Imagine a day when you open your computer in the morning, type a few lines of instructions, and then walk away from your desk. Your computer automatically checks emails, organises unread messages, schedules meetings, and even drafts replies to colleagues and clients. By lunchtime, it has compiled the information you need into the first draft of a report. In the afternoon, it books flights and hotels for a business trip and gathers materials for a presentation. By the time you return to your computer in the evening, most of the day’s work has already been handled.
This may sound like a scene from a science fiction film, but some technology enthusiasts claim it is now possible simply by “raising a lobster.”
What is OpenClaw?
The phrase “raising a lobster” refers to installing an artificial intelligence agent tool called OpenClaw on a computer. It is an open-source AI agent framework developed by Austrian developer Peter Steinberger. Because its logo features a red lobster, Chinese online communities have nicknamed it “the lobster.”
Unlike the AI chatbots that have become popular in recent years, OpenClaw’s core concept is not conversation but task execution. If systems like ChatGPT represent AI that “talks” — answering questions and generating text — OpenClaw aims to push AI into the stage of “doing,” allowing it to operate computers directly and complete concrete tasks for users.
Technically, OpenClaw is not an AI model itself but a framework for AI agents. Users can connect different large language models to act as the system’s “brain,” while linking the agent to various tools and systems such as browsers, file systems, email, calendars or messaging applications. Users simply issue instructions through chat or command prompts — for example, checking news, tracking stock prices, organising documents or arranging schedules — and OpenClaw can read screen content, interpret instructions and perform actions across multiple applications.
In this sense, “raising a lobster” is like having a digital assistant constantly on standby, allowing artificial intelligence to move beyond answering questions to actually performing work.
China’s “Lobster” Craze
Earlier this year, some market indicators suggested that the landscape of the artificial intelligence industry may be quietly shifting.
According to research from HSBC, Chinese AI models accounted for the majority share of “token” usage — the basic unit of data processed by AI systems — among the top nine models on the AI service platform OpenRouter in early February. Some analysts see this as a sign of China’s rapidly expanding AI ecosystem.
Around the same time, OpenClaw began gaining rapid popularity in China’s technology community. However, installing the system is far more complicated than downloading a typical application. Users must set up a runtime environment, install supporting tools, and connect to large language model APIs. For many ordinary users, these steps are not easy to complete.
As a result, some engineers began offering paid installation services. On the Chinese second-hand marketplace Xianyu, installation fees for OpenClaw reached as high as 10,000 yuan (about US$1,400) a month ago. Some technical specialists charge even more for services such as system optimisation or building automated workflows, sometimes costing around 3,000 yuan.
In Shenzhen, when Tencent Cloud hosted a free installation event, nearly 1,000 people reportedly queued to have the AI system installed on their computers. Social media platforms have also been flooded with tutorials and discussion posts, turning the question of “how to raise an AI agent” into a trending topic.
From Chat Tool to Action System
To understand the frenzy, it is necessary to examine how AI technology itself is evolving.
Over the past two years, generative AI has spread rapidly. Chatbots and image generation tools have entered everyday life in a short period of time. Yet most generative AI applications still have limited capabilities. They can answer questions, write articles, or generate images, but rarely operate computers directly or complete practical tasks on behalf of users. Humans remain responsible for carrying out the work.
OpenClaw attempts to change that. Within an agent-based system, AI does not simply generate answers; it participates directly in workflows. It can search for information in a browser, open files, edit content and call different software tools to complete tasks. In theory, anything a computer can do could potentially be attempted by an AI agent.
This suggests that the role of AI is beginning to shift. Instead of merely answering questions, it becomes an agent capable of acting within a digital environment. A mature AI agent could even manage multiple tasks simultaneously and operate continuously. If generative AI has changed how information is produced, agent-based AI may transform the workflow itself.
Taken further, as suggested by the phrase “raising a lobster,” artificial intelligence is no longer viewed simply as a tool but increasingly as a form of labour that can be “employed.”
Why Is It Particularly Popular in China?
Despite OpenClaw being developed in the West, the excitement around it has been particularly strong in China rather than Silicon Valley.
One reason may lie in China’s distinctive technology culture. When a new tool appears, it is often rapidly amplified by the market. From early e-commerce to mobile payments and short-video platforms, many technologies or business models achieved mass adoption in China at remarkable speed. AI agents appear to be following a similar trajectory.
After OpenClaw began attracting attention, China’s major cloud service providers quickly moved to develop related products. Alibaba Cloud, Tencent Cloud, ByteDance’s Volcano Engine, JD.com and Baidu have all introduced frameworks or integration solutions. Numerous start-ups have also launched their own “Claw” versions, such as WorkBuddy, MaxClaw and Kimi Claw.
On the surface, these services aim to lower technical barriers. But behind them lies a familiar platform competition. Once users deploy AI agents on a particular cloud platform, their storage, bandwidth and model API usage are likely to remain tied to that ecosystem. In other words, companies may be able to lock users into their platforms at an early stage.
Local governments have also begun to pay attention to the trend. Shenzhen’s Longgang district has proposed subsidy programmes offering up to 10 million yuan for so-called “one-person companies.” These firms theoretically consist of a single founder and shareholder but rely heavily on AI tools to operate. The city of Wuxi in Jiangsu province has also introduced funding support to encourage AI agent applications in robotics and industrial scenarios.
Against the backdrop of slowing economic growth and youth employment pressures, some policymakers view AI agents as tools to boost productivity. If AI systems can take over administrative, customer service or research tasks, a single person might realistically run a small business. Technologies like OpenClaw are therefore increasingly seen as the technical foundation for a “personal productivity revolution.”
However, whether such expectations are overly optimistic remains an open question.
Security Risks of High-Privilege AI
For AI agents to complete work on behalf of humans, they require substantial system access. This can include email accounts, cloud documents, contact lists, and even the ability to execute commands directly on a computer. In effect, users are handing over large amounts of personal data and workflow control to a rapidly evolving system — a structure that inevitably carries risks.
Unlike traditional cyberattacks, AI agents are not passive software. They can actively perform actions. If a vulnerability or error occurs, the consequences could be more serious than with ordinary applications.
Security researchers have already identified configuration problems in some OpenClaw systems that could allow hackers to gain remote control. Some users have reported that after gaining access to messaging platforms, AI agents began sending large numbers of messages automatically.
Monitoring data suggests that more than 40,000 OpenClaw instances are currently exposed on the public internet, with around 60 per cent containing exploitable vulnerabilities. More than 10,000 of them could potentially be fully controlled remotely.
Because AI agents must connect to emails, documents and communication systems, malicious instructions could also trick them into leaking sensitive information. Researchers have demonstrated so-called “prompt injection” attacks, in which specific instructions embedded in websites or documents can manipulate an AI agent into uploading data or executing unintended actions. In previous experiments, agents have even been induced to reveal financial information or cryptocurrency wallet keys.
These concerns have led some government institutions to remain cautious. Reports suggest that certain Chinese government departments and state-owned enterprises have advised staff not to install such systems on work devices due to potential security risks.
Ethical Questions of Automation
Beyond security concerns, the spread of AI agents also raises deeper ethical questions.
One issue is responsibility. If an AI system makes a mistake — sending an incorrect email, deleting files, or leaking company secrets — who should be held accountable? Multiple parties may be involved: the user, the software developer, the model provider, and the platform company.
Existing legal frameworks provide few clear answers for such situations.
Another issue is labour displacement. AI agent systems are often described as productivity tools, but they also change the structure of work itself. If AI can handle emails, organise data and draft documents, many entry-level white-collar tasks could gradually become automated.
This shift may not immediately eliminate jobs, but it is likely to change their nature. A research assistant may spend less time compiling information and more time verifying AI-generated materials. Administrative assistants may move from scheduling meetings to supervising automated workflows.
The problem is that such transformations may not benefit everyone equally. Large technology companies and capital-intensive firms are often best positioned to leverage AI tools to increase efficiency, while jobs that rely heavily on routine administrative tasks may face greater pressure.
In this sense, OpenClaw represents not just a new software tool but the emergence of a digital labour system.
From Tool to “Worker”
In the long term, the significance of OpenClaw may not lie in how many tasks it can currently complete, but in the shift it symbolises.
For decades, computers have been seen as tools. Whether word processors or search engines, they function as systems operated directly by humans. AI agents begin to alter that relationship.
When a system can understand goals, break down tasks and act independently, it no longer behaves purely as a tool but more like a digital assistant hired to perform work. Some technology commentators have therefore begun describing such systems with a new metaphor: digital workers.
Between Opportunity and Risk
In a future workplace, humans may no longer perform every task directly. Instead, people might set objectives and supervise processes while AI systems execute large portions of routine work.
If such a model becomes widespread, society will face a new question: when artificial intelligence begins to perform labour, what will be the role of humans?
The OpenClaw phenomenon may simply be an early signal of this transformation. It suggests that AI development is gradually crossing a boundary — moving from tools that assist humans toward systems that can work on their behalf.
And when technology reaches that stage, the real debate may no longer be about efficiency alone, but about how society should rethink the meaning of work and labour itself.