Latency Is a Budget. Humanlike Is the Goal. — Jesse Hall, LiveKit

Why watch

Jesse Hall from LiveKit explains why voice agents feel robotic despite good dashboard metrics, covering the gap between measured and perceived latency, how pipeline errors compound differently than model errors, and audio-based turn detection techniques. He demonstrates a hotel receptionist demo handling interruptions and provides an open-source benchmark for evaluating voice agents, targeting developers building conversational AI systems.