Welcome to Multi-Agent Systems, the seventh module. So far you’ve built one increasingly capable agent — it loops, uses tools, remembers, plans, and retrieves. But as you pile responsibilities onto a single agent, its prompt bloats, its tool list sprawls, and its focus slips. The fix is the same one human teams use: divide the work among specialists and coordinate them. This module teaches how to do that with agents.

You’ll start with why and when to reach for multiple agents — the real wins (specialization, focused context) and the real cost (coordination), and why you should exhaust a single agent first. Then you’ll build the three patterns that cover most real systems. Agents as tools: expose a specialist agent to a supervisor as if it were an ordinary tool, so the supervisor can delegate. Routing: classify each request and hand it to the right specialist. Orchestrator-workers: decompose a big task into subtasks, run a worker on each, and synthesize their outputs into one answer. The module ends by turning Atlas into a small team — an orchestrator coordinating a destination researcher, a budget analyst, and an itinerary writer.

Every coordination pattern here — the supervisor delegating to a sub-agent, the router dispatching to a specialist, the orchestrator fanning out to workers and synthesizing — is real, runnable Python built on the exact run_agent loop you already have, and verified end to end against an SDK-shaped mock. Start with Lesson 1 on why and when to use multiple agents.