In the spring of 2026, China's AI battlefield witnessed its most disruptive strategic shift. Moving beyond simply competing on the parameters and benchmark scores of large-scale underlying models, tech companies are now focusing on AI agents capable of actually controlling computers and software—a concept jokingly referred to in the industry as "shrimp farming." Tencent not only announced in its financial report that it would double its AI investment this year, but also quickly launched a public beta of "QClaw," transforming WeChat into a super remote control for office computers; while Xiaomi also released its MiMo-V2 series of large-scale models and..."MiMo Claw"It demonstrates powerful cross-modal system-level operational capabilities.
When you receive an urgent message from your boss on a crowded subway, requesting the immediate transfer of a file stored on the company computer, you no longer need to rush back to the office. Simply speak a sentence to AI on WeChat, and the remote computer will automatically locate the file and send it for you. This is precisely the tremendous impact that the "shrimp farming" era has brought to modern offices.
Tencent's QClaw surprise attack: Using WeChat to remotely control a computer, Ma Huateng's "decentralization" ambitions.
During Tencent's earnings call on March 18, Tencent, which delivered a stellar fourth-quarter report for 2025, clearly stated that its investment in the Hunyuan Big Data Model and new AI products will at least double in 2026.
Even more noteworthy is that Tencent CEO Pony Ma publicly discussed his strategic vision for "lobster farming" for the first time. He believes that AI should not be limited to the centralized entry point of ChatBot, but should be implemented in a variety of rich scenarios through "lobster" applications.
This aligns perfectly with WeChat Mini Programs' consistent "decentralized" philosophy. The future AI-powered intelligent entity ecosystem will be a combination of centralized traffic entry points and decentralized services.
Along with this vision, Tencent quickly launched the beta version of QClaw, based on the WeChat ecosystem. This tool precisely addresses the pain point of the previously high barrier to deploying AI agents (eliminating the cumbersome download of GitHub source code and Python environment configuration). Users can remotely control their PCs using natural language simply by accessing the platform through WeChat.
Even more strategically threatening is QClaw's strong "pragmatism." Instead of forcibly bundling with Tencent's own Hunyuan model, it embraces the OpenClaw open-source ecosystem, allowing users to access any large language model, such as Kimi and DeepSeek, via API. Tencent's intention is clear: it doesn't care whose underlying model it is; what it wants is to control the "distribution rights and interaction gateway" in the AI era—as long as users get used to issuing commands through WeChat, Tencent will win this battle to defend its underlying communication infrastructure.
Xiaomi claims the mystery model: MiMo-V2 joins the battle, creating a system-level native intelligent agent.
Just as Tencent was heavily promoting QClaw, Xiaomi dropped a bombshell. Xiaomi officially stated that the anonymous models Hunter Alpha and Healer Alpha, which recently topped the charts on OpenRouter, the world's largest API aggregation platform, are early test versions of its newly released MiMo-V2-Pro and MiMo-V2-Omni.
Led by Luo Fuli, a former core member of DeepSeek, Xiaomi's large model team's release this time is entirely focused on "optimizing intelligent agent capabilities":
• MiMo-V2-Pro:With total parameters exceeding 1T, it ranks third on the OpenClaw leaderboard (second only to the Claude 4.6 family). Its ultra-long context and powerful tool calling capabilities can automatically orchestrate complex workflows (such as automatically crawling stock market data and generating web pages).
• MiMo-V2-Omni:The multimodal base model can understand complex environments across modalities. For example, after understanding commands, it can open a browser on its own to search for product guides and compare specifications on Xiaohongshu, and finally automatically jump to JD.com to place an order. It can even connect to human customer service to help "bargain" the price.
• MiMo-V2-TTS:The speech synthesis model, trained on hundreds of millions of hours of speech data, supports multiple dialects and subtle emotional control, enabling the intelligent agent to communicate more like a real person.
Xiaomi also launched a free trial zone for "MiMo Claw" at the same time. As a terminal hardware manufacturer, Xiaomi's release of three models specializing in Agents suggests that it is very likely to deeply integrate the underlying big model, system permissions and Mi Home ecosystem services in the future to create a truly seamless "system-level native smart agent".
Convenience and hidden dangers: Security anxieties behind extreme productivity
While QClaw and MiMo Claw demonstrate impressive automation capabilities, they also amplify serious cybersecurity and privacy risks.
Taking Tencent's QClaw as an example, because it requires extremely high system privileges to simulate mouse and keyboard operations and read local files, once a user's mobile phone is lost or their WeChat account is stolen, malicious individuals can easily remotely view the user's computer screen and even manipulate the user's computer desktop and browser history. In the current stage of AI agent development, "absolute convenience" and "absolute security" are still almost mutually exclusive.
Analysis of viewpoints
If 2024 and 2025 were an arms race for "conversational AI," then 2026 will undoubtedly mark the beginning of the real-world application of "actionable AI."
The "shrimp farming" battle between Tencent and Xiaomi reveals a shift in the strategic thinking of tech giants: instead of pouring endless computing power into pursuing an unattainable AGI (Artificial General Intelligence), they are leveraging open-source communities (such as OpenClaw) to focus on solving the user's "last mile" interactive experience.
Tencent's approach was both clever and fatal. By leveraging WeChat's extremely low communication costs, it directly bypassed the "moat" of heavy-duty enterprise SaaS applications like DingTalk and Lark, which emphasize "All-in-One" solutions. When users discover that they can handle most of their office needs remotely using the free WeChat paired with QClaw, the operating system in the AI era may no longer be Windows or macOS, but rather the communication software you use every day.
Besides Tencent and Xiaomi, OPPO has also recently revealed plans to launch its own...Shrimp farming serviceNVIDIA has recently also proposed solutions to enterprise deployment needs.The software solution named NemoClawIt touts "one-click installation" and aims to inject enterprise-grade security, privacy, and scalability into this fastest-growing open-source project in history, attempting to introduce it to the commercial market with enterprise-grade security.
