In this article, I examine Japan's newly updated guidelines for AI in education and argue that they get one thing right while missing something crucial. They tell teachers how to use generative AI responsibly but they do not ask whether a given lesson needs it in the first place. Drawing on evidence that AI can speed up homework while quietly eroding exam performance, I argue that the question of 'when' has to come before the question of 'how', and I lay out a framework schools can use to test that assumption rather than take it on faith. You can read the full analysis on Society & AI: Japan's AI Education Guidelines Skip the Question That Matters Most