Human as an async API: Temporal shows durable human-in-the-loop agents that survive worker crashes
AI Engineer · youtube · 2026-10-05
Melanie Warrick (Developer Relations, Temporal) demonstrates how to build human-in-the-loop agents that don't block or lose state, in a talk from AI Engineer World's Fair 2026.
- Live demo: a multi-agent ice cream delivery system (fleet, customer, dispatch agents) built on Google ADK and LangGraph, including a customer changing an order mid-flight.
- Key insight: humans don't respond in 200ms, so a blocking "ask the human" call breaks in production — durability belongs in the agent harness, handled by Temporal's workers, workflows and activities.
- Two primitives — wait conditions and signals — let an agent pause for a human for minutes or weeks; she kills the worker mid-approval and recovers it with replay, losing no work.
- Also covers deciding when a human should be in the loop: cost of being wrong vs alert fatigue.
Code shown for both the ADK and LangGraph integrations, both open-sourced.
More from coding & agent
- Muse ships reliability fixes after SEVs left cron jobs and scheduled tasks unrecovered — alexandr_wang · 2026-10-05
- Formalizing long PDE and probability papers now takes just 24-48 hours, says mathematician open-sourcing Lean skills — kfountou · 2026-10-05
- Reddit dev ships 'anti brain rot' AI skills that force you to think before prompting agents — sakibbai · 2026-10-05
- Your Agent Doesn't Need an LLM for Half Its Decisions—Use a Classifier — Arindam_1729 · 2026-10-05
- 48-page handbook on harness engineering for AI agents makes the rounds — blaizedsouza · 2026-10-05
- How to design APIs that AI agents can actually use, per freeCodeCamp guide — blaizedsouza · 2026-10-05