Across our AI-Native Foundations classes, the same frustrations repeatedly come up among people using Microsoft Copilot. "It loses the thread." "It sounds convincing, but it gets things wrong." Trust is a big issue, particularly for people who don't have enough expertise to know when it is making things up. I have a theory about what is going on. It almost always turns out to be the same thing.

Here's what Microsoft doesn't make obvious: there are two different products. The first is Microsoft ...