that used to cost me a little began to cost me a lot, and the things that used to restore me began to feel like more items on the list. I kept showing up, and showing up became the whole of what I could do. In the middle of September I took two weeks away from work, which I had been postponing for months on the reasoning that a better time would come. The better time never volunteers itself.

I did not plan to build anything during those two weeks. Somewhere in the second week I opened a small tool I had started in April and abandoned, and for a few days I could not stop working on it. I have spent the time since trying to understand why that felt like rest.

The Thinking

The easy explanation is that I found my passion again, and I do not trust it, because the passion was never missing. What was missing was something more specific, and it took the contrast to show it to me.

For months the fuel had been spent every day on work where the cost of each step was unclear and the result of each step was invisible. I could put in a full day and be unable to say, at the end of it, whether anything had landed. Effort went in and nothing came back that I could point at, and after enough of those days the tank does not just run low, it stops believing it will ever be refilled.

The tool was the opposite in every respect. Every problem had an edge. Something either worked or it did not, and when it did not, I knew why. What came back during those days was not energy from nowhere. It was the experience of effort turning into something I could see, which I had been running without for far longer than I had noticed. I used to think rest meant doing nothing. It turns out that for me, at least some of the time, rest means doing something whose result I can actually hold.

The Work

I did not write that tool line by line. Coding agents wrote almost all of it, in languages I could not have written from a blank page, and the experience changed what I believe about building with AI more than anything I have read about it.

The honest version is neither that I built it despite the agents nor that the agents built it for me. It is that the work moved. The part that used to drain me, the syntax and the version mismatches and the hours lost to a missing flag, became cheap almost overnight. What remained was every decision about what the thing should do, what it must never do, what counts as safe, and what a stranger should see when it breaks. By the end, the only part of the project that felt like it was written in my handwriting was the list of choices I had made and why. That list was the work.

For years I had quietly measured my capability by the output I could not produce myself, and judgment never counted because it never looked like output. Watching the cost of typing collapse made the scarce input obvious. I had it all along. I had just been spending it in places where nobody, including me, could see it being spent.

There was a less comfortable discovery inside the same week. An agent does not get tired. It will build for as long as I keep asking, it will never tell me the loop has gone on too long, and it will cheerfully accept a sixth idea at two in the morning with the same energy it brought to the first. When the machine has no limit, the person directing it becomes the only brake in the system. I had just spent a summer proving I was not a reliable brake.

The one rule in the tool I keep coming back to is the smallest. If the thing crashes three times in ten minutes, it stops restarting itself and waits for someone to look, instead of looping forever. I wrote it for a piece of software and only noticed afterwards that it was the rule I had never written for myself. For most of the year my response to a crash was to restart immediately and keep restarting, each restart costing more than the last, with nothing anywhere allowed to say that the loop itself was the failure.

The Edge

I am not certain the two weeks were rest. I have a theory that they were, because the tank is fuller now than it was in August, and because the building felt nothing like the grinding that emptied it. I also know that a problem with clean edges is the most comfortable place to stand when the questions without edges are waiting outside, and that some of those days may have been a very productive way of not sitting still.

What I am taking back with me is smaller than a conclusion. The fuel was not gone, it was going somewhere I could not see, and the fix is less about finding more of it than about noticing where it leaks. Work that shows me its result refills me, and work that hides it drains me, and for most of this year I did not know there was a difference. The next time the floor starts lowering, I would like to recognise the third crash for what it is, and stop before the fourth.