Module 10

Multi-Agent Teams

When one agent is not enough

TL;DR

When one agent is not enough

> Overview

Complex products may need multiple specialized agents working together: a researcher, a writer, a reviewer, like a team of digital employees. Each agent has a specific role and they pass work between each other through defined protocols. This module also introduces Google's Agent-to-Agent (A2A) protocol, which standardizes how agents from different systems communicate.

> Why This Matters for Your Product

Multi-agent systems are powerful but expensive and complex. Each conversation between agents burns tokens and adds latency. The PM decides whether the task complexity justifies multiple agents or if a single well-designed agent can handle it. Most products start with one agent and scale to multi-agent only when needed.

> Interactive & tools

Topologies

Agent team topologies

Hub-and-spoke

  • Coordinator delegates to specialists
  • Assembles results
  • Single point of control

Best for: Complex tasks with clear roles

Pipeline

  • Agents pass work sequentially
  • Researcher → writer → editor
  • Predictable flow

Best for: Staged workflows

Peer review

  • One agent generates, another critiques
  • Generator revises
  • Quality gate

Best for: High-quality output

Competitive

  • Multiple agents generate options
  • Judge picks best
  • Diversity of solutions

Best for: Creative or varied outputs

Cost comparison

Token cost comparison

Multi-agent total tokens: 30,000

Can be 3.0x a single-agent flow if each agent sees full context.

Related Engineering Patterns

These are the technical patterns your engineering team will implement. Understanding them helps you have better conversations.

Multi-Agent CollaborationInter-Agent Communication (A2A)

See the full decision framework

Sign up free to see this module's Key Decisions, the questions to ask your engineering team, and play the interactive Multi-Agent Teams game.

Sign Up Free