What Is Actually Changing With Xiaomi's miclaw: Phones Shift Who Is in Charge

On March 6, 2026, Xiaomi, the Chinese phone and electronics maker, announced Xiaomi miclaw, which it describes as a test product for AI interaction on mobile. It is built on Xiaomi's own MiMo model. A small, invitation-only closed beta opened the same day, and for now it supports only the Xiaomi 17 series. The product's core capabilities span three levels: the system layer (more than 50 system-level tools packaged), understanding of personal context, and ecosystem connections. It can call first-party apps and Mijia IoT devices, and it runs on a reasoning and execution engine that decides the order of tool calls by itself. Because the product is at an exploratory stage, stability, power consumption, and the success rate on complex tasks are still being tuned. In its official Q&A, Xiaomi said clearly that user interaction data is not used for model training, that this is a small-scale test, and that it does not represent a mass-production form. That Xiaomi is following into mobile agents is no surprise, since it has the model, the devices, and the ecosystem. How miclaw is framed, though, raises some points worth thinking through. Xiaomi understands what agents are for Completing complex tasks across apps means taking down the capability walls between application systems. The mobile ecosystem has been divided by app. Data is isolated between apps and so are capabilities, and each app solves only its own use case. When a user wants to finish something that crosses several apps, all the coordination falls on the user: search, switch, copy and paste, then act.