Agentic RAG

Agentic RAG turns search from an automatic preparatory step into a tool managed by the agent in a ReAct cycle: think → search → assess sufficiency → refine query and search again. For simple questions with a clear need, classic RAG is faster and no worse; for complex multi-step questions, agentic RAG is noticeably better — it breaks down the question, searches iteratively, and cross-checks. Safety boundaries: found documents are a channel for indirect prompt injection; found content should not directly trigger risky operations.

Related: RAG, [ReAct cycle], [Prompt injection]