Agentic RAG puts an AI agent in charge of retrieval, so the model decides when to search, what to look up and whether to search again before answering.
In plain words
Plain RAG follows a fixed recipe: search once, add the results to the prompt, answer. That works for simple questions but struggles when an answer needs several steps. In agentic RAG, an agent treats search as a tool. It can plan, reflect on what it found, rewrite its query, call other tools, or skip searching when it already knows enough.
A quick example
Imagine asking a company assistant which of last year's product launches had the most support tickets. An agentic RAG system might first search for the list of launches, then run a separate search for tickets about each one, notice a gap, search again, and only then write its answer.
Where you'll meet it
- Assistants that answer multi-step questions over company documents.
- Research tools that combine several search sources.
Related
Sources
- PaperAgentic Retrieval-Augmented Generation: A Survey on Agentic RAG, Singh et al. (arXiv) · read 29 Sept 2026
- DocsBuild a custom RAG agent with LangGraph, LangChain documentation · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.