AI isn't a subject. It's a way of working.

A school that adds an AI class has done something reasonable. It has also, usually, done something ineffective. The class gets forty minutes on a Thursday, a teacher who learned the material two weeks ahead of the students, and a curriculum written before the current generation of tools existed. Everyone leaves having heard about AI. Almost nobody leaves able to use it.

The problem isn't effort. It's category. AI isn't a subject like chemistry, with a stable body of knowledge that can be sequenced and examined. It's closer to writing, or spreadsheets, or English as a second language: a capability that only becomes real when it's applied to something else.

What "knowing AI" actually looks like

Ask an adult who uses AI well what they know, and they won't describe a syllabus. They'll describe habits.

They know which task is worth handing to a model and which isn't. They know how to describe what they want with enough precision to get something usable back. They know how to check the output, because they've been burned by confident nonsense before. They know when the tool has taken them as far as it can and the rest is their own judgement.

None of that is teachable in the abstract. All of it is learnable through repetition on real work.

Why projects do what lessons can't

A student who watches a lesson about AI image generation learns that AI image generation exists. A student who has to produce the cover art for a book they're publishing learns what the tool is actually like: that the first attempt is generic, that the third attempt is better because the description got sharper, that some things it simply refuses to do well, and that the final choice is still theirs.

The second student didn't study AI. They used it, under the pressure of needing a result. That pressure is the entire teaching mechanism, and no lesson plan can substitute for it.

This is why every Fuvia program ends in a finished piece of work rather than an assessment. The project isn't a demonstration of what was learned afterwards. The project is where the learning happens.

The fluency that transfers

Parents sometimes ask whether the specific tools their child learns will still exist in five years. Some won't. That's fine, and it's not the point.

What transfers is the underlying posture: treating these systems as capable but unreliable collaborators, being specific about what you want, verifying before trusting, and knowing where your own contribution sits. A student who has built that posture on today's tools will pick up next year's in an afternoon. A student who memorised menu positions in a particular app will not.

The skill isn't the tool. The skill is knowing what to ask of it, and what to check afterwards.

What this means for schools

Schools don't need to build an AI curriculum from scratch, and most can't. What they need is a structure where students do real work with real tools regularly enough for the habits to form, with someone available when a student gets stuck.

That's a solvable problem. It doesn't require hiring an AI specialist, and it doesn't require rewriting the timetable. It requires giving students something to build, the tools to build it with, and enough time for the second attempt to be better than the first.