i was the model
I.
I made the mistakes on purpose. Someone would tell me what to type and I’d type it, including the off-by-one I could see coming, then run it, watch it throw, and fix it without making anything of the error. Every demo they had ever watched ran clean on the first try, which is a lie about how the work goes. The lesson, though none of us would have called it that, was in watching the screen go red and seeing that it didn’t matter.
A beginner spends everything they have on the syntax. There is nothing left over to plan the next line, let alone hold the shape of the whole program.
There’s a clean way to say what I was. I took instructions in plain language and returned working code. There was no design opinion I was allowed to act on. Quick, and unbothered by being wrong, which is most of what autocomplete is. I’d been my students’ LLM, a person-shaped version of the thing everyone now types into.
II.
The spec leaves out the part that taught them anything. With the syntax on my keyboard instead of theirs, they had attention left over, and for the first time they could plan past the line they were on. The freed attention only made room. What filled it was watching someone treat a red screen as ordinary, a picture they could keep for the first time they broke something alone. An LLM is a model that runs. I was a model they copied.
The tool clears the same attention, and then it does the thing I spent years refusing to do. Ask it for the answer and it comes back finished, the wrong guesses edited out, the way every too-perfect tutorial does. It teaches the same lie: that the work looks like its output, and that your own mess is a verdict on you. Ask it to work in the open and it will hit the red screen and recover in front of you, which looks like the real thing. But my students couldn’t get a line out of me without reasoning their way to it first. The tool asks for nothing, and on a deadline you will let it do the thinking too.
So the question my room settled by accident is open again, for everyone at once. The attention is freed. Does it go to judgment, or does it just go?
III.
In my room it went to judgment, and I know because I saw it later, at a remove. Students who had sat through enough of my red screens started hitting their own, alone, without freezing. They planned past the line. They treated their own mess as the work, not a verdict on it.
I want that to be the lesson, which is a good reason to distrust it. But the ones who came in frozen had mostly been shown things working. The attention can become judgment, and it needs a real process to copy. I’m betting on anyone who got to watch the work go wrong and come back, and less sure about the ones who only saw it land clean.
It is the ordinary trade. A tool makes a hard thing skippable, and the craft behind it goes quiet. Some years on, the work starts missing a quality nobody can name, and people go looking for whoever still remembers how it was done. When animation gave up the painted cel, the look of it took years to approximate, and only because people had studied what it was while it was still ordinary.
The thing to keep is the room: a person doing the work in front of you, badly and then less badly, who made you reason out every line before it would appear.