From Chaos to Choreography: Multi-Agent Orchestration Patterns That Actually Work
summary
Demonstrates concrete fixes: from one agent to five the coordination work scales ~25x and a credit pipeline shows 20% incorrect risk ratings when cache invalidation fails; prescribes immutable state snapshots with versioning, explicit data contracts, circuit breakers and saga/compensation rollbacks; maps an orchestration implementation using LangGraph, Agent Bricks, Unity Catalog, Delta Lake and MLflow with append-only state, handoff validation, retry thresholds and observability that make multi-agent systems debuggable and auditable.
AI-written summary · about
chapters (14)
- 0:00Multi-agent AI systems intro
- 1:04Scaling agents in production
- 2:29Race conditions war story
- 4:17Complexity with multiple agents
- 5:43Choreography vs orchestration
- 7:50Choreography or orchestration?
- 11:24State management without races
- 15:40Data contracts enforcement
- 16:46Agent failure recovery
- 18:12Circuit breaker pattern
- 19:58Compensation pattern
- 21:01Production architecture overview
- 22:50Databricks implementation
- 24:58Final thoughts on reliability
related talks
45mAn Agent Walks Into a Bar... and Meets Another Agent: Multi-Agent Systems in Action
50mHarnessing Event-Driven and Multi-Agent Architectures for Complex Workflows in Gen...
14mOrchestrating Agentic Workflows with Governance and Trust
25mBeyond the IDE: Toward Multi-Agent Orchestration
59mOrchestration vs. Choreography: The good, the bad, and the trade-offs
42mOrchestration vs Choreography, A Guide To Composing Your Monolith
52mFrom Microservices to Agent Services Architectural Patterns for Autonomous System..
29mAgent-Ready Orchestration: Engineering the Execution Layer.