Author: Shenwan Tencent News

Just as OpenClaw became a top player in the AI field with the frenzy of "shrimp farming" and the controversy of "shrimp killing," leading smartphone manufacturers deeply researching edge AI have also been unable to contain themselves, jumping into the fray to deploy and "tame" their own Claw.

On March 6, Xiaomi's mobile Agent—Xiaomi miclaw officially began a limited internal beta test through an invitation code system, becoming the first domestic smartphone manufacturer to test "lobster." After that, Huawei, Honor, OPPO, and others announced information about their own Claw beta testing.

Among them, Huawei announced that its Xiao Yi added an OpenClaw mode and then released the Xiao Yi Claw Beta version; Honor announced the launch of the "Honor Lobster Universe," supporting one-click shrimp farming on PC and tablets, and in the future allowing compatibility with other ecological device connections; OPPO's ColorOS design director Chen Xi showcased some functions of the Xiao Bu Claw on social media and stated, "Xiao Bu Claw still has issues such as security that need to be resolved."

In other words, smartphone manufacturers' foray into "shrimp farming" is currently mainly in the internal testing phase, with no clear timeline for large-scale launch yet.

For example, Xiaomi's miclaw is currently limited to a closed internal test for the Xiaomi 17 series, Xiaomi 15S Pro, and Redmi K90 series. By receiving an invitation code and updating the system, users can access the Xiaomi miclaw app. "During the beta test, there are currently no charging plans," stated Lu Weibing, Xiaomi Group partner and president.

Regarding smartphone manufacturers deploying mobile "lobster," industry insiders revealed, "OpenClaw is essentially an open-source framework, containing third-party Skills and plugin ecosystems, and can also call various large models. For ordinary users, the threshold to deploy OpenClaw is high, but for smartphone manufacturers, there is no technical difficulty; the real challenges are issues like permission access, user information security assurance, and compliance with laws and regulations."

“Mainstream smartphone manufacturers are dealing with hundreds of millions of ordinary users, and any AI feature must undergo thorough verification to ensure the experience is mature, secure, and stable before it can be pushed out,” said an employee from a smartphone manufacturer.

Smartphone Manufacturers Gather to "Farm Shrimp"

Large model manufacturers are keen on deploying "lobster," which can be simply understood as a business of "monetizing computing power," letting the agent (Agent) call the model more frequently and perform complex tasks, thus consuming more Tokens and directly boosting API revenue.

However, this logic is difficult to implement in the smartphone industry. After spending thousands or even tens of thousands on devices, users are rarely willing to pay extra for each specific task. Given that they cannot profit directly from "task sales," why are leading smartphone manufacturers still willing to bear the costs of computing power and Tokens and conduct exclusive internal testing for mobile "Claw"?

One reason is that, on the road from traditional smartphone AI assistants to "personal agents," OpenClaw is approaching the ideal form of a "super assistant" without limits.

Unlike past voice assistants that could only respond passively, OpenClaw is more like a "digital employee" available around the clock, allowing ordinary users to truly experience the potential for AI to replace human labor.

From the underlying logic of OpenClaw, its core value lies in its powerful "autonomy." It breaks through the boundaries of the chat box: as long as it is equipped with the corresponding Skills and granted sufficient Tokens, OpenClaw can remember users' habits and tasks, autonomously plan steps, call tools, and operate software until it provides the final result.

However, to truly tame this ethereal "autonomy" down to a compact mobile device, relying solely on the accumulation of App features is clearly insufficient; it requires smartphone manufacturers to deeply reconstruct the operating system from the ground up.

In terms of implementation, whether it is Huawei's Xiao Yi Claw or Xiaomi's miclaw, both have conveniently chosen to enter as "system-level applications." This approach essentially unifies previously disparate software functions, system permissions, and even cross-device capabilities into Skills that the agent can call, and then organically connects them through a self-developed inference-execution engine.

Taking Xiaomi miclaw as an example, it integrates over 50 system tools and ecological services to build a closed-loop engine of "perception-inference-execution." In response to user commands, the engine autonomously breaks down steps, matches tools, determines parameters, and continuously adjusts based on execution results until the task is completely delivered.

On the other hand, Huawei's Xiao Yi Claw is constructed directly on the HarmonyOS architecture. "Xiao Yi Claw has three major advantages: system-level permissions (no need for third-party APP to switch; directly calls underlying functions), full-scene collaboration (seamless interaction among phones, PCs, car systems, and smart home devices), and data security isolation (local processing of user privacy data)," revealed an insider from Huawei.

However, when deploying "lobster" on smartphones, the challenges are not only technical and ecological but also involve properly handling sensitive data under the premise of safety and compliance, breaking down barriers across applications and platforms, and even restructuring the entire industry's profit distribution pattern.

“To deploy lobster on commonly used smartphones, the most important thing is to ensure information security,” emphasized an employee of a smartphone manufacturer.

This concern for information security is not unfounded. Due to the default weak security configuration of OpenClaw, it can be easily attacked, giving attackers complete control over the system; there have already been risks such as prompt injection, misoperation, and function plugin poisoning.

Faced with these hidden security "reefs," security governance has become an insurmountable red line for smartphone manufacturers when scaling up the deployment of "lobster."

Taking Xiaomi miclaw as an example, to prevent the Agent from executing high-risk operations like payments unilaterally in the cloud, miclaw directly "castrated" all tools involving money transfers and orders at the code level. This means that without user explicit confirmation through fingerprint verification or password input, any financial transaction actions will not be triggered, thus fundamentally locking down the risk of automatic deductions.

The Battle for AI Ecological Entry Begins

The nearing ideal form of OpenClaw as a "super assistant" is merely a surface incentive for smartphone manufacturers to rush into "farming lobsters." The deeper game lies in when users gradually get used to the interaction method of "just saying it to get things done," the old order dominated by traditional mobile internet centered on Apps, where smartphone manufacturers control the distribution rights of application stores, begins to loosen.

As Nvidia's founder Jensen Huang said: "Mac and Windows are the operating systems of personal computers, while OpenClaw is the operating system of personal AI."

In the PC era, whoever controlled the operating system held the gateway to the ecosystem. The same applies in the AI era, but the battle for entry has shifted to agents.

Imagine if users got used to resolving all problems on a third-party Agent (like a web portal or an independent app similar to OpenClaw), smartphones could become merely "hardware bases."

As internet giants rush to deploy mobile "lobster," the sense of crisis among smartphone manufacturers is obvious.

Just as smartphone manufacturers announced the launch of mobile "lobster," internet giants like Baidu and Alibaba have also quickly acted and begun free internal testing of mobile "lobster."

On March 12, Baidu launched the "Red Hand Operator" application on Android, allowing users to directly experience mobile AI assistant capabilities for tasks like hailing a taxi and ordering delivery across applications. Following that, Alibaba Cloud launched its mobile version of OpenClaw "lobster"—JVS Claw—on the next day, focusing on "out-of-the-box usability," where users can perform operations, handle files, and complete complex tasks in a securely isolated cloud space using simple natural language commands.

Regarding the deployment of "lobster" by smartphone manufacturers and internet giants, IDC China's research manager Guo Tianxiang stated, "Currently, the practical application value of 'farming lobsters' on smartphones is limited. The key bottleneck lies in API authorization issues if trying to call third-party Apps. If forced to call, it could end up disabled by those third-party Apps, like the earlier Doubao Phone."

With the lessons of "Doubao Phone," smartphone manufacturers like Huawei and Xiaomi are prioritizing validation within their ecological closed loop while deploying mobile "lobster."

For example, Xiaomi miclaw is currently focused on verifying the task execution capabilities of large models in the "car, home, and whole ecosystem," while Xiao Yi Claw prioritizes collaboration among Huawei devices like smartphones and tablets.

However, running "lobster" in a relatively closed ecosystem can avoid some risks, but it also somewhat restricts the potential of "lobster," since users' high-frequency demands often scatter across national-level third-party applications like WeChat and Douyin.

In the quest for balance between safety compliance and comprehensive functionality, manufacturers have not chosen to completely abandon cross-application collaboration but have turned to explore a more cautious and controlled technological path.

In regard to collaboration with third-party applications, technicians close to Xiaomi have revealed that currently, Xiaomi miclaw's collaboration with third-party applications mainly achieves two industry-standard methods: one is to drive applications or trigger specific actions through Intent (SendIntentTool); the second is to encourage applications to adapt to its AppTool SDK (based on the AIDL protocol), enabling deeper feature calls and task collaborations through preset data formats, allowing third-party Apps to actively push notifications to trigger tasks in Xiaomi miclaw.

“Distant Water” Cannot Quench “Near Thirst”

Currently, deploying proprietary "lobster" at the underlying system level is a key step in the evolution of smartphones to "AI phones." However, for manufacturers eager to seek growth in the AI wave, the primary challenge in building a super agent is cost pressure.

Deploying localized "lobster" is not just a simple software upgrade; it also requires upgrading hardware such as the core processor and storage. The high-frequency inference and real-time response demands of large models bring higher requirements for the NPU computing power of the core processor (SoC) while significantly raising the specifications’ threshold for operational memory and storage chips.

“Running large models on the smartphone side is affected by a series of technical limitations like storage space and power consumption. The larger the parameter count, the harder it is to run on a smartphone. A model with 1 billion parameters occupies 1GB of the phone's memory, 7 billion occupies 4GB, and 13 billion occupies 7GB," revealed a director of AI solutions at a leading smartphone manufacturer.

Currently, the storage chip price is on the rise, and every GB memory upgrade is directly squeezing the hardware profits of the entire device.

More difficult than a one-time hardware investment is the ongoing usage costs incurred once the mobile "lobster" is activated. On the PC side, each task execution corresponds to substantial Token consumption and computing fees. The earlier news that "earning 20,000 per month isn't enough to support lobster" directly brought this "cost anxiety" to the users' attention.

“Before using 'lobster', you need to first clarify what you want to do with it,” explained Feng Nian, founder of a content production agency (MCN), “There is actually a significant difference in Token consumption between editing video and generating video in the video production process, but many newcomers can't distinguish what 'lobster' can actually do.”

Feng Nian illustrated with the actual operations of his team: "We mainly deploy OpenClaw on Mac mini4 to assist our work. Specifically, 'lobster' is responsible for generating scripts for shop visit videos based on local trends; some of these scripts feature real human shots while others use AI (like Jimeng’s Seedance 2.0 or Sora2) to generate. The lobster can control the Mac mini to edit videos while calling Sora2’s interface to generate videos. Some steps are cheaper when done by people, while others are more cost-effective when given to AI. By the end of the day, about 12 original and mixed videos can be produced, with corresponding Token consumption cost of around 15 yuan.”

“The core difficulty of decision-making lies in balancing Token computing costs with the wages of entry-level editing personnel,” Feng Nian added, “Clarifying which tasks to hand to 'lobster' and which to leave for human workers is critical to using lobster effectively. Unfortunately, many companies participating in 'lobster farming' are merely following the trend without generating actual productivity.”

The daily Token consumption of 15 yuan seems low, but it cannot withstand the massive user scale of smartphone manufacturers. When hundreds of millions of users are accustomed to the model of "buying hardware and receiving free services," whether smartphone manufacturers can afford the subsequent vast computing power and Token costs in the long term remains uncertain.

“In the future, smartphone manufacturers might adopt a model of 'buy a phone and get computing power'," predicted an industry insider, “For example, offering a certain amount of free Tokens when purchasing the device, to handle daily light tasks like writing reports or booking tickets. For more complex high-consumption operations like video generation, they might charge separately based on task complexity, or users may have to bear the excess Token costs themselves.”

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。