Essay 1 of The Almost-Right Machine is out. Each essay gets a short Whisper here: the scene it opens with, the sentence it turns on, and one question for you. The full essay is on Leanpub, free.
I do not remember exactly what I asked for. I remember that the answer came back in less than a minute.
It was the unglamorous half of a benchmark: filter fourteen million programs down to the few that deserved the name, recover their tests, generate new ones where there were too few. An afternoon of care and then a week of being wrong in small ways. I have done that week many times. Instead, I described it to a coding assistant, mostly to see what would happen.
The shape was right. The order was mine, the names were ones I could have chosen, and the files looked like a dataset. It was not right. It was almost right. Wherever I had left something open, it had decided for me, quietly, and moved on. I found the decisions in minutes, because I knew which lines to read slowly. Until that morning I had not thought of that knowledge as separate from the ability to write the code myself.
Then I kept working, and I enjoyed it. Only days later did I notice what had been happening to me while I was having fun. Something had generated, in under a minute, the thing I had planned my afternoon around. That is an inferiority experience, and I use the phrase deliberately. Most of what is written about engineers and AI describes everything except that moment at the keyboard.
The essay is about the third kind of result. Wrong is cheap, and right is cheap. Almost right runs, passes the tests, and somewhere inside has made a decision you did not ask for and were not told about. It will not go away as the models improve, because it is not a failure of capability. It is the shape of authorless work.
The machine was almost right. What I did next was the only part of that day that was mine.
A question for you, since this is why I am writing in the open: when something you were good at came back finished in under a minute, what did you feel, and what did you do next? Reply and tell me. If you think the inferiority framing is wrong, even better.