A practical, engineer-level read on how AI learned to search the enterprise. It traces the path from keyword search to dense retrieval, then breaks down the three architectures in play today, baseline RAG, Microsoft's GraphRAG, and the LLM-maintained markdown wiki, with the mechanics, the honest pros and cons, and the tools for each. It closes on a decision frame for picking and combining them and on the question that outlasts every technique: where your answer comes from, whether you can prove it, and under whose law the documents behind it sit.