Airbnb's agent harness, Insight Miner, transforms unstructured data exploration by encoding scientific methodology into scalable and reproducible infrastructure. Traditional AI responses often lack the underlying investigative methods, making it difficult to trust or reproduce findings. Insight Miner addresses this by governing an AI agent's operations, from question framing to evidence selection and decision recording, ensuring auditability and shared methodology. This system was crucial for Airbnb's AI customer service assistant launch, enabling rapid analysis of diverse customer situations across languages and geographies. It shifts data scientists' focus from execution to refining analytical techniques, increasing human judgment in strategic data inspection. Insight Miner is now used by dozens of teams beyond data science, including operations and product insights, allowing subject matter experts to conduct scaled analyses without coding. This new category of full-stack infrastructure integrates proprietary data, deep workflows, and expert "taste," where methodology itself becomes a cumulative product. Such harnesses are vital for adopting AI-first knowledge work in various fields, continuously improving with every use.