Agentic pattern lab → https://goo.gle/agenticpattern 
Multi-agent pattern blog → https://goo.gle/multiagentpattern 
Design agentic pattern → https://goo.gle/agenticpatterndesign

Learn advanced AI agent design patterns for building dynamic, multi-agent systems. In Part 2 of this agentic AI tutorial, we use the Agent Development Kit (ADK) to build Loop, Coordinator, and Agent-as-Tool workflows.

In this AI tutorial, we cover:
* The Loop Pattern (Review & Critique): Build self correcting AI using a generator and critic agent to iteratively refine outputs until strict conditions are met.
* The Coordinator Pattern (Dynamic Routing): Create a "smart manager" that uses hierarchical task decomposition to dynamically route complex LLM requests to specialized sub-agents.
* The Agent as Tool Pattern: Learn the crucial difference between delegating tasks (Coordinator) and treating sub-agents as stateless tools to maintain full system control.

Chapters:
0:00 - Intro
1:09 - Pattern 1: The loop pattern (review and critique) 
3:08 - Pattern 2: Coordinator (Router) pattern
3:26 - Hierarchical task decomposition with LLM agent
5:33 - Pattern 3:  "Agent as tool" pattern
6:49 - Recap

More resources:
ADK Doc → https://goo.gle/4bmJMdR 
Foundations of multi-agent systems with ADK → https://goo.gle/4lw3EA7 
Workflow agents and communication in ADK → https://goo.gle/4lsAlhD 

Watch more AI agent crash course → https://goo.gle/AIforBeginners
🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech

#GoogleCloud #AIAgents #ADK

Speakers: Annie Wang
Products Mentioned: Agent Development Kit

YouTube

Covers three advanced AI agent design patterns: loop review and critique, hierarchical task decomposition with a coordinator/router agent, and agent as tool. The loop pattern uses a generator and critique agent in an iterative cycle until output meets specific conditions. The coordinator pattern acts as a smart project manager delegating subtasks to specialized sub-agents. The agent-as-tool pattern differs by keeping the primary agent in full control, treating sub-agents as stateless tools. Each pattern is demonstrated with code examples and a web UI, with trade-offs discussed around latency, cost, flexibility, and complexity.

3 Advanced AI agent design patterns