The human component here is to understand what the child is trying to do, what the available options are, and what level of guidance he needs to steer him towards the option that makes sense (ie easy enough for a simple game for a child, just a small step from what he already knows, etc).

This all comes down to "judgement".

The current state-of-the-art isn't yet there yet (saying this as someone who spent the last year building on current LLM tech).

I think it's possible for AI to take on the role in the future, when it's capable of reasoning at a much higher level than it's at now. We already have big context windows, ability for AI to "see" your screen, etc. But what we're missing is good judgement.