Not every AI problem should be solved with one model and one prompt. Complex applications increasingly use specialized agents for planning, research, tool execution, validation, and synthesis. This session explores how to design multi-agent systems using Microsoft Foundry and modern agentic architecture patterns. We will examine agent responsibilities, orchestration, context sharing, tool access, handoffs, failure recovery, evaluation, and human-in-the-loop controls. Through an end-to-end example, attendees will learn when multiple agents improve a system, when they create unnecessary complexity, and how to design agent workflows that remain observable, testable, secure, and maintainable.