Adaptive agentic RAG

This architecture adjusts its strategy based on the difficulty of the query. It uses a classifier to assess query complexity and then chooses the appropriate path: no retrieval at all for a trivial query, single-step retrieval for a moderate one, or multi-step reasoning for a complex one.

The value is efficiency through matching effort to need. A simple factual question does not trigger an expensive multi-hop process, while a complex question gets the full treatment. This is the same idea as effort scaling from the multi-agent topic, applied at the level of retrieval strategy. The tradeoff the lessons chapter notes is that adaptive systems trade some correctness guarantees for responsiveness.