The work was the point

Norway is about to tell its elementary schools to put the AI away. The government moved this week to bar most generative-AI tools from primary classrooms, on the grounds of what happens to a child's learning when the answer shows up before the effort does. A tech-forward Nordic country, pumping the brakes on the thing every ministry calls the future. The reason it's doing that is the interesting part.

Set the decision beside two other stories from the same week and a pattern surfaces that none of them states alone. In each, someone draws a line around a piece of work and says: don't let the machine do this part — not because it can't, but because the doing was the point. The output was never the product. The friction of producing it was.

The output was never the product. The friction of producing it was.

The sharpest version comes from a site-reliability engineer dreading a future of LLM-written incident reports. His objection isn't that the model writes badly. It's that writing is how you find out whether you actually understand what happened — the report is a test the writer has to pass, and the machine quietly sits the test for them.

The seduction is that you can just ask, and it'll do it. What disappears is the step where a human confronts whether the explanation actually fits the evidence — the step that catches the coupling the model invented and the interaction it missed. A postmortem can have the right shape and be wrong, and unlike code, no test fails. Nobody notices, because nobody did the synthesizing.

Norway is making the same bet about nine-year-olds. The fear isn't that a chatbot gets the arithmetic wrong; it's that it gets it right, instantly, and the child never builds the thing the exercise existed to build. And a working mathematician, writing to Tyler Cowen this week about how AI will reshape his field, arrives at the same worry from the top of the ladder: once the proof and the paper can be generated, what happens to the understanding that writing the proof used to force?

The obvious objection is that this is the calculator panic in a new coat. We let machines do arithmetic and the sky held; offloading drudgery is how every field moves forward. True — and the incident-report engineer concedes it without flinching. Use the model to gather the logs, build the timeline, kill the toil; he has no issue there. The line was never between hand work and machine work. It runs between the parts of a task that are merely laborious and the one part that is the task — the synthesis, the confrontation, the moment a fuzzy understanding gets forced into something exact. Automate the toil and you free people up. Automate the confrontation and you keep the document while losing the only thing it was ever standing in for.

That's the line all three are drawing at once — in a classroom, a postmortem, and a math department. The essay, the report, the proof were always just the residue. The understanding was the product, and it only ever came from doing the work. Hand the work to the machine and the residue still shows up on time. You just won't have learned a thing.