Melanie Warrick demonstrates an ice cream delivery simulation with customer, fleet and dispatch agentsA multi-agent system contains multiple AI agents that interact, coordinate, divide work, or influence one another while pursuing tasks.. A human can approve a changeHuman-in-the-loop command approval requires a person to review and authorize a consequential AI action before it executes. to one order while other work continues.
She explains Temporal’s worker, workflow and activity primitives and describes how wait conditions and signals can preserve a pending human decision without blocking the entire system.
Using Google ADK and LangGraph integrations, Warrick shows a worker going offline and resuming from event historyDurable workflow execution preserves progress so a multi-step process can resume after interruptions using recorded completion state.. She advises weighing the cost of errors against the risk of human alert fatigueRisk assessment identifies potential harmful outcomes and examines their likelihood, impact and uncertainty in a specified context. It provides evidence for deciding how a system should be managed. when choosing approval points.
Watch on YouTube
