Google has recently released a new voice feature update for Gemini Live, aiming to allow users to invoke capabilities such as chat, Spark, and Daily Brief through natural voice commands, completing tasks such as information processing and task execution. Google stated that the upgraded Gemini app can automatically understand user needs without users having to determine which specific function should be used for a particular task.
However, this design also highlights the interaction challenges faced by current AI applications. Although Google aims to simplify operations through a voice interface, Gemini still presents chat, Spark, and Daily Brief as separate functions, each with different icons and entry points, increasing the cost for users to understand and choose.
The Daily Brief relies on data from Google's ecosystem such as Gmail and calendar to provide users with proactive personalized updates, but its information filtering capability still has shortcomings, with some reminders possibly not being what users currently need. In contrast, Spark, as an AI agent with task execution capabilities, is closer to the future direction of AI assistants, but Google's branding of it separately requires users to understand the differences between various functions.
This issue is not unique to Gemini. The AI industry currently普遍存在 the problem of directly exposing internal model architectures and functional patterns to users. For example, some AI products require users to distinguish between different modes such as chat, collaboration, and agents, while consumers prefer to propose their needs through a unified entry point, allowing the AI to automatically determine and call the appropriate capability.
Industry experts believe that the future direction of AI assistants may be to reduce the complexity of interactions, hiding complex model, agent, and tool calls in the background. Apple's strategy around Siri, as well as some AI agent products based on text interaction, emphasize completing tasks through familiar chat methods. As AI capabilities continue to improve, how to allow users to avoid learning new operation logic may become an important direction for competition in the next stage of AI products.
