Multi-agent systems use several agents for large workloads, complex tasks, or diverse expertise. Benefits: specialization, scalability, and fault tolerance. You need agent communication, coordination, and visibility (logging, dashboards). Common patterns: group chat (many agents in one conversation), hand-off (one agent passes the task to another), and collaborative (multiple experts contribute to one recommendation). Human-in-the-loop is often required for approval steps.
Multi-agent conversation flow (e.g. trip booking)
Splits task, calls sub-agents
Multi-agent patterns
Group chat
Multiple agents in one conversation; messages go to the group. Good for collaboration and support.
Hand-off
One agent passes the task to another by predefined rules. Good for workflows (e.g. triage → specialist).
Collaborative filtering
Several agents with different expertise contribute to one recommendation (e.g. industry + technical + fundamental).
Example: Refund process
Customer agent starts the refund; seller, payment, and resolution agents handle their parts. General agents (shipping, notification, analytics) can be reused across processes. Each agent has a clear role; they communicate via a shared protocol and optionally a coordinator.