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.
Key Product Decisions
- [01]Does your use case genuinely need multiple agents or can one suffice?
- [02]What are the distinct roles/specializations needed?
- [03]Which topology (hub-spoke, pipeline, peer review, competitive) fits your workflow?
- [04]What is the cost ceiling for multi-agent interactions?
Ask Your Engineering Team
- →What is the token cost per multi-agent conversation?
- →How do we trace and debug issues across multiple agents?
- →Can we start with a single agent and upgrade to multi-agent later?
- →What orchestration framework are we using (CrewAI, LangGraph, AutoGen)?
Unlock the decision framework
Free account — no credit card required. Sign up to see the full decision checklist and the questions to ask your engineering team.
Sign Up FreePlay the interactive Multi-Agent Teams game
Practice the decisions from this module in an interactive game. Sign up free to play and save your progress.
Sign Up Free to PlaySee 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