AI Agentic Workflows & Design Patterns
summary
A practical migration path from LLMs to agentic systems with concrete patterns: implement prompt chaining to split large requests into sequential classification, priority and summary steps; run independent tasks in parallel; use an orchestrator-worker pattern to dispatch dynamic workers; add an evaluator/optimizer loop with programmatic guardrails (demo uses max 3 iterations); route issues to the right domain and call tools. Demo builds a local helpdesk that runs on Gemma 4 20 12B, auto-triages tickets, detects exposed passwords and recommends forced resets.
AI-written summary · about
chapters (10)
- 0:00Introduction to AI agents
- 2:56From LLMs to agentic AI
- 6:57Prompt chaining workflow
- 15:08Parallel workflow execution
- 18:38Orchestrator workflow
- 25:16Evaluator optimizer pattern
- 36:12Prompt routing workflow
- 41:05Combined workflow demo
- 46:43Key takeaways and conclusion
- 47:56Q&A session
related talks
35mAgentic Workflows: Stories from the Engine Room
26mBuilding Agentic Workflows: Patterns for Orchestrating Intelligent Systems.
49mAgentic AI Patterns
22mAgentic AI for Engineering Leaders | Strategy & Execution Guide
16mEffective agent design patterns in production
27mEmerging Patterns Agentic AI
24mAgentic Al in SW Development: Evolving Patterns & Protocols
14mOrchestrating Agentic Workflows with Governance and Trust