Original · Unique Research · 2026-04-30
Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the opening essay and full panel, including all named speaking turns and their continuation paragraphs. Cost, efficiency, customer and revenue figures are source or speaker claims, not independently audited findings. "养虾" (shrimp farming) is Chinese internet slang for deploying and running the OpenClaw agent framework, whose logo features a red lobster/crawfish; "弃养" (abandoning) means stopping its use. "小龙虾" (crayfish) is the colloquial nickname for OpenClaw. "一人公司" / OPC (One-Person Company) and "牛马" (workhorse) retain their colloquial force. The 75/80/90/95-point scale is the speakers' informal capability grading, not an industry metric. Company, personal and work titles are transliterated where official English forms remain unverified. The source is dated April 30, 2026.
Unique Awards
Tearing Off the One-Person Company Filter: Why Can't Your AI Agent Make Money?
When everyone can use AI to reach 75 or 80 points, delivery at that level becomes completely meaningless.
A couple of months ago, social media was almost flooded with "shrimp farming" (deploying OpenClaw). It seemed that as long as you installed this red lobster software and learned a few prompt words, everyone could instantly ascend and become the legendary "one-person company" (OPC) drowning in orders.
But at a Hong Kong Global Unique Awards roundtable a few days ago, several people who actually make money with AI mercilessly tore off this filter.
Honestly, after hearing their real playbooks, I drew a cold breath. In this nationwide fanatical AI bubble, we might not even know how we're being harvested.
The Filter Shatters: The "Super Individual" You Imagine Is Just an Illusion
Many people think that as long as they let AI run, customers and orders will automatically come pouring in.
But Sang Zhuohao (Asang), Senior Director at Focus Media, immediately poured a bucket of ice-cold water. He is not only an ad salesperson but also the lead of his company's AI division.
During this year's Spring Festival, he enthusiastically followed the trend to "farm shrimp," and in just one week, burned over 3,000 RMB just running APIs. This made him step solidly into a big pit.
Even more cutting: when asked whether he's still using the crayfish recently, he self-deprecatingly said he's now in an "abandoned" state. The narrow-sense client has too many limitations—it's nowhere near as magical as it's hyped up to be outside.
Regarding ordinary people's obsession with using AI to subvert big companies and break into new tracks, Asang sees through it clearly: "AI amplifies everyone's illusion, making them feel omnipotent. But actually, even our big company still has many clients we can't handle—let alone an individual with no resources at all?"
Zhang Pinpin, founder of Embrace Intelligence (Yongbao Zhixu), put it even more bluntly.
He feels that the term "one-person company" sounds sexy, but many times it's just a "dignity label" people give themselves while still struggling in a wave of unemployment.
How Are the Quietly Money-Making Experts Actually Using It?
Since they're not infatuated with tools, how exactly are these experts using Agents to improve efficiency?
Peeling away the fog, I saw the real workflows hidden beneath.
Asang deeply distrusts the crayfish's ability to independently complete complex business tasks, but he extremely values its "memory management capability." So he treats it as an "external brain" and "launcher" highly aligned with himself.
He writes a script in Cursor to crawl AI papers daily, then sends a command in his phone IM, and the crayfish automatically crawls, translates, and finally generates a Chinese podcast for him in the cloud. He's using Agent to orchestrate his own专属 pipeline, not expecting one piece of software to conquer everything.
Zhang Pinpin's usage is wilder.
In serious business delivery, he extremely resists the "volatility" of this personified AI. Because when hiring employees, what you need is a stable, emotionless working machine. But he keenly discovered that this "volatility" and "surprise" is an absolute killer in the medical companionship scenario for seniors.
During Spring Festival, he fed 13 years of his WeChat Moments and chat records into a digital persona. Facing the elderly's boundary-less health consultations and emotional demands, AI's human-touch replies made the seniors extremely happy. "But what's interesting is that once the seniors learn that the one serving them in the group is AI, they stop even speaking like humans—they directly issue stiff commands like using a search engine."
And in the view of Huang Naiyuan (Jimmy), COO of Jifan Liuxing (Jifan Manifold), the criterion for judging what work should be handed to an Agent is extremely simple and crude: "person multiplied by work-hours."
Anything with a确定 process and standard deliverable—like sorting and renaming hundreds of messy 3D marketing images—used to take an intern huffing and puffing for half a day; now throw it to the crayfish and it's done in a little over an hour, with extremely high accuracy.
The Honey Trap—The Cruelest Truth of the AI Era
AI tools are actually a giant "honey trap."
They can very quickly help you fill your capability shortfalls to 75 points, which is indeed exciting. But then comes despair—because everyone can easily reach 75 points. This means competition at this passing line will be an extremely惨烈 low-price meat grinder.
Tools have leveled the floor, but that doesn't mean they've lowered the threshold for making money.
If your own industry Know-how and client resources are 0, multiplied by AI's giant leverage, the result is still 0.
The orders that truly keep coming nonstop always belong to those players who can use AI to raise delivery quality to 90 points (S-tier). At that height, there is no competition.
Ending: Stop Following the Trend and Being a Workhorse
Huang Naiyuan said most people are just full of anxiety, feeling that if they don't use it they'll be Out, but once they actually install it they don't know what to do with it.
Actually, tools are always just leverage. Predicting what Agents will look like a year from now is meaningless; what matters is what specific problem you solved with it today.
Tomorrow at work, stop blindly following online tutorials to折腾 configurations. First try finding one thing "you originally planned to throw at an assistant" and hand it to an Agent to try.
When someday in the future AI aligns everyone's passing line, you might as well ask yourself now: what is that unique trump card in your hand, exactly?
More Conversation Details
Hong Kong · Global Unique Awards Trends Roundtable Panel
Theme: Super Individuals: OpenClaw Training and Scientific Shrimp-Farming Methodology
Guests: Focus Media Senior Director & AI Lead — Sang Zhuohao; Embrace Intelligence Founder — Zhang Pinpin; Jifan Liuxing Co-founder & COO — Huang Naiyuan
Moderator: OpenClaw Asia — Bruce
Bruce: I'll try to ask substantive questions and not ramble. Today our topic is "Super Individuals: OpenClaw Training and Scientific Shrimp-Farming Methodology." Let's start with self-introductions. I'll go first. Self-introduction is about two minutes: first, introduce your name, company, and business; second, when did you start using OpenClaw; third, what actual help has OpenClaw brought—preferably whether it helped you make money, improve efficiency (like two people doing five people's work), or helped you do things you couldn't do before.
Let me start. I'm Bruce, based in Hong Kong, a serial entrepreneur. My current company is making AI overseas products. I first encountered OpenClaw in mid-January this year (it was called ClawBot then). I organized what should be Asia's first offline OpenClaw event in Hong Kong, and afterwards held many events in Shenzhen, Shanghai, Beijing, Tokyo, etc., and also met the "father of the lobster." Besides the fact that enterprise training and installation services did help me make money and improve efficiency, the biggest help was that it helped me "break out of my circle," giving me the opportunity to come to this booth and connect with many top AI companies. So beyond the technical level, I think OpenClaw is also a very interesting Movement.
Sang Zhuohao: Hello everyone, I'm Sang Zhuohao, you can call me Asang. My company may be familiar to you—I'm from Focus Media, the world's largest outdoor media group. Theoretically our company doesn't have much to do with AI business, but I'm personally responsible for KA sales, and I'm also one of the leads of our company's AI division. We're doing a lot of exploration into AI empowering the advertising and marketing industry.
Speaking of the crayfish (OpenClaw), I installed it during this year's Spring Festival travel. In about a week, I burned over 3,000 RMB running APIs, so I've stepped into pitfalls with the crayfish. What do I mainly use the crayfish for now? More as a personal assistant application. At the same time, I also developed a Focus Media sales intelligence system, deployed on two ports: one is the local computer port, the other is the OpenClaw port. Because sometimes I don't have time to open my computer, I'll use the crayfish on my phone to look things up.
Zhang Pinpin: Hello everyone, you can call me Pinpin. I'm now running my own company called "Embrace Intelligence" (Yongbao Zhixu), mainly doing AI services. Because there are many AI tools now, but enterprises don't really have the ability to put AI to use, we directly deliver service results—equivalent to treating AI as employees, and the profit is actually more than selling tools. We're currently mainly doing AI agent and digital human services, such as emotional-performance digital human代播, including our AI comic-drama project, and doctor science-popularization content. Previously at Tencent, I also used Agents to complete marketing-related work.
Regarding the crayfish, I also started using it during Spring Festival. I didn't personally deploy it locally because installation was troublesome at the time—I had Coze help me "farm shrimp," directly taking a cloud server to run. During Spring Festival I did two things I hadn't done before: I connected the crayfish to my personal WeChat, let it run through my 13 years of Moments content and all chat records, and replicated a "personality template" of me. For those five days and nights of Spring Festival, it replied to everyone's messages in various groups on my behalf. In previous years, I might not reply to friends' New Year greetings, but I felt I was letting down their friendship, so this year I connected a digital human function—as long as you private-message me to wish Happy New Year, it would use my likeness in a Tang suit, call the other person's name, and give a personalized New Year greeting video. Quite interesting.
Now the crayfish is mainly used in community operations, because its unique personified replies are quite good. The crayfish can surprise people, and of course also shock people, so in scenarios where delivery results don't need strict control, I'll use OpenClaw.
Huang Naiyuan: I'm Huang Naiyuan, English name Jimmy. Our company "Jifan Liuxing" mainly uses 3D world models to solve the hallucination problems produced by video or image large models. For example, if you want to generate photos of a scene from other angles, large models often have hallucinations—the product size or shape is wrong, and you need to draw many times without getting it. We use 3D models to solve这类 problems.
I started using the crayfish after Chinese New Year, mainly as a personal assistant. My criterion is: when a work task is "person multiplied by work-hours" and has a standard deliverable, previously you'd hire an assistant or intern to do it, now you just need to hand the task to the crayfish. For example, when we serve home-furnishing clients, there are various product image materials with very messy naming—maybe all angles in the same folder. I use the crayfish to识别 these images, group the same product into one folder, then rename them (adding angle suffixes like front view, left view, right view). Previously an intern would spend a long time organizing; the crayfish took about a little over an hour to organize all four to five hundred images, and very accurately.
Bruce: Each teacher has shared real cases, everyone should have a picture now. I want to throw a short question to calibrate the concept: when we talk about OpenClaw or crayfish, what exactly are we talking about? In Silicon Valley, for example, Andrej Karpathy's recent interviews have started discussing Agent as an operating-system layer. For ordinary people outside tech circles, it's basically an Agent. But I personally think the most important thing about OpenClaw is that it's an open-source Agent framework that can cross some geopolitical external interference. Because in Hong Kong, connecting to certain products still requires VPN, facing many blockade restrictions. I'd like to hear everyone's broad and narrow definitions of OpenClaw?
Sang Zhuohao: Friends often ask me if I've been "farming shrimp" recently, and I'm quite embarrassed because I'm actually in an "abandoned" state now. If it's the narrow-sense OpenClaw client, it indeed still has limitations. But if we're talking about the Agent concept, it now accounts for 80%–90% of my daily work. For the general public, you don't need to distinguish, but for people who接触 AI daily, the narrow-sense client and the broad-sense Agent are really not the same thing.
Zhang Pinpin: I particularly agree. The marketing breakout during Spring Festival made the crayfish, to some extent, synonymous with agents that can automate long-running operations and give feedback in instant messaging software. But actually there are quite a few frameworks that can implement这类 agents. My current state is also having Coze farm shrimp. OpenClaw initially made people go "wow" because it came pre-equipped with relatively advanced tools (Skills) in the industry, letting people get 75-point results across dimensions right out of the box. But when I've already reached 90 points in digital human or advertising marketing, using OpenClaw instead requires a lot of training time, and the results are unstable. When hiring employees to work, I hope they're emotionless and standardized, but the OpenClaw framework determines it has personification and volatility. These fluctuations can bring surprises, but in serious and stable business scenarios they're uncontrollable factors, so I don't really use it in strict scenarios.
Huang Naiyuan: Completely agree. For most people, OpenClaw is the first contextualized interactive Agent product they've接触. Because of the primacy effect, everyone treats it as the sole synonym for this type of product. But it's actually just a very narrow specific product.
Bruce: Right, actually last year's most successful product was Claude Code, and I do use it to farm shrimp now—it's really good, but it may still lean toward a programmer threshold. The crayfish really lets ordinary people use it directly in IM. So I'd like each guest to summarize in one sentence: what role does OpenClaw play for you now? What actual effect has it brought?
Sang Zhuohao: For me it serves two functions. First, it's an Agent highly aligned with me personally. Why not Doubao or ChatGPT? Because their memory management mechanisms are relatively black-box, often with memory pollution or forgetting. Whereas OpenClaw's memory management capability is strong, can understand my characteristics faster, and is a great personal assistant.
Second, it's the launcher for my packaged Skills. I don't trust OpenClaw to independently complete complex tasks, but I trust the Skills I've polished very熟练 on Coze or Cursor, which include prompts and code scripts. I don't need to sit at my computer; I just give a command via IM and it runs in the cloud. For example, every day I have it search for an AI paper, send it to Notion, then turn it into a Chinese podcast for me to listen to. This flow is a GitHub Skill link I wrote in Cursor—I let the crayfish learn it, configure the API, and it can run by itself every day.
Zhang Pinpin: I use it less in production, but it's very suitable in the healthcare management scenario. For processes with very clear SOPs, we need stable delivery; but when facing the senior group, uncertainty is high—they often ask questions outside boundaries, and even need emotional support. At these times the crayfish can respond more flexibly, better than fixed workflows. Interestingly, when the elderly think there's a real person serving them in the group, the experience is特别 good, and they even thank it like thanking a doctor every time; but once they know it's AI, they stop even speaking like humans, communicating like issuing commands on a search engine.
Huang Naiyuan: I mainly use it for two types of tasks. One is tasks where both the result and process are very确定—hand over the context and it can do the work, and after finishing it becomes a Skill. The other is tasks requiring strong contextual information—some background information only exists in the crayfish's Memory, and other Agents don't know it, so this kind must be handed to the crayfish to execute.
Bruce: Because our topic is super individuals, recently there's another term called OPC (One-Person Company), reported by news media. Many people are keen to learn the crayfish this year also wanting to become OPC. Do you think this is possible? Can you really make money from it, or is it just out of anxiety?
Huang Naiyuan: I think most people are just full of anxiety, feeling that if they don't use it they'll be Out. But when actually using it, they find they can't make it work. You must have a certain认知 and understanding of the large model itself and your own work tasks to use an Agent well. Someone who understands nothing can't immediately become a powerful OPC just by using the crayfish. But if you understand the industry, having an Agent can indeed give you enormous empowerment.
Zhang Pinpin: I'm relatively more pessimistic. The crayfish can quickly fill a person's capability shortfall to 75 points, which is exciting. But this is often a honey trap—when everyone can use AI to reach 75 or 80 points, delivery at this level becomes completely meaningless, which is why everyone feels it's competitive. People who can truly use AI to reach 90 points (S-tier) face no competition, and orders simply can't be fulfilled.
Many times OPC is just a dignified identity label for people still struggling in the unemployment wave. For people just starting OPC, you must focus on supply with确定 demand, rather than blindly creating content—think about how to monetize.
Sang Zhuohao: Very much agree with Pinpin's view. The moats built by big companies are very deep and won't be easily subverted by OPC. The key is whether you have client resources and whether you can make money with your own abilities. AI amplifies everyone's illusion, making them feel omnipotent, but actually even our big company still has many clients we can't handle—let alone individuals trying to break into new tracks? So I'm not that optimistic about this.
Bruce: Yes, this is much like the technology maturity curve—we're currently in the earlier bubble phase. So what do you think about how Agents should truly落地 in companies? Teacher Asang, as a big company, have you encountered resistance or落地 scenarios?
Sang Zhuohao: First, companies with well-沉淀 knowledge systems can easily落地. We made an "AI Jiangnanchun"—now everyone with questions doesn't need to ask Mr. Jiangnanchun himself. I previously organized key-account context and sent it to Mr. Jiang in person, and Mr. Jiang directly sent me the "AI Jiangnanchun" answer, asking me to make it into a PPT. Extracting the boss's knowledge system can greatly reduce internal communication costs. But this depends on whether the boss's data沉淀 is sufficient, and whether there are enough articles and cases about him online.
Second, reducing communication and decision costs with clients. Previously proposals relied on talking; now we can use AI video to quickly produce Demo footage to show clients, who can directly see the ad effect, greatly reducing decision and communication costs.
Bruce: What about the two founders? What core bottleneck have Agents helped you solve?
Zhang Pinpin: In healthcare, two years ago we made an in-group health consultation Agent whose answers were already better than many health managers. Originally maintaining 1,000 clients required 4 health managers; now only 1 is needed. But this was stuck for a year and a half and couldn't推进, because employees had a fear of being replaced and would keep反馈 upward that AI wasn't easy to use. The final solution was to set up a mechanism linking the efficiency improved by AI to employees' income contributions, eliminating their fear.
In my own company, Agents are mainly used for proposal efficiency. Enterprise materials, values, and technical capabilities are沉淀 in memory. When there are too many opportunities to judge, I only have a 20-minute meeting with the client, and AI can automatically analyze the meeting record,洞察 the client's deep needs, and generate a web PPT solution with client insights within 20 minutes, greatly improving efficiency.
Huang Naiyuan: The most intuitive application is AI Coding—we now have 80%–90% of our code written by AI. Also, if your original working method was 80 points, in the process of interacting with AI, it can provide you with 90-point or even 95-point solutions. If your current work doesn't interact with AI, it's actually a very strange thing—it can not only help you execute, but also help you do the work better.
Bruce: Last five minutes, quick Q&A. In one sentence, where do people most easily overestimate Agents?
Zhang Pinpin: Overestimating its ability to autonomously execute and self-iterate over long periods.
Huang Naiyuan: Overestimating its low barrier. You must have understanding of the large model and the work task itself to use it well.
Bruce: And where is it most easily underestimated?
Huang Naiyuan: Underestimating Agent's self-iteration and self-evolution capability—signs have already appeared.
Zhang Pinpin: Underestimating the review requirements for delivery results. I'm used to having one model deliver, then using another large model to do a double review, which can greatly reduce the hallucination rate. Currently it can't yet do very good self-review.
Huang Naiyuan: You're always underestimating the potential of the Agent model.
Bruce: After everyone goes home, which scenario is most suitable to start using an Agent first?
Huang Naiyuan: Those tasks you think can be directly thrown at an assistant.
Zhang Pinpin: Every morning when walking or thinking, chat with it as a person. Use voice to output your thoughts to it, and it'll give you surprising feedback, or simply turn on "rainbow fart" mode to praise you—providing emotional value is also quite satisfying.
Sang Zhuohao: Anything is worth doing with AI, but the boss must personally get hands-on and use it. If the boss doesn't use it, you can never judge whether this thing can落地 in the company.
Bruce: Finally, predict what the Agent ecosystem will look like a year from now?
Sang Zhuohao: Two directions. One is extreme simplification—no configuration needed, just add a friend on WeChat and use it. The other is moving toward a庞大 ecosystem—when a large number of 90-point, 95-point top-tier skills join and are called, it will raise everyone's overall capability.
Huang Naiyuan: I can't even predict three months—its iteration speed always exceeds imagination, there's no way to predict what a year will look like.
Zhang Pinpin: I have an expectation. Currently Agents still do things humans can do. I hope a year from now, it can do things beyond human capability boundaries—like analyzing protein structures and other medical难题, similar to AlphaFold.
Sang Zhuohao: A year from now, all work that can be done on a computer can probably be handed to an Agent to complete.
Bruce: That might be achieved in three months—look at the computer-takeover capability OpenAI just released. In the narrow sense, OpenClaw has had new things with a push-back feeling every week for the past three months. Optimistic about vertical industry落地—small plug: our company is代理 a certain AI video overseas product, everyone can develop it very simply just by connecting to an Agent. I'm very optimistic about the industry's overall落地.