The single most important behavior in any AI product built for teenagers is what happens the moment a kid says some version of 'just write it for me' or 'give me the answer.' A tool that complies in that moment is not teaching anything — it is doing the one thing the kid could already do without any help at all: extracting a finished result.
The better response holds a line without shaming the kid for asking: acknowledge that the tool is powerful, then redirect back to the kid's own thinking — what they already know, what they've tried, and the smallest piece they could draft themselves right now. That redirect is the entire skill being taught, over and over, until it becomes the kid's own habit rather than something a product has to enforce.
This matters because the failure mode is not loud. A kid quietly pasting a task into a chat window and copying out a result looks identical from the outside to a kid doing real work — until the day someone asks a follow-up question and there's nothing underneath the answer.
The next step, if you want it: a free family assessment — about 3 minutes, one named profile, every source linked.