Your Child Doesn't Need AI Tools Yet. They Need Computational Thinking.

Parents ask, often in the same week, two opposite questions. Should my eight-year-old be using AI, so they are not left behind? Should I ban it, so they still learn? Both treat the tool as the subject. The subject is whether the child can hold a problem still long enough to break it into steps.

That is computational thinking. Unfashionable phrase. It means: say what happens first. Notice the repeat. Name the branch. When it fails, look at the failure. None of that requires a model. All of it is ruined if a model writes the function and the child reads it out in a confident voice.

“Left behind” is the fear that sells child AI products. The fear is misplaced. The models will be better next year. The habit of looking at an error will not install itself at sixteen if we skipped it at nine. You cannot prompt your way into that habit. You get it by sitting with a red line.

What we refuse in First Code and App Builders

We do not let a 7–14-year-old paste a generated game and call it theirs. We will not mark a page they cannot change. If the mentor asks “why did it turn left?” and the answer is “the AI said so,” the work is not done. The child has used a search engine with better manners.

Between sessions we ask for a small artefact — a function, a scene, a log — not a generated essay about what they would have built. If the week produced nothing they can defend, that is the conversation. We do not paper it with a chatbot paragraph for the family portal.

This is not a moral panic about technology. It is a standard. Calculators did not remove arithmetic; they moved it. Children still had to know what they were asking the calculator to do. A language model will write fluent rubbish with the same confidence as a correct function. If the child cannot tell the difference, they are not learning to code. They are learning to submit.

If they cannot explain the answer, the tool did the thinking. That is not a class. That is a search.

What to do at home instead

Play the game they wrote. Ask what a bee does. Open the page on the family laptop. Do not install a “learn to prompt” app for a nine-year-old and call it computing. If they are stuck, they can bring the stuckness to the next live class. That is what the four-person room is for.

If they are curious about ChatGPT, tell the truth: it is a tool adults use, and it lies in complete sentences. They will meet it properly when they can catch the lie. Until then, the honest homework is the function that does not run yet.

When AI belongs — and how

AI Engineers (15–18) is where the tools enter as objects of study, not as invisible homework ghosts. They train something small. They show a score or a log. They write down what the system will not do. They can pause it. That is a different sentence from “ChatGPT made my app.”

A camera that names a cup is interesting only if they can show you a wrong label and a next experiment. A chatbot that answers from their notes is interesting only if they can show you a question it should refuse. We are not training prompt technicians. We are training people who can still think when the tool is closed.

If a family wants a child who can prompt well and think poorly, other products will sell that. We would rather they think first. The tools will still be there at sixteen. The habit of looking at the error will not appear by magic if we skip it at nine.