0 views
Most voice-controlled browser agents feel broken because every click or page load blocks the voice stream.
In this episode, we build a live voice agent that controls a real browser, clicking, typing, and navigating, without killing the conversation. The secret: execute browser actions asynchronously in the background, then feed the next screenshot back to the model so it sees what happened.
Speaker: Annie Wang
Products Mentioned: Agent Development Kit, Gemini Live API
Date: October 12, 2026
