This Is What Voice Coding Was Supposed to Be (Qwen Audio Agent)

Qwen Audio Agent takes a different approach to voice coding: instead of acting like simple dictation, it works as a full-duplex voice runtime that can keep talking to you while Claude Code, Codex, OpenCode, or another coding agent handles real work in the background. In this video, I test how the runtime actually works, including barge-in, live status updates, async coding tasks, ACP handoffs, and the moment a finished task gets injected back into the conversation without killing the voice session. 🔗 Relevant Links Audio Agent Repo - https://github.com/QwenAudio/qwen-audio-agent ❤️ More about us Radically better observability stack: https://betterstack.com/ Written tutorials: https://betterstack.com/community/ Example projects: https://github.com/BetterStackHQ 📱 Socials Twitter: https://twitter.com/betterstackhq Instagram: https://www.instagram.com/betterstackhq/ TikTok: https://www.tiktok.com/@betterstack LinkedIn: https://www.linkedin.com/company/betterstack 📌 Chapters: 0:00 Why Voice Coding Still Feels Broken 0:42 Qwen Audio Agent Explained 1:29 Live Demo: Claude Code While You Keep Talking 4:58 How Qwen Audio Agent Works With ACP 5:44 Barge-In and Full-Duplex Voice Explained 6:18 OpenAI Realtime vs Pipecat vs LiveKit Agents 6:52 Open Source, DashScope and Local MLX Mode 7:30 Qwen Audio Agent Limitations and Latency 8:33 Who Should Actually Use This? 9:08 The Real Future of Voice Coding