Agentic Design Patterns Where You Are the Hero
summary
Guillaume Laforge (Google) presents concrete agent design patterns for building multi-agent systems using Java tooling (LangChain4j, ADK for Java). He explains how to combine basic components—sequential/parallel flows and loops—into powerful patterns (reflection loops, critic/reviewer duos), discusses routing and swarm coordination strategies, and explores remote agent ecosystems with MCP and A2A, including code examples and guidance on choosing abstractions to avoid technical dead ends.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 5:18Agent pattern selection
- 10:33Avoid deterministic agents
- 13:21Token savings progressive disclosure
- 18:57Demo: Progressive disclosure
- 21:18Sub-agent decomposition determinism
- 25:28LLM as judge pattern
- 30:08Demo: LLM as judge
- 34:18Goal Oriented Action Planning
- 38:57Demo: Goal Oriented Action Planning
- 40:25Continuous improvement feedback
- 42:31Demo: Coding agent feedback
- 45:20Summary
related talks
41mPatterns & Practices for building Multi-Agent Systems
49mAgentic AI Patterns
56mDesign Patterns for Multi Agent Systems
47mAgentic AI Patterns
52mChoose your own adventure in agentic design patterns
51mAgentic AI Patterns
27mEmerging Patterns Agentic AI
24mAgentic Al in SW Development: Evolving Patterns & Protocols