Flashcard: how is agentic RAG different from single-shot RAG?

Attempt it

Agentic RAG puts retrieval inside the agent loop as a tool the model can call multiple times. The model decides when to query, what to query, whether the result is sufficient, and whether to follow up with another retrieval. The shape is a ReAct loop where one of the available Actions is search or retrieve. Single-shot RAG runs retrieval exactly once, before generation: the user query is embedded, the top-K chunks are retrieved, and the model generates an answer conditioned on them. Agentic RAG handles multi-hop questions, missing-context cases, and queries that need decomposition naturally, because the agent can query, evaluate, and re-query. The cost is N retrievals plus N model calls per task instead of 1+1. Single-shot RAG is faster, cheaper, and more debuggable; agentic RAG is more accurate on hard or open-ended questions where one retrieval is not enough.

Your answer

**Agentic RAG** puts retrieval inside the agent loop. The model has access to a search or retrieve tool, and it decides when to call it, what to query, whether the result is enough, and whether to follow up with another retrieval. The shape is a ReAct loop where one of the available Actions is a retrieval call. The Thought reasons about what to look up. The Action issues the retrieval. The Observation is the retrieved context. The next Thought reads that context and either generates an answer or decides another retrieval is needed. **Single-shot RAG** is different. Retrieval runs exactly once, before generation. The user query is embedded, the vector database returns the top-K chunks, and the model generates an answer with those chunks as context. No second retrieval, no agent decision-making. The key difference is whether retrieval is a runtime decision or a fixed pre-processing step. Agentic RAG handles hard cases single-shot misses: multi-hop questions need multiple retrievals, and queries where the first retrieval misses need reformulation. The cost is more LLM and retrieval calls per query.

**Agentic RAG** puts retrieval inside the agent loop as a tool the model can call multiple times during a single task. The model decides at each step when to query, what to query, whether the returned context is sufficient, and whether to follow up with another retrieval. The shape is a ReAct loop where one of the available Actions is a search or retrieve call. **Single-shot RAG** runs retrieval exactly once before generation. The user query is embedded, the vector database returns the top-K chunks, those chunks become context, and the model generates the answer. There is no opportunity for the model to ask for more context or refine the query mid-generation. The defining difference is whether the number and content of retrievals is a runtime decision or a fixed pre-processing step. The consequence: agentic RAG handles multi-hop questions (chain several retrievals), missing-context cases (reformulate and retry), and verification cases (retrieve evidence for specific claims). Single-shot has no chance to do any of this. The cost is N retrievals plus N model calls per query instead of 1+1. Latency stacks because retrievals happen serially. Single-shot RAG remains the right choice for simple lookup questions; agentic RAG earns its cost on harder, open-ended, or multi-hop ones. Production systems like Perplexity research mode, ChatGPT browsing, and Claude with web search default to agentic for complex queries and single-shot for simple ones.

Quick reference

TL;DR

Agentic RAG puts retrieval inside the agent loop as a tool the model can call multiple times. Single-shot RAG retrieves once before generation. Agentic RAG handles multi-hop and missing-context cases at extra cost.

Memory aid

Sign in for the mnemonic that makes this stick. See Pro pricing.

Easy to grasp

Imagine answering a question with a stack of books. One way: someone hands you five pages they think are relevant, and you write your answer using only those pages. That is single-shot RAG. Another way: you can ask for any page at any time, read it, decide if you need more, ask for another, and so on, until you feel you have enough. That is agentic RAG. The first is faster because you only look once. The second is better when the answer needs you to chase down related facts, but it costs more because you keep looking. Most modern AI products like Perplexity research mode and ChatGPT browsing use agentic RAG for hard questions and fall back to single-shot for simple ones.

Key concepts

Agentic RAG RAG Multi-hop Retrieval as Tool

Detailed answer & concept explanation

~7 min read

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.

Sign in for 5 free unlocks/day or go Pro for unlimited. See Pro pricing.

Quick scan · 2–4 min

Everything important, quickly.

Define agentic RAG as retrieval as a tool inside an agent loop; single-shot RAG as one retrieval before generation. Name the runtime decision versus pre-processing step distinction. Cover the three problem classes agentic RAG wins on. State the cost reality and the production routing pattern. Close with the loop-cap requirement to avoid runaway loops.

Real-world examples

Real products, models, and research that use this idea.

Perplexity research mode runs agentic RAG: the model decides which sources to search, reads results, decides whether to search again, and iterates until it has enough evidence for the answer.
ChatGPT browsing and Claude with web search both implement agentic RAG: a web-search tool is exposed to the model, which calls it as many times as needed during the task.
LangGraph's prebuilt agentic-RAG template wraps a vector retriever as a tool inside a ReAct state graph, with conditional edges that decide when to retrieve again versus when to answer.
Cursor's codebase search inside an agent loop is a form of agentic RAG over code: the model decides what to grep or semantic-search for, reads the results, and decides whether to search again before editing.

Interviewer follow-ups

What an interviewer would ask next. Try answering before peeking at the approach.

QHow does agentic RAG handle multi-hop questions better than single-shot RAG?

Sign in or go Pro to see the approach for this follow-up.

QWhy do production agentic-RAG systems route queries between single-shot and agentic paths?

Sign in or go Pro to see the approach for this follow-up.

QHow do agentic RAG and Reflexion compose?

Sign in or go Pro to see the approach for this follow-up.

Don't say this

Red flags and common mistakes that signal junior thinking. Click to expand.

Most common mistake

Describing agentic RAG as RAG with reranking or with better embeddings. The defining feature is that retrieval is a runtime tool call the agent decides to make, not a fixed pre-processing step.

Sign in or go Pro to see all red flags and common mistakes. See Pro pricing.

60-second cheat sheet

The night-before-the-interview bullets. Scan these on the way to the call.

Sign in for 5 unlocks/day or go Pro for unlimited revision sheets. See Pro pricing.

References

Primary sources. Skim if you want the original framing.