Tencent's Gander AI keeps talking while working in background

Tencent released Gander, an AI system that processes speech, images, and text while simultaneously handling tasks like searching files or writing code in the background. A "cerebellum" component sustains the conversation while a swappable "brain" does the heavier task work, and users can interrupt or redirect it mid-task. In benchmarks it interrupted users unnecessarily in 8% of cases, less often than rival systems, but scored lower on task accuracy.

Why it matters: This targets a real usability gap in current assistants, which typically either block while "thinking" or ignore interruptions entirely. Splitting a fast conversational layer from a slower task-execution layer is a distinct architecture from the single-model approach most labs use, and if the accuracy tradeoff narrows, it could become a common pattern for voice agents that need to stay responsive during long-running work.