AI voice agents are moving from demos to real mobile apps. But building one that feels natural on Flutter is very different from adding a chatbot UI. A production-ready voice agent needs to handle microphone capture, real-time audio streaming, low-latency responses, interruptions, live transcripts, network changes, background/foreground app behavior, and fallback states when the AI pipeline is slow or unavailable. This session breaks down the mobile architecture behind real-time AI voice agents using WebRTC and modern mobile app patterns. We will look at how audio moves from the device to the AI pipeline and back, how to manage conversation states in the app, how to design interruption-friendly UX, and what developers should monitor when voice quality or latency breaks down. Attendees will learn: - How AI voice agent architecture differs from traditional chatbot flows - How WebRTC enables low-latency audio communication for mobile AI agents - How to manage voice agent states such as listening, thinking, speaking, interrupted, and fallback - How to design mobile UX for real-time voice conversations - What production issues to watch for around latency, network changes, permissions, and audio quality