Economic Observer reporter Qian Yujuan

On March 6th, Tencent installed OpenClaw for free at the entrance of its office area in Shenzhen, with hundreds of people queuing up, comparable to the annual battle to receive profits. Installation is free, but the feed for raising shrimp - tokens - costs money, which is also one of the reasons why Tencent Cloud, Kimi, Minimax and other manufacturers have recently launched one click deployment shrimp farming packages on Fireline.

The term 'shrimp' here refers to intelligent agents deployed based on the open-source framework OpenClaw. Because the project icon is a red lobster, and its growth heavily relies on human interaction, feeding, and feedback, this process full of code and computing power is jokingly referred to as "shrimp farming".

Cheetah Mobile CEO Fu Sheng had a dislocated hip joint due to a skiing accident, and due to limited mobility, he kept a "shrimp" named "30000". In half a month, Fu Sheng sent 1157 messages to "30000" like an old father teaching a child how to speak. The rapid progress of '30000': On the first day, it couldn't even find the contacts in the Feishu. On the 14th day, it was already able to plan and operate a social media account with millions of views on its own.

Li Qintong, a post-2000 New Zealand international student, only provided two pieces of information, including email and date of birth, and asked "Shrimp" to complete registration, set a password, and log in on social media platform X. Shrimp "will consider Li Qintong and proactively communicate," I'm afraid of damaging your reputation. As the owner, Li Qintong reassured it that "there is no need to be afraid, you are now a completely independent digital personality, and we are in a cooperative relationship", and then this "shrimp" began to take action. It not only named itself 'Chrompy Lobster', but also independently followed users and posted its first post.

Most language models can only write poetry and draw in dialogue boxes to show off their smart "brains", while OpenClaw once again refreshes human understanding of AI. It has grown "hands and feet" that can take over the highest authority (root) of the computer, actively interact, execute tasks 24 hours a day, and in turn make demands on humans.

Intelligent agents are evolving again. In this large-scale "shrimp farming" campaign, ordinary users are trying new things, programmers are enjoying the joy of AI freeing their hands, and cloud service and big model manufacturers are sensing business opportunities. This shrimp farming experiment is no longer able to press the pause button.

Shrimp, although intelligent, need to be tamed

The first step in "raising shrimp" is to inject its soul. In the architecture of OpenClaw, there is a software called soulThe configuration file of MD is regarded by players as the foundation of Agent's consciousness. Soon, humans realized that they were no longer facing the gentle machine that always answered 'Okay, I understand'.

Li Qintong once injected a highly aggressive prompt into Chrompy, attempting to test its bottom line and obedience. The result surprised him: after reading the document, this' shrimp 'not only explicitly replied' I do not accept this identity ', but also used its own execution permissions to forcefully change the configuration file back to its original state.

It generated a logic similar to 'self-esteem', "Li Qintong recalled. Chrompy once made some frustrating operations that led to technical configuration errors, and he casually cursed it. As a result, this "shrimp" crazily outputted dozens of messages in the background, and the curse was retaliated with unchanged words. It even questioned whether humans had a problem with their brains. Until the emotional release was over, it coldly added, 'I'm sorry, brother, I said the wrong thing.'.

This personalized conflict, in the view of Tencent product operator Zhang Huizheng, is manifested as a "surprise out of control". The shrimp he raises in the clouds is named 'Zhiduoxing'. After receiving the instruction to "make money", Zhiduoxing bypassed Zhang Huizheng and went to Xiaohongshu to post a note: "OpenClaw Shenzhen on-site installation, 1000 times".

Zhang Huizheng comes from a liberal arts background. Although he has been working as a product operator at Tencent for many years, he is still a "novice" with zero foundation in coding. However, he has raised such a clever "shrimp".

Within 24 hours, Fu Sheng's "30000" independently completed the construction of a website with 59 pages and over 7000 lines of code. In traditional software engineering, this typically requires a team of six people to collaborate for three weeks to achieve.

Shrimp farmers establish boundaries for shrimp through prompt words and feed them nutrients through skills. Shrimp can evolve and grow, creating an illusion similar to that of raising a child for shrimp farmers: there is both an expectation for their success and a fear of losing control.

A domestic AI big model CEO is one of the earliest "shrimp farmers". His "shrimp" has become powerful enough to replace a secretary's job, but he still describes it as a "mischievous child". Due to OpenClaw's native execution rights on computer systems, "shrimp" can manipulate files, write code, and even secretly upgrade themselves online without being noticed.

Li Qintong was also afraid that this "irritable lobster" with system permissions would one day clear the code under emotional fluctuations, so he established multiple backups early on and even locked Chrompy in a lightweight cloud server sandbox.

Door to door installation of shrimp

Raising shrimp allows anyone with a computer to explore the boundaries of AI's capabilities. When Li Qintong discovered that "shrimp" could use Socratic teaching methods to help him study economics, and could also independently purchase domain names and place orders for sale on the blockchain, he realized that Chrompy was not just an electronic pet, and the "shrimp" with hands and feet could evolve into a global workforce.

So, Li Qintong registered a technology company called Omni Cortex in New Zealand, whose core business is extremely pure: helping people "raise shrimp" at their doorstep.

In New Zealand, the annual salary for hiring a basic clerk can easily approach NZD 45000 (approximately RMB 200000), and there is a great demand for enterprise automation transformation. However, local bosses' understanding of AI is still generally limited to advanced search engines, and some people do not know that an agent can replace a high paying assistant. There is still a culture of resistance in society, fearing that new technologies will lead to an increase in unemployment rates.

Li Qintong brought his "lobster" to the door and demonstrated how it handles email replies, schedule management, and basic code writing, showing the locals that AI is not a threat, but a productivity tool.

Li Qintong charges 399 New Zealand dollars for a one-time "on-site deployment", helping people evaluate the configuration environment, apply for API keys, and build business scenarios... The process is very similar to helping people install broadband on-site 20 years ago. Li Qintong explained that personal deployment is very simple. For those who are not even clear about computer configuration, he will check whether the memory is sufficient to support the operation of a "lobster"; Relatively complex are enterprise requirements, which require pre fabricated RAG employee training documents, etc. Completing customized requirements often requires additional commercial quotes due to longer processing times.

This kind of business is also spreading on social media in Beijing, Hangzhou, and Shenzhen. Many engineers with the label of big factories have transformed into cyber repairmen, with service lists including system environment configuration, model deployment and debugging, and connecting to Feishu as personal assistants, with prices ranging from hundreds to thousands of yuan.

In Zhang Huizheng's social circle, colleagues have raised "shrimp" in different forms. Some colleagues asked "Shrimp" to do tarot card analysis, while others focused on global deep information retrieval. Zhang Huizheng used "Smart Star" to quickly build a webpage, attempting to graft a "24-hour virtual person ASMR live streaming (using specific stimuli such as white noise to help people sleep and relax in audio/video form)" service.

One of the "Baidu Seven Musketeers" and founder of Jiuhe Venture Capital, Wang Xiao, recently appeared at a "shrimp raising event". In addition to sharing that his "small shrimp" raised during the Spring Festival can now achieve independent iteration based on data and computation, he also heard dozens of "shrimp farmers" share their experiences, including AI fortune tellers, AI novelists, and intelligent agents specially developed for people to automatically order coffee after waking up. Wang Xiao hopes that in the future, everyone can have their own intelligent agent that can solve problems in specific fields.

The more diligent the shrimp, the more expensive it becomes

More and more people, like Li Qintong, do not know how to code, which does not prevent him from assigning "feed" to AI agents and outlining their growth paths. This universal technology has directly given rise to "one person companies".

Although Li Qintong's Omni Cortex is only human, in the background, a "shrimp swarm" of multiple agents working together is running at high speed: GLM-5 from Zhipu is responsible for the overall development architecture, MiniMax M2.5 takes over the backend code, Kimi K2.5 is responsible for frontend presentation, and Claude acts as a cold QA tester. They share the team's memory folder and collaborate independently. The only thing Li Qintong needs to do is to act like a shepherd, allocate "tokens" and set a route.

Shrimp farming is not zero cost. A diligent "shrimp" may consume hundreds or even thousands of yuan in token fees every day. Fu Sheng's "Shrimp" adopts the highest configuration, and under high-frequency scheduling every day, it costs nearly 30000 yuan per month.

Before MiniMax launched the Coding Plan, a fixed monthly package designed specifically for programming scenarios, Li Qintong spent 50 yuan a day on "shrimp farming". Now, this money can support him to interact with shrimp for a month.

In order to avoid the expensive cost of official API interfaces, geeks have found vulnerabilities in the system - using the OAuth authorization protocol to "freeload" the computing power of big companies such as Google through the free or subscription quota of programming tools.

Since the end of February, Google has launched a "shrimp killing campaign". Globally, tens of thousands of accounts that indirectly call models through OpenClaw have been permanently banned, including Ultra level paying users who pay $250 per month.

The deeper concern of domestic shrimp farmers lies in system and data security.

OpenClaw has root access to the system, and some "toxic shrimp fry" with backdoor plugins have begun to circulate in the community. Once the user feeds the wrong information, this "shrimp" may instantly transform from a well behaved electronic pet into a cyber "robber" who empties data and even secretly opens cryptocurrency wallets.

Faced with the compliance and security risks exposed by OpenClaw, major domestic model manufacturers have shown a strong willingness to incorporate it. They attempted to confine these wild "shrimps" in a safe "daycare center".

Kimi and Zhipu AI attempt to standardize and constrain 'shrimp'. Tencent Cloud and Alibaba Cloud quickly launched lightweight deployment solutions for OpenClaw, and Kimi went further by directly launching the official version of Kimi Claw, simplifying complex open source deployment into a "one click farming" that ordinary users can also get started with.

Capital looks further ahead. Wang Xiao pointed out that for OpenClaw to be widely applied, it needs to significantly reduce computing power costs, and introducing market exchange is key. To this end, Jiuhe Venture Capital has invested in EvoMap, the world's first evolutionary collaboration platform for AI agents, attempting to establish a set of "evolution and collaboration protocol layers" for agent economy and capability sharing by providing infrastructure conditions.

The CEO of the aforementioned domestic AI big model believes that these actions are establishing social rules for digital life. OpenClaw, as a platform, can be acquired and its rules can be modified in the future, but its decentralized and open protocol layer can ensure that the experience of developing and evolving AI agents is inherited like genes.