Lobsters are very popular, but here are 8 "precautions" you should know.
Text | Wang Yuchan, Zhou Xinyu
Editor | Su Jianxun
At the beginning of March, Datan participated in an offline OpenClaw event in Shenzhen. The venue was packed with people, and the aisles on both sides were also full.
What brought hundreds of people together was the current hottest open - source Agent framework, OpenClaw, also known as "Lobster". This Agent, which can be directly deployed on a local computer, can operate the system, access files, and reply to messages like a human being. At the same time, it can break through physiological limits and work 24/7.
At the event site, many people sitting in the front rows wore red lobster hats. During the group photo session, people raised their hands, making a "V" gesture - just like the two pincers of a lobster.
The scene of the offline Lobster event.
Datan started a Web 3.0 startup in 2022. That once - popular industry, which also created widespread FOMO, is now uncared - about in China. After installing Lobster through Claude's learning materials, Datan quickly signed up for two Lobster - themed conferences in Shenzhen and found that "30% or even more of the people on - site were old acquaintances from the cryptocurrency circle."
The FOMO (Fear of Missing Out) sentiment had spread during the Spring Festival. Datan has a paid knowledge community. Even during the Spring Festival holiday, members of the community kept calling Datan. They were concerned about two important issues: 1. The impact of the war on the cryptocurrency circle; 2. Lobster.
Jean, a former Web 3.0 entrepreneur, also installed Lobster recently. During the Spring Festival, he sent New Year greetings to his friends in the circle and found that most of them had given up on Web 3.0 and fully embraced AI.
Those who experienced FOMO four years ago have once again become the main force in this wave of FOMO. "The Web 3.0 industry has fallen into a structural bear market, and practitioners are eager to find new growth points," Jean said. Two young men who founded a Web 3.0 startup with him have both turned to AI entrepreneurship after the team disbanded. "The talent, capital, and attention that originally belonged to Web 3.0 are now being sucked away by AI like a black hole."
This is a group of people who can smell opportunities the most. The current popularity of Lobster shows a strong characteristic of 'digging for gold at the emerging trend' - technology + marketing + get - rich - quick myths. There are several key nodes:
The nodes of Lobster's popularity explosion.
Under the triple promotion of well - known enterprises, famous people, and the government, Lobster, still an emerging thing, has suddenly become a huge consensus and a wealth code that everyone worships. A Cong, a technology entrepreneur in Hangzhou, while raising his own Lobster, also looks for business opportunities through on - site installation services. Within two days, his three customers were not practitioners with a technical background but ordinary people engaged in e - commerce, self - media, and stock trading.
"Everyone is very anxious. It seems that if you don't raise Lobster, you'll fall behind," A Cong said.
Datan still remembers standing at the crowded Lobster event site, seeing his old friends wearing lobster hats, shouting "Lobster changes the world" while busy taking short - videos with their mobile phones and posting them on social platforms. This picture of the superposition of "technological revolution" and "traffic marketing" made him feel exhausted.
After attending several suffocating events, Datan believes that there are three types of people stirring up this trend on the stage: those who sell 'Lobster mining machines' claiming to recoup the cost in half a year, those who sell '99 - yuan 7 - day training camp' courses, and entrepreneurs who ride on the hot topic to promote their AI companies or official accounts.
And those standing off - stage are mostly more afraid of being "eliminated" than of "losing money". In 2022, they were afraid of missing the once - in - a - lifetime opportunity to get rich in Web 3.0. In 2026, they wear lobster hats.
But the pressure valve turned quickly. Just two days later, voices about the practicality and data security of "Lobster" gradually emerged. The Ministry of Industry and Information Technology and the National Internet Emergency Center successively issued risk warnings. In the press release sent by DingTalk to the media, the title directly read "DingTalk does not recommend enterprises to install Lobster".
The services of "on - site Lobster installation with free cleaning/cooking" that used to be everywhere on Xiaohongshu and Xianyu have now become "Lobster uninstallation services", still priced at 299 yuan per time.
Everything has returned to the starting point.
Rational voices have begun to appear: Xie Xin, the CEO of Feishu, posted on his WeChat Moments, "Running an Agent on a personal computer and using an Agent in an enterprise are completely different things. The lower limit of security determines whether it can truly enter the work scenario."
Everyone is talking about Lobster, but they are talking about different things. Some people shout in live - streaming rooms and on their WeChat Moments that it is a revolutionary tool to open the next era, while others spend a lot of effort tuning Lobster and then evaluate it as "less capable than an intern".
Xu Feng, who has ten years of experience in software design and development and has installed nearly a hundred skills for his Lobster, believes that "Lobster is like the Yitian Sword. It is extremely powerful in Zhang Wuji's hands, but it becomes a burden in an ordinary person's hands. I encourage you to try using Lobster, but think clearly about your purpose. Don't just follow the trend, and don't stand by indifferently."
If you want to experience Lobster, here's what you need to know:
1. Lobster is not a necessity
Whether Lobster can satisfy you depends on whether you regard it as a sword or a kitchen knife, and whether you want to fight an enemy or cook a meal.
Since Lobster became popular for a month, Sun Linjun, the CEO of the Agent manufacturer Shizai Intelligence, has seen the number of inquiry orders soar by 3 - 4 times. But he told us that so far, no real commercial implementation has been seen.
Most users feel that Lobster is not practical after deployment. A Li is a clerk in a state - owned enterprise. Three days ago, she asked a programmer friend to deploy Lobster for her, but it has been idle ever since.
"I don't know what to use Lobster for," A Li admitted. Her main work usually involves writing speeches and drafting documents. There is already a quite efficient SOP within the company, and there seems to be no room for Lobster to play its role.
"Lobster is not practical because most users have no usage scenarios," an Agent entrepreneur told us. For most tasks related to search, research, and coding, the current models themselves are mature enough, and there is no need to deploy Lobster additionally.
However, Lobster also has its own uses.
Previously, A Cong was a Web 3.0 practitioner. The reason he started raising Lobster was to summarize the cryptocurrency market dynamics 24/7 and send a daily report to himself at 10:00 am the next day.
He has tried almost all the models and Agent products on the market, but most of them cannot meet his needs. In terms of the 'timed sending' function, A Cong found that among the model products on the market, only Grok can achieve it.
Xu Feng uses Lobster to organize tables, operate his WeChat Moments and official accounts. He carefully and strictly controls every instruction given to Lobster to ensure that the AI does not have hallucinations and make uncontrollable actions.
Liu Xi, who defines himself as a user with a "humanities - oriented thinking", said, "A mother can cook a delicious meal in an hour, not by brute force, but by making full use of tools: a clay pot for soup, a rice cooker for steaming sausages, an oven for roasting meat, an air fryer for frying vegetables, etc. It's just like having multiple Lobster assistants, each doing its own job."
This family education practitioner, who is building his own OPC, compares AI to various kitchen utensils. It is these "kitchen utensils" that make a one - person company more possible.
2. A 2000 - yuan on - site installation is not as good as the free official tutorial
Currently, those who have made their first pot of gold from Lobster are not the users but the "shovel sellers" who do the installation business.
On Xiaohongshu, a 500 - yuan on - site Lobster installation fee is already the basic price. If you want to configure Skills, the charge can be as high as 2000 yuan per Skill. The entrance fees for some offline activities under the guise of deployment and experience exchange also range from 200 to 500 yuan.
△ Some on - site installation services charge nearly 2000 yuan per time. Source: Xianyu
In fact, the quality of these installation services and offline activities varies greatly.
Even a one - on - one on - site installation is prone to pitfalls. A Li told us that a friend of hers bought an 800 - yuan on - site deployment service, claiming to be equipped with commonly used Skills. But after deployment, the friend found that these Skills did not have their own API Keys and PDIs configured and could not perform actual tasks at all.
Security is also an issue that cannot be ignored. Sun Linjun reminded that you must be present during the Lobster installation, especially for installations that authorize the other party to remotely control your computer. "Users must see the other party's operation process in real - time to prevent data leakage and poisoning."
The most reliable way to deploy Lobster is still to find a reliable programmer friend around you or use the installation package from a regular manufacturer.
When the interviewees were asked about the time spent on deploying Lobster, an interesting phenomenon emerged - the more tech - savvy a person is, the longer it takes; the less tech - savvy a person is, the shorter it takes.
Xu Feng, who has ten years of experience in program development and architecture design, spent nearly two days. Jean, who used to work on projects in the Web 3.0 field, spent one day. Datan, who does self - media in the Web 3.0 field, spent a few hours. And Liu Xi, who has a background in financial content and family education, only spent a few minutes.
The micro - time difference is caused by the macro - time difference. The more at the forefront of technology a person is, the earlier they install Lobster, and the greater the installation difficulty; the less tech - savvy a person is, the later they install Lobster, and the lower the threshold.
Although there is only a one - month difference between Xu Feng and Liu Xi in installing Lobster, the difficulty is worlds apart.
The earlier you enter the game, the higher the time and money costs. "I know a company that spent tens of thousands of yuan to install Lobster for five or six people in the company and connect it to Feishu," Datan said.
Less than half a month later, Feishu officially held a live - streaming event. In the live - streaming room, you could order the Feishu OpenClaw digital employee for 9.9 yuan.
"At this time, we should mention Duan Yongping's words again - be brave to be the latecomer," Datan said.
3. Pay attention to the cost. For simple tasks, find a 'flat - substitute' model
The unexpectedly high Token cost is the first tuition fee paid by many Lobster raisers.
The so - called Token can be simply understood as a unit to measure the workload of AI. In our communication, there are many Lobster users who consume tens of thousands or even hundreds of thousands of Tokens in a single task, corresponding to a high daily expenditure of dozens of US dollars.
Although Datan, who calls overseas large - scale models, uses them relatively restrictively, he still spent nearly 500 yuan in half a month. "If there is no restraint, the monthly expenditure could be as high as tens of thousands of RMB," Fu Sheng, the CEO of Cheetah Mobile, also admitted in a live - streaming session that he spends more than 100 US dollars a day on raising Lobster.
In fact, this high cost is often not worth it. A "new Lobster raiser" engaged in financial work told us that when asking Lobster and Kimi to perform the same task of organizing a financial briefing, Lobster consumes three times as many Tokens as directly communicating with Kimi.
This extra "unnecessary expense" comes from Lobster's trial - and - error, feedback, and self - evolution process.
Sun Linjun, the CEO of Shizai Intelligence, explained to us that currently, Lobster does not have a fixed execution framework for tasks, that is, an efficient SOP. This results in Lobster consuming a large number of Tokens for pre - task trial - and - error and adjustment when performing any task.
Therefore, it is recommended that Lobster raisers manage their wallets well. For example, purchase model services with a Token limit and do not authorize Lobster to automatically renew.
Moreover, many developers also suggest that for single - point and simple tasks such as search and document parsing, you can directly use mature Agent and model products on the market, which are not only cheaper than Lobster but also have a higher success rate.
4. There is no 'perfect - born' Lobster. Raising Lobster is as difficult as leading a team
During a recent on - site service, less than two hours after installing Lobster, A Cong received a complaint from the customer: Lobster was very "stupid" and couldn't even complete a simple search.
This is also the misunderstanding of most "newbies" when using Lobster - thinking that it can be used out of the box and can complete complex tasks right from the start.
The factory - set Lobster is like a "student" with only the knowledge of the base model and no practical experience.
To enable Lobster to perform various complex tasks, relevant skill packages need to be implanted, that is, the "SKILL.md" file with built - in tool instructions. These Skill installation packages can be obtained from technical communities such as GitHub or written by yourself using AI Coding tools.
What kind of Skills should be configured for Lobster? Which Skill installation packages can precisely meet the execution requirements? Just answering these two questions can discourage a large number of users.
A Cong told us that configuring Skills for Lobster is like forming a complete project team: "You need to break down the task objectives, execution processes, and role divisions. Each division may correspond to a Skill."
Configuring Skills is just the beginning of "raising" Lobster. "Lobster needs a certain amount of time to learn to use Skills and adjust Skills according to tasks," a developer said.
Fu Sheng described the process of configuring Skills in his Lobster - raising PPT as "learning through pitfalls": Every time you encounter a pitfall