Agentic RAG overview

Agentic RAG as a cross-cutting topic — agents that actively reason about what to retrieve, iterate on retrieval results, and update knowledge rather than passively returning the top-k nearest vectors. Covers query planning (decompose complex questions into sub-queries), self-reflective retrieval (evaluate if retrieved context is sufficient, re-query if not), tool-augmented retrieval (agent uses calculator, code execution, or API calls alongside vector search), corrective RAG (detect hallucination, re-retrieve), and the shift from "retrieve then generate" to "reason, retrieve, reason, retrieve" loops that characterize agentic knowledge work.