Agentic search
A search pattern where an agent — not a single retrieval call — runs the query: it plans, queries multiple sources, evaluates results, refines, and returns a synthesized answer with citations.
Classic search returns 10 blue links. RAG returns one synthesized paragraph from one retrieval. Agentic search is a multi-step loop: decompose the question, query multiple indexes or web sources, judge each result, reformulate, re-query, and finally synthesize a cited answer.
It is the production form of "deep research" features. Perplexity Labs, Gemini Deep Research, OpenAI Deep Research, and ChatGPT's search all use agentic search under the hood for hard questions. Latency is 10–60 seconds versus a fraction of a second for classic search.
For internal docs and B2B SaaS, agentic search is the next step beyond RAG. The cost is much higher per query, but answer quality on multi-hop or ambiguous questions is materially better.
Frequently asked
Agentic search vs. RAG — which do I need?+
RAG for fast, single-shot questions ("what does our return policy say"). Agentic search for hard, multi-hop questions ("compare our return policy to our top 3 competitors and flag risks"). Most production apps use both behind one search box.
How slow is agentic search?+
10–60 seconds typical, sometimes 2–3 minutes for deep research. Show progress UI; users will wait if they can see the agent working.
