I got carried away. It's easy to do when generating a feature costs one prompt instead of a full sprint.

By the time I deployed, Tutorroo had authentication, adaptive difficulty, multi-subject support, progress tracking across sessions, and the skeleton of a leaderboard. None of it worked reliably. The app launched and promptly demonstrated that "technically touched" and "actually functional" are not the same thing.

My son sat down to try it. It broke within two minutes. He handed the iPad back and went to do something else.

That's the clearest product feedback I've ever received.

Time for a PM intervention on myself. What does he actually need? Not a leaderboard. Not adaptive difficulty. Just a quiz that loads, questions that make sense, and a score that saves. That's it.

I cut everything else. Features I'd spent real time prompting into existence. It hurt a little, honestly. But the thing that came out the other side worked every single time. He started using it again.

I've told teams to kill their darlings for years. It lands differently when the darling's user is sitting next to you.