Skip to slide
144 / 170

CH 15 · Deep Research Agents and How They Are Trained · 07/11

Read as one page

Why it matters

DR agents are where the harness patterns meet the training methods, and they preview where coding agents are heading: planned, multi-agent, self-verifying, and increasingly trained rather than only prompted. Even if you never train a model, knowing the vocabulary (SFT, RL, GRPO, CBR) lets you read the field and understand why an agent behaves as it does. And the architecture (MasterAgent, SubAgents, ReviewAgent) plus the fact-checking loop are directly reusable in any agent that must produce trustworthy, sourced output.

Finished this idea?

Your learning progress stays in this browser.