← All drills

Multi-Agent Patterns

What you'll be able to do

Pick how several agents should be organised before you build them, and know the failure each arrangement brings - the decision that comes before every multi-agent system.

⌁ Support-automation and research agents built with LangGraph, Google ADK, AWS Strands Agents and AutoGen, and the planner-and-workers designs inside coding agents.
Start this internship
Create an account to unlock the 9 sections, the workbench, and askThili.
Begin

Sections

1. Multi-Agent Patterns: Orchestrator, Supervisor, Hierarchy, Pipeline, Fan-Out, MapReduce, Debate
🔒 locked
2. Lesson 1 - Orchestrator-Subagent: plan once, delegate to specialists
🔒 locked
3. Lesson 2 - Supervisor: route, watch, check, loop back
🔒 locked
4. Lesson 3 - Hierarchical: supervisors of supervisors
🔒 locked
5. Lesson 4 - Sequential Pipeline: fixed stages, validated at every hop
🔒 locked
6. Lesson 5 - Parallel Fan-Out / Fan-In: independent checks at once
🔒 locked
7. Lesson 6 - MapReduce: the same worker over many chunks, then combine
🔒 locked
8. Lesson 7 - Debate: two sides argue, one judge decides
🔒 locked
9. Lesson 8 - Choosing: which pattern for which job
🔒 locked

Dig deeper

🔗Building Effective Agents - Anthropic Engineering
article
📄AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation (Wu et al., 2023, arXiv:2308.08155)
paper
📄MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework (Hong et al., 2023, arXiv:2308.00352)
paper
📄Improving Factuality and Reasoning in Language Models through Multiagent Debate (Du et al., 2023, arXiv:2305.14325)
paper
📄MapReduce: Simplified Data Processing on Large Clusters (Dean and Ghemawat, OSDI 2004)
paper
🔗Groq - free API keys for open models
docs