Skip to slide
Chapter 15 · Deep Research Agents and How They Are Trained
117 / 142

CHAPTER 15 · Deep Research Agents and How They Are Trained · 7 / 10

How it works: a research loop with planning, retrieval, and a review pass

The MVP is a dynamic single-agent research loop: plan subtasks, run each as a (simulated) retrieval, then a review pass that cross-checks claims against sources before producing a cited answer. It reuses the orchestration shape from Chapter 11 and the fact-checking idea above.

← → arrow keys work too