The sommelier analogy: understand what the client wants before deploying AI

A good restaurant doesn't just hand a guest a glass of Pinot Noir without asking. It has a sommelier whose job is to understand your palate — do you like dry wines, wines from southern France or northern France — and then recommend accordingly. An FDE works the same way. You have to understand what the client's system and set of tasks actually are before you recommend or deploy a specific model or tool. If you just hammer a specific model without that understanding, it might work for some clients, but it won't work for most — and that's exactly why most AI pilots fail.