Chinese AI Startups Pivot Toward Agent-Centric Mobile Experiences
Zhipu AI's AutoGLM aims to abstract the smartphone interface, allowing AI agents to operate apps via voice commands.
The traditional relationship between smartphone users and their applications is facing a fundamental shift in China. AI startups are developing "mobile agents" designed to operate device interfaces on behalf of the user, potentially rendering manual app navigation obsolete.
Zhipu AI has launched AutoGLM, a mobile agent capable of simulating human operations on smartphones to complete a variety of tasks. By utilizing multimodal perception, the agent can understand what is appearing on a screen and execute complex actions—such as ordering coffee or sending messages—through simple voice commands. Rather than the user clicking through menus, the agent interacts with the phone's UI directly to achieve the requested outcome.
The Evolution of the Mobile Ecosystem
This transition arrives in a Chinese mobile landscape that is already distinct from Western markets. Due to the absence of Google Play and the pervasive dominance of "super-apps" like WeChat, the region has long been accustomed to centralized digital hubs. The move toward AI agents represents the next logical evolution of this trend, aiming to abstract the app layer entirely. Instead of a user managing a dozen different interfaces, the agent serves as a single, intelligent layer that mediates all interactions with the device's software.
Shifting Industry Economics
If AI agents successfully replace the need for direct human interaction with app UIs, the economic incentives for software developers will change. For years, the mobile industry has prioritized "attention-grabbing" interfaces designed to keep users engaged within a specific app. In an agent-centric world, value shifts toward "agent-accessible" APIs and seamless backend integration. The competitive advantage will no longer be based on who has the most intuitive visual design, but on whose service is most easily discoverable and executable by an AI agent.
The Path Forward
As these agents evolve, the industry is watching whether this shift will extend beyond the smartphone. According to Yicai Global, Zhipu AI has packaged AutoGLM's capabilities into an API, which could allow other wearable hardware, such as smartwatches and glasses, to function as intelligent assistants. While Zhipu AI markets the tool as a pioneering effort in the space, the broader challenge remains whether these agents can maintain reliability and security across the diverse and fragmented ecosystem of third-party applications.