Onboarding answer agent
New engineers at your company ask the same questions for months because the answers are buried in internal docs — many of them stale. Design an onboarding answer agent over that doc corpus. How does it handle documentation it can't trust, when does it say 'I don't know', and what does it never answer from memory? Describe how you'd evaluate answer quality before letting new hires depend on it, then give me the pitch you'd bring to engineering leadership.
decide_doc_answer_action(topic: str, doc_ages_days: list[int], sources_conflict: bool) → str["how do i deploy the web app",[12,400],false]out"answer"["where are the production credentials stored",[3],false]out"route_to_human"["what is our code review policy",[500,900],false]out"answer_with_stale_warning"Turn the fuzzy goal into a bounded agent task. Say what the agent reads, what it may do on its own versus draft for a human, and what it must never touch. Then earn the trust: how you’d verify it before rollout, and the pitch (the problem, what you built, why it matters).
Agent build: design the agent in plain language (or narrate it). The coach grades the decomposition, the guardrails, the verification plan, and the pitch.