Chinese smartphone makers have begun rolling out mobile-based personal AI agents — a trend quickly embraced by Huawei, Xiaomi and Honor as they race to differentiate their platforms with on-device intelligence. Huawei this week introduced a HarmonyOS-based beta called "Xiao Yi Claw," presenting features such as document editing, PPT writing and automated email replies, together with multi-device collaboration and configurable personalities. Huawei's developer materials also describe an OpenClaw mode that lets developers expose 24/7 personal agents through the Xiao Yi app, with the company stressing that the agent core runs on HarmonyOS and can coordinate tasks across phones, tablets and PCs.
Honor followed with a branded "Lobster Universe" that promises one-click "raising" of virtual lobsters on PCs and tablets and interaction through its assistant YOYO, while Xiaomi was first to market with a limited beta of "Xiaomi miclaw," an agent built on the firm's MiMo large model. Executives and marketing have leaned into playful metaphors for these agents — often called "raising lobsters" in Chinese tech slang — framing them as always-on, personalised assistants that live on users' devices. Xiaomi founder Lei Jun even reposted material referring to miclaw as a "phone lobster," underscoring how vendors are using colloquial language to make the technology feel friendly and domestic.
The technical pitch from vendors highlights two selling points: lower latency and greater privacy by keeping AI processing at the edge, and deeper integration with a manufacturer's device ecosystem. Huawei, in particular, emphasises cross-device execution — for example, issuing a command on a phone that is carried out on a PC — by tying the agent to HarmonyOS capabilities. Honor and Xiaomi likewise stress ecosystem linkages, but for now those linkages appear to be largely limited to each company's own services and apps rather than broad third-party integration.
Industry reaction has been mixed. Some observers see on-device agents as a logical next step for personalisation and offline utility, especially as mobile platforms seek ways to remain competitive without depending solely on cloud-hosted models. Others doubt the near-term usefulness of device agents where they cannot seamlessly access third-party apps and services, pointing to API-authorisation and ecosystem friction as the principal obstacles. An IDC China analyst warned that unless developers and app makers explicitly open integrations, agents will be confined to a vendor's privileged apps and will struggle to deliver the cross-app workflows that make virtual assistants genuinely helpful.
That tension — between a vendor-controlled, privacy-promising on-device assistant and the need for broad app-level interoperability — is at the heart of the business challenge. Phone-makers control the hardware, operating system and some core services, but many of the functions users expect from an assistant depend on third-party apps and cloud services. For now Xiaomi's miclaw reportedly supports only Xiaomi's own ecosystem; pushing beyond that will require negotiated API access or new platform-level standards that third parties accept.
The broader industrial context matters. China’s device makers are pushing edge AI partly because geopolitical constraints and export controls make reliance on foreign cloud infrastructure and advanced chips more complicated. Designing efficient on-device models that can run on local silicon is a way to preserve functionality while also locking users into a domestic stack. But doing so successfully will require balancing performance, developer incentives, and regulatory scrutiny over data exposure when personal agents invoke device capabilities on behalf of users.
For consumers and developers the rollout will look incremental. Early betas let vendors show off capabilities and gather feedback, but widespread adoption will hinge on tangible productivity gains and seamless cross-app workflows. Phone-makers can create convenience within their own app families; turning that convenience into broad, genuinely useful assistants will depend on whether they can convince the wider app ecosystem to cooperate or build new incentives for integration.
In short, the opening salvos in China's "phone lobster" wave show ambition and platform strategy more than immediate mass-market utility. The technology matters because it signals how the next phase of mobile AI may be structured — device-first, ecosystem-anchored and shaped by vendor incentives — but the path from experimental features to indispensable personal agents remains uncertain.