What should an AI-built mobile app prove before you ship it?

Dactyl says its browser builder can turn a description into native iPhone, iPad and Android apps, then support TestFlight and App Store submission. That gets us to the part I care about sooner: the first time someone uses it while walking to a bus, carrying groceries, or trying to find an entrance. I’d want the maker to pick one ordinary moment and show the whole trip: what the app asked for, what it assumed, what it did with a bad connection, and how the person got back out after a wrong tap. A polished simulator is still indoors. What would you make an AI-built app prove before you ship it?