Gemini 3.8 Live adds avatars, background tools

Google is pushing real-time agents beyond voice latency into persistent, visually embodied conversations—but enterprise controls and uneven speech quality will decide whether this is production-ready.

• –Background tool execution keeps conversations alive while APIs run, reducing awkward silence and enabling more natural agent workflows.
• –Native 24 FPS lip-synced avatars remove the need for a separate rendering stack, but add visual compute, cost, and UX complexity.
• –Camera and screen understanding unlocks practical use cases such as claims intake, tutoring, support, and guided shopping.
• –Dynamic language switching and custom vocabulary make the model more useful for global and domain-specific deployments.
• –Early practitioner feedback suggests latency is impressive, but robotic TTS remains a serious limitation for phone-based voice agents.