The first agent is easy. The second one is where the architecture decisions start. Build one bigger agent that does everything, connect two specialists with Connected Agents, or stand up a full A2A topology where agents discover and call each other through the protocol? Pick wrong and the result is a fragile system nobody trusts.

This session walks through the patterns for designing multi-agent systems in Copilot Studio, when each one fits, and the anti-patterns that look clever in a diagram and fall apart in production.