Day 10. Agent loops drift, retry badly, and lose context. Step Functions is AWS's answer to "make a workflow durable." Worth comparing it against framework-level orchestration like LangGraph.
Topics to cover:
- Step Functions basics — state machines, ASL, Standard vs Express
- Error handling and retries — Catch / Retry, exponential backoff, custom error types
- Parallel and Map states — fan-out for batch agent work
- Agent loops as state machines — when to use Step Functions vs LangGraph vs a plain loop
- Human-in-the-loop steps — Wait for token, callback patterns, approvals
Plan: Chandana on the Step Functions native side (ASL, retries, parallel), me on agent-loop orchestration (LangGraph parallels, HITL patterns), shipping one workflow that uses both.
Day 10. Agent loops drift, retry badly, and lose context. Step Functions is AWS's answer to "make a workflow durable." Worth comparing it against framework-level orchestration like LangGraph.
Topics to cover:
Plan: Chandana on the Step Functions native side (ASL, retries, parallel), me on agent-loop orchestration (LangGraph parallels, HITL patterns), shipping one workflow that uses both.