Agentic RAG

Standard RAG runs one fixed search per question. Agentic RAG puts that search inside a decision loop: the model can choose whether to retrieve at all, rewrite a vague question into a sharper query, search several times across multiple steps, and judge whether what came back is good enough before it answers.

The extra control helps on genuinely multi-step questions, but it is not free. Measured against a well-tuned single-pass pipeline, an agentic loop can spend several times the tokens and time, and on simple questions it sometimes does worse by retrieving when it should have just answered. It earns its cost on hard questions rather than on every one.