How do you decide when an AI feature is ready for the homepage vs. hidden away?

The rule I use: a feature earns top-level navigation by being reliable, not by being exciting. New AI features are exciting on day one and embarrassing on day three when someone hits an edge case. So I embed them inside a relevant page first — low stakes, real users, real failure data — and only promote them to the main menu once they stop surprising me badly. There's a discipline reason too. Every menu item you add is a promise to the visitor that this thing works and matters. A navigation full of half-baked features makes the whole site feel unfinished. I'd rather have four menu items that all deliver than eight where three are experiments. The AI question box graduates to the menu the day I stop wincing at its mistakes — not before.