Week 5: Multi-Agent Architectures
Communication topologies, AutoGen, MetaGPT, ChatDev
Week 5 of 12
Learning Objectives
- Implement message passing between agents
- Build coordination patterns
- Design multi-agent workflows
Topics Covered
- Multi-agent communication patterns
- AutoGen framework
- MetaGPT software development
- Role-based agent teams
Resources
Jupyter Notebooks
Required Readings
| Paper | Authors | Year | Link |
|---|---|---|---|
| AutoGen: Enabling Next-Gen LLM Applications | Wu et al. | 2023 | arXiv |
| MetaGPT: Meta Programming for Multi-Agent Collaboration | Hong et al. | 2023 | arXiv |
Reading Guide: Multi-Agent Systems Survey
Comprehensive survey of LLM-based multi-agent collaboration
Primary Paper
A Systematic Survey on Large Language Model-based Multi-Agent Collaboration
Tran, et al. (2025)
arXiv arXiv
Tran, et al. (2025)
arXiv arXiv
Exercise: Multi-Agent Design
Design and implement multi-agent coordination
Learning Objectives
- Create: Design multi-agent workflows
- Apply: Implement message passing between agents
- Analyze: Compare coordination patterns
Tasks
| Task | Points | Description |
|---|---|---|
| Agent Team Design | 30 | Design roles for a multi-agent team |
| Message Passing | 40 | Implement agent communication |
| Coordination Analysis | 30 | Measure coordination overhead |
Exercise
Build a software development team simulation with:
- Product Manager agent
- Architect agent
- Developer agent
- Reviewer agent
- Coordinated workflow
Discussion Questions
- When is multi-agent better than single-agent?
- How do you handle conflicting agent outputs?
- What coordination patterns work best for different tasks?
Additional Resources
Discussion & Questions
Join the Conversation
Have questions about this week's material? Want to discuss concepts with fellow students?