2.3. Orchestrating Multiple Agents
💡 First Principle: Multi-agent orchestration is specialization plus routing, not duplication. You build several agents each with focused instructions and tools, then connect them so work flows to the right specialist — via a coordinator, handoffs, or connected-agent calls. The win is that a narrow agent is more reliable than one trying to do everything.
Why care: real systems outgrow a single agent. A triage assistant that also writes code, queries databases, and drafts legal language will do all of them mediocrely and choose tools poorly. Splitting into specialists — each with a tight toolset — improves accuracy and makes the system maintainable. The exam tests recognizing when to split and how the pieces communicate.
⚠️ Common Misconception: "Multi-agent means many copies of the same agent running in parallel." It means different agents with distinct roles. Parallel copies of one agent is just scaling; multi-agent orchestration is division of labor among specialists.
