AI Agents path

🤖 Using AI & AI Agents

Chapter 14 of 24

👥 Chapter 14: Multi-Agent Patterns

When to use multiple agents; group chat, hand-off, collaboration; visibility

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)

User: "Book a trip to Honolulu — flight, hotel, car"
Orchestrator agent (Travel Agent)

Splits task, calls sub-agents

Flight agent
Hotel agent
Car agent
↓ each returns result
Orchestrator combines: flight details + hotel confirmation + car booking
User sees one reply with full itinerary

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.