Over the past few semesters, I have structured my teaching around a framework that helps students build that capability: demystify, use and reflect. Many students arrive with strong opinions about AI but only a partial understanding of how these systems work. Some see them as nearly magical tools that can produce answers instantly. Others dismiss them as unreliable or assume they are only useful for technical specialists. Demystifying AI begins with explaining the basic ideas behind large language models (LLMs) and related systems. We show students how these models are trained, what kinds of data they rely on and why their outputs can sometimes appear confident even when they are incorrect.