Multi-agent systems are powerful, but without structured orchestration and orientation they become unpredictable and impossible to control. This session shows how to build reliable, governable multi-agent workflows using Azure AI Foundry and the Microsoft Agent Framework. We break down a real scenario into coordinated agent roles – researcher, planner, critic, executor, safety-checker – each with clear autonomy boundaries and escalation rules, along with MCP tools and human-in-the-loop protocols. Foundry’s native multi-agent layer replaces custom routing logic with explicit schemas, deterministic interactions, and policy-driven constraints that keep behaviour stable and observable.