An AI agent that answers a research question by planning its own searches, reading many sources and writing up a referenced report, usually over several minutes rather than seconds. Instead of answering from memory, it works more like a junior researcher: it decides what to look up, follows leads and gathers evidence. The output can save hours, but it can still misread or misquote its sources, so the references need checking.
Technical view
Google introduced a feature under this name in December 2024, and several other developers followed with similar products. The pattern combines a planner that drafts a multi-step research plan, iterative search and browsing with queries rewritten in light of what has been found, reading and note-taking across many documents, and a final synthesis with citations. It is agentic retrieval run over a long horizon with a large context budget. Failure modes include citations that do not support the claim attached to them, over-weighting low-quality pages that rank well, and confident synthesis over thin evidence. Evaluation is hard because there is rarely a single right report.
Related
Part of AI Terminology Index, section: Agents and Tools.