Agentic search goes beyond submitting one query and returning a ranked list. The agent can break a question into subqueries, choose sources, inspect results, follow gaps and revise the search plan as evidence accumulates.
The approach can improve coverage for research tasks but also increases the risk of compounding weak sources or wandering outside scope. Reliable systems retain citations, distinguish evidence from inference and stop when the required coverage or confidence is reached.
ELI5
Agentic search is searching performed as a multi-step investigation by an AI agent. The agent decides what to look for next based on what earlier results did or did not answer.
For example, an agent researching a product claim may search for the official announcement, inspect the documentation and then look for independent test results. It should keep the source links and avoid treating a repeated unsupported claim as stronger evidence.
