Agentic AI architecture patterns
summary
Argues that agents replace symbolic rule cortices with LLMs and that an agent is perceive‑reason‑act plus short‑ and long‑term memory, then maps concrete patterns to distributed systems: prompt chaining (linear LLM relay), agent routing (LLM classifier dispatch), parallelization/scatter‑gather, and orchestration/step‑function supervisor. Implements examples using Amazon Bedrock, MCP, vector stores, Lambda, Step Functions and EventBridge for RAG, tool‑execution, computer‑use, coding, voice and workflow agents. Provides architecture templates and an open‑source SDK so these Bedrock‑based patterns are reproducible in real systems.
AI-written summary · about
chapters (13)
- 0:00Welcome and introduction
- 1:05What are software agents
- 3:34History of agentic AI
- 7:27Building blocks of agents
- 10:40Common agent patterns
- 17:42LLM workflow patterns
- 21:37Agent workflow patterns overview
- 23:22Prompt chaining pattern
- 24:25Routing and delegation pattern
- 26:36Parallelization pattern
- 28:41Orchestration pattern
- 30:28Implementation and tools
- 31:09Q&A and closing
related talks
15mServerless & Agentic AI: Better Together
22mAgentic AI for Engineering Leaders | Strategy & Execution Guide
27mEmerging Patterns Agentic AI
44mBuilding Agentic AI Systems: From Transactions To Conversations
49mAgentic AI Patterns
29mAgentic AI: What It Is and Why It Matters
46mDemystifying AI Agents
40mAgents in Action: Advancing Open Source AI Missions