Building the Voice OS at Willow

You'll leave this session knowing what it takes to build toward a real voice operating system, not just another transcription tool. Willow's team will share how they think about the shift from short, isolated voice commands to a system that understands how each person communicates, learns from their edits, and acts across every application. You'll hear about the advances in speech recognition, low-latency language models, synthetic data, and personalization that have gotten them this far, including how they train and iterate on these models using Serverless RL instead of managing their own training infrastructure. Willow will share what's working, where the broader voice ecosystem still falls short, and the gaps, including continuous learning, personal context, cross-application action, privacy, and trust, that stand between today's voice products and a true voice OS.