While most discussions of machine learning concentrate on the underlying mechanics, this talk discusses their capabilities, strengths, and weaknesses at the level of how they can be applied to a wide variety of real-world applications. Capabilities discussed are classification, end-to-end behavior, generative outputs, and foundation model applications. Challenges discussed include bias, validation, edge cases, hallucinations, autonowashing, AI safety, and accountability. The evergreen concept of the 90/10 principle cuts both ways for AI, with solving the last 10% of building dependable systems likely to make the difference between winning and losing technology bets.