BQ — GenAI Usage (How / Failure / Tradeoff)

A GenAI-themed BQ appears in nearly every Amazon round in 2026. Standard prompts: 'tell me about a time you used GenAI', 'a GenAI failure', or 'how do you use AI day-to-day'. Some technical rounds also probe Agent / multi-turn design depth. Requirements Prepare 2-3 concrete GenAI stories, each in STAR form: situation, task, action, result. At least one success, at least one failure or tradeoff. Be ready to defend the choice of model (Claude vs GPT vs in-house), the prompting pattern (chain-of-thought, retrieval, agent), and a quantified outcome. For Applied Scientist / MLE candidates, also prep stories that fold in agent design — multi-turn, context overflow, tool-use. Examples Verbatim prompts collected from recent loops: "Tell me about a time you used GenAI." "Tell me about a GenAI failure." "Describe a time AI gave you a wrong answer — how did you catch it and what did you do?" "How do you use AI in your day-to-day?" "How would you design a multi-turn agent? What happens when context overflows?" "How would you architect an AI chatbot?" "Where would you use GenAI, and where would you not use it?" "When AI output differs from what you expected, how do you adjust it?" "Talk about a time when you had to learn about a new generative AI capability." Notes Interviewers follow up aggressively. Expect at least 2 levels of 'why' and 'what would you do differently'. Vague 'I used ChatGPT to summarize' stories score poorly — bring measurable impact (latency, accuracy, time saved). AS / MLE candidates should weave in agent / multi-turn / context-management vocabulary; SDE candidates can stay at the application-layer. Each interviewer typically owns 1-3 Leadership Principles and pushes 2-3 BQ prompts in the slot; expect 10-15 minutes of recursive follow-ups on a single story rather than rapid topic switching. AI tools are not permitted inside the live interview itself — the GenAI BQ slot is asking about your past usage, not an invitation to demo a tool on screen. For AS / MLE rounds, name the agent design vocabulary explicitly (tool-use loop, context window management, eval harness) — generic "I prompted the model" answers consistently underperform. Hiring-manager rounds increasingly open with an AI-literacy segment before the technical portion — expect the usage question even in rounds billed as OOD or coding. Preparation Write 3 GenAI stories in STAR form: success, failure, tradeoff. Memorize the metric each one closes on. Practice the 'how would you do differently' beat for each story — Amazon LPs reward iteration mindset. For technical stories, prep at least one Agent or multi-turn example with explicit context-management discussion.