You Have 3 Weeks to Prove AI Value. What Do You Build First?

A customer has twelve possible AI projects—but only three weeks to prove that one of them deserves further investment. What do you build first, and what evidence would convince the customer to continue? Interview question #9 in an ongoing series for Forward Deployed Engineer, AI engineer, solution architect, and applied AI interviews. This is not really a model-selection question. It tests whether you can narrow an ambiguous opportunity, choose a measurable workflow, establish a baseline, and build a pilot that supports an actual business decision. In this video: how to select the right workflow instead of the most impressive demo, why you need a baseline before building, what a thin but complete end-to-end pilot looks like, how to use each week to reduce a different uncertainty, how to respond when the customer asks for a reusable platform first, and how to define success as a go, revise, or stop decision. CHAPTERS 0:00 The Clock Is Already Running0:23 The Question Behind the Question0:37 Prove One Outcome0:50 Find the Useful Wedge1:14 Measure Before Building1:42 Build a Thin, Complete Slice2:03 The Three-Week Plan2:50 What If They Want the Platform First?3:13 Value Is a Decision, Not a Demo3:34 Build Evidence Worth Funding This is part of an ongoing series on real AI engineering interview scenarios, covering production AI, system design, evaluations, reliability, architecture trade-offs, and customer-facing engineering judgment. Subscribe for more practical AI Engineer and Forward Deployed Engineer interview breakdowns. #AIEngineering #ForwardDeployedEngineer #InterviewPrep #GenerativeAI #ProductionAI #SystemDesign #AIInterview #AppliedAI #LLMOps