The National "Shrimp Farming Craze": From Lining Up to Install to Paying to Uninstall, What Signal Does It Send?

Source: Securities Times Network Author: Wu Shun, Chen Yukang

“Did you raise a lobster today?” This seemingly is a greeting in the aquaculture circle, but it has recently exploded in popularity, becoming a nationwide buzzword since 2026. The “shrimp” here refers to an open-source AI agent called OpenClaw, affectionately nicknamed “Little Lobster” by netizens due to its red cartoon lobster icon. The process of deploying, training, and using this agent is humorously called “raising shrimp.”

Since the first official version was released over a month ago at the end of January, this nationwide craze driven by open-source technology has rapidly spread from the tech community to the general public: offline “raising shrimp” experiences in cities like Beijing, Shanghai, Guangzhou, and Shenzhen have surged; related online topics flood social media; major companies have launched “cloud shrimp” services; local governments have initiated “shrimp-raising” competitions; the AI concept in capital markets has reignited. It seems a revolution in popularizing AI agents is sweeping the country.

But behind this “shrimp-raising” boom, questions remain: Is there a real demand for “raising shrimp,” or is it an irrational expectation amplified by marketing narratives? Does the high barrier to entry and learning curve mean it’s still far from ordinary users? How can AI with extensive permissions ensure user safety? These issues will continue to test how far this “shrimp-raising” trend can go.

“Raising Lobsters” Becomes a Hot Topic: A Fierce “Policy Race”

OpenClaw can directly control computers and devices to complete tasks, enabling a leap from conversational assistance to autonomous execution, thus “freeing hands” and becoming a hot topic in tech and capital markets. On March 6, a long queue even formed outside Tencent’s headquarters: nearly a thousand developers and AI enthusiasts gathered at Tencent Tower, assisted by Tencent cloud engineers, to install OpenClaw on the cloud, collectively becoming “cloud shrimp raisers.”

Zhang Cheng, Assistant Dean of Fudan University School of Management and Professor and Department Chair of Information Management and Business Intelligence, told Securities Times that “Lobster” not only processes content but can also flexibly call various tools and combine strategies like humans to complete tasks. How it achieves this black box is left to AI to experiment freely, allowing users to focus only on what they want, shielding them from technical complexities in automation.

As of March 10, in recent trading days, A-share market OpenClaw concept stocks experienced a surge, with several listed companies responding.

UCloud stated that its lightweight cloud hosting products with OpenClaw images have not yet formed a scaled product system; technological iteration and commercialization progress may fall short of expectations, with limited short-term impact on overall performance. Autonomous AI frameworks like OpenClaw are still in early development stages, with uncertain future market potential, technical stability, and data security.

Previously, Nubia (ZTE) and ByteDance’s Doubao collaborated to launch Doubao AI phones, which became popular for features like “completing tasks in one sentence.” Regarding the impact of “Lobster” on AI phones, a ZTE terminal executive told Securities Times that they welcome the entry of “Lobster phones” into the market to foster growth. However, compared to the various competitive advantages of “Lobster,” AI phones also have core strengths.

“‘Lobster’ has a high usage threshold, requiring local deployment, manual skill configuration, and is prone to errors and security vulnerabilities. Doubao AI phones are ready to use out of the box, eliminating complex debugging, and all key functions are user-driven in the final step, forming a last line of defense,” the executive said.

Zhang Cheng believes that the high autonomy of “Lobster” comes at a cost. “The additional reasoning and programming test steps generated when AI tries different paths directly consume more tokens. Since more processes are handed over to AI, and AI itself can hallucinate or be fragile, this increases the risk of task loss,” he explained. “‘Lobster’ essentially represents a new balance between user-friendliness, computational costs, and task risks.”

In this “raising lobster” wave, local governments have also quickly followed suit, offering “real money” to attract developers and enterprises, creating a fierce “policy race.”

For example, Longgang District in Shenzhen issued the “Ten Policies for Lobster,” offering up to 2 million yuan in subsidies; Wuxi High-tech Zone explicitly supports up to 5 million yuan per project.

Hu Bo, Honorary President of Zhejiang Investment and Financing Association and Founding President of Suzhou Industrial Park Development Promotion Association, told Securities Times that recent government policies supporting “raising lobsters” are a continuation and upgrade of previous OPC (one-person company) community support policies.

“Since last year, many regions have focused on building OPC communities, mostly supporting co-working spaces or incubators, without deep involvement in the technical foundation,” Hu said. “Now, with the rise of OpenClaw, these OPC communities have a technical base and enabling tools, making the development logic more solid, which has attracted local government attention and support.”

High Learning Curve: “Raising Shrimp” Guides Reach 800 Pages

As the “shrimp-raising” craze continues, many ordinary users are researching how to raise this “shrimp” well and have high expectations for OpenClaw’s capabilities. In fact, from the OpenClaw deployment on a major company’s cloud observed by Securities Times, the learning curve is extremely steep. For non-technical “newbies,” even a simple computer term can cause hours of confusion.

Currently, many platforms have released guides and tutorials for OpenClaw. One such guide, with nearly 800 pages, covers beginner basics, four core functions, advanced skills, and practical cases. It contains many technical terms, making it seem like a “bible” to users without programming backgrounds. However, this tutorial is only a superficial introduction; most skills still require users to explore and learn on their own.

What tasks OpenClaw can perform depends on its underlying “brain”—the large model. The smarter the model configured, the better OpenClaw performs. Its specific capabilities rely on skills (skills), each capable of executing particular commands. Currently, the official skill marketplace ClawHub hosts nearly 20,000 skills, making it challenging for users to find suitable ones.

Even after finding the right skills, usage can be difficult. For example, if a user wants to automate posting on Xiaohongshu (Little Red Book), they need to log into Xiaohongshu on their local browser to get cookies, open developer tools with F12, refresh, find a request, copy the cookie, and send it to OpenClaw. For ordinary users, locating and copying this cookie is a significant challenge.

Senior AI investment expert and special researcher at Wangjing Society E-commerce Research Center Guo Tao said that, from an industry evolution perspective, the current popularity of “raising lobsters” is more of a phase of technological application hype rather than a mature AI terminal form. Its core driver is the inclusive nature of open-source technology and users’ curiosity—OpenClaw, as an open-source project, lowers the barrier for ordinary users to access advanced AI agents. People can easily debug it to perform basic tasks like file organization and information retrieval, which quickly sparks social sharing. But beneath the surface, its core remains experimental: functions are mainly on the PC side, limited to niche enthusiasts’ “geek” play, far from the public’s expectation of a “smart terminal.”

Guo Tao further pointed out that while open-source projects like “Lobster” provide valuable exploration for AI terminals, current agents lack stability, scene adaptation, and human-computer interaction standards for commercial use: low accuracy in understanding commands, limited planning ability for complex tasks, over-reliance on text instructions, and a lack of physical-world perception and interaction. These flaws mean they are more suitable for technical experiments at this stage rather than mass-market products.

Beware of Recurring Risks: Accelerate Building Permission Governance Framework

While OpenClaw has become popular, issues like accidental email deletion and privacy leaks have emerged. Many ads for paid “Lobster” uninstallation appear on social media, turning this trend into another hot spot. The mixed reputation of “raising lobsters” has prompted warnings from various departments, injecting a dose of caution into the fervor.

The Ministry of Industry and Information Technology recently issued the “Six Do’s and Six Don’ts” advice on preventing security risks of OpenClaw (“Lobster”) open-source AI agents, highlighting risks such as supply chain attacks, internal network infiltration, sensitive data leaks, hijacking, and financial transaction errors. Recommendations include using the latest official versions, strictly controlling internet exposure, adhering to the principle of least privilege, cautious use of skill markets, preventing social engineering and browser hijacking, and establishing long-term protective mechanisms.

Local authorities also advocate rational use of OpenClaw. On March 11, Suzhou Artificial Intelligence Industry Association issued a statement urging the promotion of professional services for OpenClaw, provided by specialized organizations for secure deployment, capability training, and trustworthy delivery, ensuring AI agents are embedded into business processes as reliable productivity tools. Strictly implementing security baseline configurations and following the principle of least privilege are emphasized.

The capital market responded swiftly. On March 11 and 12, several OpenClaw concept stocks in A-shares experienced sharp declines.

Guo Tao told Securities Times that, under current technology, the ambiguity of AI agent permissions could lead to excessive data collection, and the liquidity of open-source community code increases the risk of data leaks. If maliciously exploited, agents could become tools for privacy theft.

“More challenging is the difficulty in defining control and responsibility. When AI agents autonomously perform tasks and errors occur—such as incorrect transfers or mistaken messages—who bears responsibility—the user, developer, or device manufacturer? Currently, there are no technical standards for permission levels or legal regulations clarifying responsibilities,” Guo Tao said. “The industry should accelerate establishing permission governance frameworks based on the ‘minimum necessary’ principle. Regulations should also be improved to clearly define ‘AI behavior responsibility chains.’”

Beyond technical vulnerabilities and security risks, the social trend of “raising lobsters” needs correction. As Hu Bo noted, “Overhype and blind follow-up in the industry can mislead resource allocation and deviate from genuine development.”

He emphasized that, on the trend front, “safety must be prioritized.” Besides micro risks like data loss and privacy leaks, if the craze turns into disorderly social movements, broader chaos could ensue—such as scams under the guise of “raising lobsters” or distributed computing power, or hype-driven claims about advanced ecosystems to attract investments. Such phenomena have appeared repeatedly in past booms like the metaverse and blockchain.

Of course, despite risks and controversies, industry experts generally believe that the widespread adoption of OpenClaw reflects a paradigm shift in AI: from “dialogue-based interaction” to “autonomous execution.” This trend and transformation could have profound long-term impacts on the economy and society.

Zhang Cheng said that the popularity of OpenClaw also indicates an accelerating trend of multi-AI agent collaboration, which may change the current emphasis on specialization. For example, AI could handle accounting, legal documents, and market analysis, allowing founders to focus more on core creativity and strategy, lowering entrepreneurial barriers. While division of labor won’t disappear immediately, it may become less fragmented, leading to greater organizational flexibility.