Many real-world questions cannot be answered with a single prompt to a language model. They require gathering information from multiple sources, evaluating the credibility of those sources, comparing conflicting claims, and synthesizing the results into a coherent answer.

Deep research agents are designed to handle these kinds of complex information-gathering tasks.

A deep research agent goes beyond simple retrieval. It can search across documents and the web, read and analyze multiple sources, identify gaps in its knowledge, and perform additional searches to refine its understanding. Through iterative reasoning and evidence collection, the agent gradually builds a well-supported answer rather than relying on a single retrieval step.

In this chapter, we will explore how deep research agents work, including the architectures that enable multi-step search, document analysis, and evidence synthesis.