Why AI agents are being adopted faster in China than elsewhere?

The OpenClaw frenzy shows that China approaches artificial intelligence very differently from Western countries.

One should be wary of technological hype. It often produces more noise than reality. But the frenzy triggered by OpenClaw in China deserves attention, because it reveals something deep about the way the country conceives AI.

In just a few days, an open-source AI agent model moved beyond the circle of developers to become a mass phenomenon, with queues, collective installation sessions, and an almost carnival-like staging. Events organized by Tencent attract children, retirees, and curious onlookers, wearing lobster-shaped hats, the emblem of the OpenClaw project.

This frenzy has also taken on a very Chinese form: competitive imitation at the national level. A platform war has begun. In just a few days, Zhipu launched AutoClaw, presented as a version with more than 50 pre-installed skills. ByteDance, via Volcano Engine, released ArkClaw. Tencent put WorkBuddy online for professional use, before extending the offensive to other components such as QClaw and then ClawBot within WeChat. Alibaba responded with its own enterprise-oriented offerings. In other words, several giants rushed simultaneously onto the same open software layer.

This is where the economic core of the story lies. OpenClaw is free. The code is open. The model therefore does not, in itself, constitute a defensible rent.

Value shifts elsewhere: default installation, interface, the implicit marketplace of skills, browser automation, integration into messaging platforms, work suites, and models. In short, the battle is not about ownership of the code but about control of distribution. It is the browser war replayed in the age of agents: the browser was free, but the real power came from controlling the entry point. The market understood this very well. Zhipu’s stock jumped by nearly 13% on the day AutoClaw was launched.

Mimetic competition

The other specifically Chinese trait is the speed with which local governments have followed suit. The most tech-oriented cities—Shenzhen, Wuxi, Hangzhou, Hefei, and the development zones of Suzhou—published within days support plans for the OpenClaw ecosystem, in the name of the national AI+ initiative aimed at integrating AI   across the entire economic fabric. For example, subsidies of up to 10 million yuan (1.3 million euros) for certain companies, computing resources offered, housing, low-cost office space, and mechanisms specifically designed for the so-called “one person companies,” these micro-structures boosted by AI agents.

The mechanism is typically Chinese: a technological signal appears, territories immediately enter into mimetic competition to capture talent, projects, and political prestige.

The “lobster paradox”

Like any speculative phenomenon in China, the rush did not escape Beijing’s attention. On March 22, the Chinese National Computer Network Emergency Response Technical Team/Coordination Center (CNCERT/CC) and the Cybersecurity Association of China published an official guide to best practices. The message is clear: an agent capable of executing system commands, reading and writing files, and exchanging data with elevated privileges can become a formidable productivity machine, but also a security sieve. The Chinese press quickly coined the expression “lobster paradox.” The precautionary rules are very concrete, notably not installing OpenClaw on one’s daily work computer but on a dedicated device, a virtual machine, or a container, with environment isolation.

This is what makes OpenClaw more of a symptom than a one-off event. Many Western observers still view Chinese AI through the lens of Silicon Valley, wondering whether Beijing is truly pursuing general artificial intelligence in the American sense of the term.

This is not the case. Chinese models are progressing but remain structurally behind so-called “frontier” models. The OpenClaw episode tells a different story: China sees AI less as a divine machine installed in a data center than as a layer to be deployed everywhere—in work, services, interfaces. And tomorrow, in robots and infrastructure.