You already know: what a forward deployed engineer does (previous lesson); evaluation and eval sets (D10); RAG, agents and prompting (D7–D9); cost per token (D12).
Why this matters for Atlas. Most failed AI projects were doomed in the first two weeks, not the last two months. They started from a technology ("we need an agent") instead of a decision, never agreed what "working" means, and discovered the data problem after the budget was spent. Scoping is the FDE's highest-leverage skill: an hour of good questions saves a quarter of rework. This lesson gives you a repeatable process: find the decision, test feasibility in a week, define success before building, register the risks, and write the one-page scope that both sides sign.
1. Start from the decision, not the model
The question. The customer says "we want a chatbot." What do you ask?
Start from nothing. Customers describe solutions ("a chatbot", "an agent", "fine-tune a model"). Your job is to find the decision or task underneath, because that's what has value and what you can measure. Five questions get you there, in this order: