Chapter 16: Multi-Agent Systems Architecture¶

Design systems of cooperating AI agents: roles, message-passing protocols, contract-net task allocation, consensus, and the orchestration patterns and failure modes behind production multi-agent applications.

Metadata¶

Learning Objectives¶

• Model agents with explicit roles, tools, memory, and a perceive-decide-act loop
• Pass messages with a typed performative protocol (inform / request / propose / accept)
• Allocate tasks with the Contract-Net Protocol and reason about greedy vs optimal assignment
• Share state safely through a blackboard and understand read/write contention
• Reach consensus across agents with gossip averaging and majority voting
• Orchestrate workflows with supervisor, pipeline, and debate topologies
• Diagnose failure modes — deadlock, infinite delegation loops, and runaway token cost
• Evaluate a multi-agent system on task success, cost, and latency, not vibes

What's Included¶

Notebooks¶

Scripts¶

• config.py — Chapter config: seeds, agent roster, topology presets
• agents.py — Agent dataclass, Message/Blackboard, Contract-Net allocation, gossip consensus
• orchestration.py — Supervisor / pipeline / debate orchestrators with cost + latency accounting

Exercises¶

• Problem Set 1: Agents & Allocation — Build an agent loop, route messages by performative, and run a Contract-Net auction
• Problem Set 2: Consensus & Orchestration — Prove gossip convergence, implement weighted voting, and add a loop guard to an orchestrator
• Solutions — in exercises/solutions/ (notebooks and solutions.py for CI)

Diagrams (Mermaid)¶

• agent_loop.mermaid, contract_net.mermaid, orchestration_topologies.mermaid

Read Online¶

• 16.1 Introduction — Agent model, perceive-decide-act loop, typed messages, blackboards
• 16.2 Intermediate — Contract-Net task allocation, gossip consensus, majority voting
• 16.3 Advanced — Supervisor / pipeline / debate topologies, failure modes, cost-aware evaluation

Or try the code in the Playground.

How to Use This Chapter¶

git clone https://github.com/luigipascal/berta-chapters.git
cd berta-chapters/chapters/chapter-16-multi-agent-systems-architecture
pip install -r requirements.txt
jupyter notebook notebooks/01_agents_and_messaging.ipynb

Created by Luigi Pascal Rondanini | Generated by Berta AI