The launch of Gemini 3.8 Live highlights a broader trend: AI companies want models to become real-time voice partners rather than systems waiting for a text message.
The important test will be less about demo quality and more about reliable operation in real tasks, different languages and environments with noise and interruptions.