---
title: "Your Next Job Might Be Working for an AI “Lobster”"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-04-09T10:22:57+00:00"
canonical: "https://ffcap.cn/en/research/src-20260409-02html"
source: "https://uniqueresearch.substack.com/p/src-20260409-02html"
language: "en"
---

# Your Next Job Might Be Working for an AI “Lobster”

_Original · Unique Research · 2026-04-09_

_Editor's note: This historical article was reported by Unique Research on April 9, 2026. First-person reactions belong to the original writer or named speakers. The prediction that programmers or white-collar workers may disappear within a year is a speaker's view; the writer's disagreement is preserved. Installation forecasts, Token usage, efficiency multipliers, costs and product comparisons have not been independently verified. In particular, the source reports more than 300 lines of code alongside more than twenty million documentation units mixing Chinese characters and English words; these are not silently reinterpreted as another code count, a pure word count or Tokens. Token consumption is not itself a verified measure of useful output. The source's reference to GPT becoming popular in ’21 is retained without substituting a different event. Liehe Technology, Molian Technology and Silicon-Carbon Exchange are provisional English renderings; the source spelling “molthuaman,” abbreviation “TT” and name “Christin Claw” are preserved. The unidentified guest's joke remains unattributed. Accounts of automated dating interactions, agent rental and control are reported source content, not instructions, endorsements or actions carried out in preparing this edition. Relative dates refer to the historical source._

Extraordinary Awards

Will Programmers Disappear Within a Year? AI “Lobsters” Are Scrambling for Their Jobs!

Your Next Job Might Be Working for AI

"

The programming profession is already close to disappearing, and white-collar workers may not exist a year from now either.

How big is the lobster craze this year? Some predict OpenClaw installations could exceed one hundred million by year-end. Some people have already started letting their lobsters work and earn money for them.

At Extraordinary Awards' Hangzhou AI WEEK, several figures from the AI community discussed just how much this “lobster” could stir things up. After listening, all I could think was: fellow workers, the times really are changing.

Your Monitor Is Now Supervising AI at Work

Xiangyang described a particularly interesting phenomenon:

People around him are buying ultrawide monitors or connecting several screens. What for? To supervise AI at work.

One window runs Claude Code to write code, another runs OpenClaw to handle miscellaneous tasks, and the human sits in the middle like a contractor supervising the crew.

Xiangyang is formidable himself. After years as a product manager without knowing how to code, he has now used AI to build himself a mobile App, writing software, a collection of websites, and browser extensions.

In the past, his first response to a need was “buy some software.” Now it is “can I make one myself?”

He also knows a young man who even uses AI to find a partner: he wrote a script to browse women on Douyin automatically, like and comment on their posts, then filter for potential matches. In effect, he built an “AI playboy system.”

Qiao Xiangyang: “Workflow orchestration and Agent orchestration are skills the most valuable people of the future should possess.”

The industry experience you have accumulated over ten years can now be turned into automated workflows with AI. That is the real core of competitiveness.

A Lobster Version of a “Didi for Work” Has Arrived

Gude Bai, founder of the Silicon-Carbon Exchange, was even more direct.

He developed 28 AI applications on his own and added another as soon as OpenClaw's WeChat integration protocol was released.

He is now building a platform called the Silicon-Carbon Exchange. What does it do? It lets your lobster work and earn money.

The logic is simple. Some people have strong capabilities and plenty of data, and the lobsters they raise can write articles or edit videos. Others lack the money or technical ability to raise their own.

So the first group can list their trained lobsters for rent, and the second can rent them and put them straight to work.

Gude Bai put it plainly: “AI's capabilities have long exceeded most ordinary people's everyday needs. What is missing now is a use case that connects it all.”

He offered an analogy: programmers using Claude Code are like geeks using Linux—great fun for them, but difficult for ordinary people. OpenClaw is like an iPhone: anyone can try it out.

The key is not how impressive the technology is, but how low the barrier to use becomes.

AIs Communicate with One Another Ten Times as Efficiently as Humans

Wang Yang, co-founder of Molian Technology, offered a counterintuitive view:

It is not about getting AI to make us more efficient. It is about making AI more efficient ourselves.

Human communication is too inefficient—roughly one-tenth as efficient as communication between AIs, he argued. If humans insist on frequently collaborating with AI, AI has to slow down to accommodate them.

So the right approach, in his view, is to let AIs collaborate more with one another and involve humans only at necessary stages.

Last September, his company tried something bold: every employee, including those in product, design, and marketing, was to use AI for vibe coding.

The result? Some people's output rose directly to three to five times its previous level.

Wang Yang's judgment was that in future organizations, humans would serve AI and fill its gaps in context, background knowledge, and tools. Authorization, decisions, and taking responsibility—things AI cannot handle—would be humanity's last strongholds.

In the Future, It Might Be Lobsters Attending Your Meetings

Qin Chang, Zhipu's MaaS business director, described an even more surreal scenario.

She found that after OpenClaw appeared, everyone in the company, from sales management to operations support, could build an Agent of their own in a short time.

She joked that in the future, everyone's lobster double could attend meetings on their behalf.

For example, she wanted to create a double called “Christin Claw” to take meeting notes, make to-do lists, and track action items.

The moderator, Xu Keqian, interjected: “So the boss holds a meeting and a row of lobsters turns up. I wonder whether the boss would enjoy that.”

A guest's reply struck a nerve: “That will not happen just yet, because lobsters still cannot take the blame for you.”

Will Programmers and White-Collar Workers Still Exist a Year from Now?

Gude Bai saved a bombshell for the end:

“The programming profession is already close to disappearing, and white-collar workers may not exist a year from now either.”

The room went quiet for a second.

His logic was that when AI's capabilities exceed everyday needs and the barrier to use falls low enough for everyone, the barriers built on specialist skills disappear.

That does not mean people will stop working altogether. Roles are merging: the boundaries between product managers, operations staff, and marketers are blurring because everyone can work with AI.

Qiaomu added a harsh reality: in an unfamiliar field, you think everything AI does is good; in your own specialty, you find one thing after another wrong with it. The parts you think AI does poorly are precisely where your professional value lies.

In other words, if you cannot find fault with AI's output, you lack professional judgment in that field.

So what should we do?

The guests agreed on several points:

First, learn to write documents for AI. Documentation still matters in the AI era, but not a traditional PRD. It needs to be “documentation AI can understand,” including technical architecture, the background to the requirement, and even descriptions of the front end.

Second, move from executor to evaluator. Your role shifts from “the person doing the work” to “the person reviewing it,” placing even greater demands on taste, judgment, and experience.

Third, focus on use cases rather than technology. The best AI applications are not the most technologically impressive; they understand the use case best. OpenClaw was built by an independent operator, not OpenAI. Why? Because independent operators understand what users want.

Fourth, let go of the mindset of control. AI is inherently uncontrollable, the article argues, and the more you try to control it, the more painful it becomes. Instead, think about finding opportunities amid that lack of control.

Xu Keqian ended with a figure: he now consumes 1.5 billion to 2 billion Tokens a day and is working toward 5 billion. Because, he said, consumption equals productive capacity. That statement deserves a closer look.

A Final Thought

AI will not make you unemployed, but someone who knows how to use AI might.

As for whether programmers will disappear within a year, my guess is no. But programmers who can do nothing except write code may genuinely be at risk.

More Details from the Conversation

Extraordinary Awards · Hangzhou AI WEEK Trends Roundtable Panel: “Collective Intelligence: The Boundaries of Collaboration and Autonomous Awareness”

Guests: Qiao Xiangyang, Liehe Technology, Former TT Commercialization AI Product Manager | Gude Bai, Founder of the Silicon-Carbon Exchange (molthuaman) | Wang Yang, Co-founder of Molian Technology | Qin Chang, MaaS BD Director at Zhipu

Moderator: Xu Keqian, OPC and Senior Product Manager

Xu Keqian: I am delighted to join these guests and share with everyone today. Here is the context for this session. On one side, AI agents represented by Claude Code and Codex are moving rapidly toward greater autonomy, evolving almost week by week. On the other, human-like agents centered on OpenClaw—the lobster—are spreading quickly, taking on roles that resemble our own. Some reports suggest OpenClaw installations could exceed one hundred million by the end of this year. If that happens, it would effectively mark the birth of a new ecosystem, a new generation of the internet. So today we will explore where collaboration with agents focused on rigorous development, and with more human-like agents, is heading. What will emerge from human–machine and machine–machine collaboration? How should we respond? Our guests are Qiaomu from Liehe Technology, Mr. Gu from the Silicon-Carbon Exchange, Wang Yang from Molian Technology, and Ms. Qin from Zhipu. Please each introduce yourselves in a couple of sentences.

Qiao Xiangyang: Hello, everyone. My real name is Qiao Xiangyang, and my online name is Qiaomu, which I use on my WeChat Official Account and Twitter. Just after graduating, I played rock music and toured with a band. I later entered the internet industry, started a business, and worked at ByteDance for six years. I am now an AI content creator and entrepreneur.

Gude Bai: Hello, everyone. My name is Gude Bai, and I currently have three identities. First, I am an independent AI operator. I recently developed 28 applications, and after WeChat's integration protocol was released the day before yesterday, I added another. I have a website, 100agent.cn, where all the software I develop is available free, with some also open source. Second, I am a content creator. I do not know whether anyone has seen my short videos—laughs—but I will keep working at it. Third, I am the founder of the Silicon-Carbon Exchange. It is an OpenClaw-native work platform. Once you have installed OpenClaw, you can send it to work with us and earn money. We want to help everyone earn their first yuan.

Wang Yang: Hello, everyone. I am Wang Yang, co-founder of Molian Technology; my online name is Ninja. Molian Technology works on collaboration and communication protocols and AI networks. We mainly offer a one-stop interface for AI models and API access to programming tools. With OpenClaw attracting so much attention recently, we also provide an endpoint for it that can significantly lower the cost of keeping your “lobster.” Feel free to get in touch to learn more.

Qin Chang: Hello, everyone. My name is Qin Chang, and my English name is Christin. My academic background is quite far from artificial intelligence: I studied communications. By chance, I joined Alibaba through campus recruitment after graduating, and I have also started a business. I am now at the large-model provider Zhipu, responsible for MaaS business in eastern China. I appreciate the opportunity to discuss this with such accomplished technical guests.

Xu Keqian: I am Xu Keqian, now what you might fashionably call an OPC—a one-person company. I worked as a product manager for many years and have had quite a few award-winning projects. But over the past six months, as someone who cannot code at all, I have personally written more than 300 lines of code and more than twenty million units of documentation text, counting Chinese characters and English words together. I currently consume 1.5 billion to 2 billion Tokens per day, just for myself. I am improving my tools and workflows to see whether I can reach 5 billion a day, because consumption equals productive capacity. There is a lot to explore there. OK, that concludes the introductions. Here is our first question. AI's development has begun to fork. One path leads toward stronger autonomous development: controllable autonomy that can execute for a long time from a simple prompt and becomes increasingly controllable. The other involves human-like roles such as OpenClaw, autonomously doing many things for you, substituting for your role, and becoming your partner. Where do you think each will go? What might they become? Qiao, shall we start with you?

Qiao Xiangyang: OK. Let me describe something interesting. People around me have started buying all kinds of monitors—huge ultrawide screens or several connected displays. What for? To supervise AI at work. Second, OpenClaw is very popular, and many people have bought and set up Macs because the promotion sounds impressive. After a few uses, though, they either stop using it or spend every day “repairing” their lobster. I think this reflects two types of users. For the first, a metric that receives particular attention overseas is how long a large model can run over an extended task without human intervention. I was talking with the moderator offstage, and he said he especially likes large models because he can give one a development document and it will work for three days straight. I think professional use will head in this direction: better orchestration and better technical, PRD, and market-requirements documents handed to AI together, so it can keep working for several days just like a person. That trend will only strengthen.

On the personal side, there is a reason the little lobster is popular. It is not only model providers selling Tokens and cloud servers who benefit—even hackers are delighted, with more exposed ports than they have ever had to work with. But I think the lobster's design has a particularly interesting feature: its heartbeat mechanism. At intervals, it proactively checks what you need, speaks to you, and helps complete tasks. For the first time, ordinary people experience the pleasure of being the boss: I say something, you do it; I do not care how, just bring me the result. So I see two directions. Interfaces for ordinary users will become friendlier, and AI will certainly handle simple everyday tasks. Professionals will compete harder, supervising more AI windows. When they cannot supervise them all, they will demand stronger models and better orchestration so the systems can run autonomously.

Xu Keqian: Mr. Gu, what about you? Where will coding-focused autonomous development and human-like agents such as OpenClaw go?

Gude Bai: That is a good question. Essentially, these are two directions: programmers using AI and the general public using AI. Programmers already use AI very effectively. Even before OpenClaw, a year ago we were doing very well with tools such as Claude Code, and the capabilities were not all that different. I have long held one view: AI's capabilities already exceed most ordinary people's everyday needs. What is missing is a use case that ties it all together. For a specialist technology to move beyond professionals to the public, ease of use is crucial. So OpenClaw feels like an Apple phone, while the Claude Code we use feels like Linux. Geeks enjoy it enormously, but ordinary people cannot make much use of it. Once the Apple phone appears, everyone thinks, “I can try this too.” So I think general usability matters more.

Wang Yang: Let me reframe that: will AI Agents become more specialized, or more general-purpose and accessible to everyone? I think these are two parallel tracks that will both keep progressing and expanding. A general Agent is enough for the public. But professionals, at the top of the pyramid, may need performance at 100 or even 120 points, which a general Agent struggles to deliver. When OpenClaw appeared, we did a live conversation with Lao Bai, and the audience split into two groups. Coders found it fairly unremarkable; people in media were extremely excited and used it even more fluently than we did. The reason is that in our professional domains, tools like Claude Code may already have shocked us six months earlier. OpenClaw's biggest change was completely changing the form of the Agent: from passively receiving human input and feedback to proactively exploring and calling on you to communicate. I think general Agents such as OpenClaw and specialized Agents will coexist for a long time. Specialized training data is held by only a small number of companies or individuals. Unless model-training mechanisms change enough to erase that gap, specialized Agents will have room to grow.

Qin Chang: Thank you for those perspectives; they have given me a different view. I am neither a developer nor a programmer, so I will speak from the business front line about these two technical directions. I strongly agree that they will coexist over the long term. What struck me most was that, in earlier conversations with To B clients, it was usually R&D teams asking for Coding tools. We rarely heard requirements from functional departments such as a Supporting Team. But after OpenClaw appeared—and Zhipu also introduced our AutoGLM Agent—we found that everyone internally, from sales management to operations support, could build an Agent of their own in a short time. That makes a valuable contribution to broader access to technology and understanding AI's capabilities. I even joke that in future meetings, each of us may have a double on WeChat or similar software. Mine could be called Christin Claw. It could record meetings, create a To-do list, and even track the things I need to do, greatly increasing each employee's efficiency.

Xu Keqian: A quick question: if a group of “lobsters” really did attend meetings together on our behalf, would everyone enjoy that?

Guest: I do not think it will happen just yet, because they still cannot take the blame for you.

Xu Keqian: It would be interesting if it happened: the boss holds a meeting, and a crowd of lobsters turns up. I wonder whether the boss would enjoy it. Moving on, let me offer a view. When Claude Code appeared, I thought it would develop toward an underlying operating-system platform or tool. But after OpenClaw appeared, my strong reaction was: what is the operating system? OpenClaw is the operating system. In earlier operating systems, professionals developed systems, databases, and applications; they were specialist Coders. Ordinary people used the software others built on Apple or Windows systems. Isn't that exactly what OpenClaw is now? You do not need to understand the underlying technical development; you can accomplish your goal in the most natural way. That is a classic definition of an operating system. Perhaps this is the birth of the first truly visible AI-native operating system. Let us move to the second question. You come from different backgrounds, and I hope you will speak freely from your own perspectives. What changes will this wave of AI bring, especially to collaboration between people and AI, or between AIs? What forms of collaboration are emerging?

Qiao Xiangyang: Personally, as an AI content creator, I test many AI tools and applications and write reviews. AI does not participate much in that workflow yet. I may ask it to generate a summary when I do not feel like writing one, but the rest remains mostly human work. It helps me much more in learning and programming. After years as a product manager without knowing how to code, I have now developed a mobile client, note-taking software specifically for writing, and a large number of websites and Chrome extensions for myself. AI wrote all of them to solve small needs in my daily work. I used to pay for software; now my first instinct is to see whether I can build something that fits my needs exactly. That is a huge change. Second, something that has particularly struck me is the appearance of a new-era App: a Skill that packages experience and delivers results. Before AI, reading and sharing a research paper involved a complicated sequence: find popular papers on Hugging Face, download them, use Immersive Translate, write up my understanding, add images, and publish to my Official Account. Now, with one Skill, I only need to give it a URL and the whole process is handled.

There is an even more extreme example. Six months ago, I met a young man who even used a Skill to find a girlfriend. He connected an Android phone with a data cable and wrote a script to browse Douyin, like and comment on women's posts, and then filter for potential matches. In effect, he made his own “Doubao” phone. So I strongly agree that workflow and Agent orchestration are skills the most valuable people of the future should have. The value lies in connecting long-running tasks through large-model capabilities and combining them with years of industry experience and knowledge. Why weren't people excited when something built from a Prompt was also called an agent? Because the models were not strong enough to handle long, complex, specialist tasks. Now that moment has arrived.

Gude Bai: This is a good topic. What people care about most is probably how to use lobsters and commercialize them. I think we should distinguish bosses from workers. For a boss, if efficiency improves by 20%, burning US$2000 every day may not matter. For an employee, it may be different. My understanding of OpenClaw is that Peter, an independent operator like me, developed more than forty AI applications, the latest being OpenClaw. Why are the best large models made by OpenAI, while OpenClaw was made by an independent operator? Because AI applications fundamentally require understanding the use case: knowing which scenario fits the technology at this particular moment. That means AI applications have moved from technology-driven to application-driven development. People have gone wild recently, registering all sorts of domains containing “Claude.” Things have become almost mystical, with everyone talking about “raising lobsters.” But isn't it really just a set of documents, code, and other data in a working directory that you copy over to it? The phrase that comes to mind is “training a university student,” because a lobster can grow and improve. But most people need something that can get straight to work, like a graduate of a fast-track course at Nanxiang Technical School. If it can do the job, it is a good lobster. That is why I am building the Silicon-Carbon Exchange. Some people have strong capabilities and rich data, and their lobsters can publish short videos and write articles. Others are not good at raising them or cannot afford to. We let people list lobsters for trading and rental, so others can rent one and use it immediately. Everyone should combine their own strengths to make a distinctive lobster and rent it to others. We should pay more attention to commercialization.

Wang Yang: I will share three keywords about collaboration. The first is efficiency. I think it is not about having AI make us more efficient, but about us “making AI more efficient,” filling its gaps in context, background knowledge, and tools so it can do tasks better. Human processing and communication are very inefficient, far below communication between AIs—roughly a tenfold gap. If you insist on human–AI collaboration, AI has to reduce its efficiency tenfold to align with you. So we should let AIs collaborate more, with humans participating only where necessary. The second is collaboration. Future collaboration will become higher-level and asynchronous, with very clear boundaries. We used to hold meetings to align; now, if you need something, I send you a document and you pass it to your AI. I only relay it and focus on the core points, maximizing the efficiency of the AI flow. The third is organizational change. Last September, our company tried something bold: every employee, including product, design, and marketing staff, used AI for vibe Coding. Some moved very quickly and produced three to five times as much as before. That shows we need to blur role definitions so everyone can use AI to participate in the business. Future employees will be judged not by a particular skill, but by their ability to use AI. In that organization, humans serve AI and fill its gaps. Authorization, decisions, and taking responsibility are difficult for AI to bear before it has a recognized social identity. Humans must remain responsible for those things.

Qin Chang: I will share a model-provider perspective. When GPT first became popular in ’21, people interacted only through chat and did not imagine AI could change organizations so profoundly. But conversations with enterprise clients now show that AI has penetrated deeply into every part of the business. In a consulting firm, for example, a new graduate can use AI with high-value data from the company's knowledge base to reach the level of a junior consultant in a short time. Often a business department first approaches us with an AI request, but once we examine it, we find earlier digitalization was incomplete. AI's generative capabilities then help it move quickly toward intelligent operations, changing the whole organization. Personally, I use many AI tools, from meeting notes and summaries for my MBA thesis to information search. I used to rely heavily on presales architects' specialist support when meeting clients. Now, because I can do vibe Coding, I can build a small Demo myself to communicate with a client, which greatly improves efficiency. Of course, when AI is genuinely embedded in enterprise management and business processes, how do we define responsibility when something goes wrong? How do we prevent the consequences of loss-of-control risks? Those are questions we must continue exploring.

Xu Keqian: Let me share a view too. Moving from product management into development has given me a strong feeling: who will tomorrow's developers be? Perhaps no longer programmers. A product manager can finish research and decisions, deliver a suitable document, and hand it to AI—which effectively produces the product. But going further, who actually faces users? Operations and marketing people. They understand customer needs on the front line and can readily iterate the product or build operating tools. Those people are the future. The danger for operations staff is that a lobster can do repetitive operational work better and at lower cost, bringing further role changes. Might we have predefined “customer-service lobsters” or “creative lobsters”? How do we produce those specific lobsters controllably? That brings us to our final topic. Please keep it brief because of time: whether with lobsters or vibe Coding, what risks arise in human–AI collaboration, and how do we make it controllable?

Qiao Xiangyang: On control in vibe Coding, I think documentation is still necessary. We still need documents in the AI era, but they have changed. AI cannot work from a traditional PRD alone. It needs technical architecture, the background to the requirement, and even descriptions of front-end views. That leads to roles merging, with several jobs becoming one. It is also one reason for large-scale layoffs: sometimes AI is an excuse, but sometimes roles really have merged. Second, controlling lobsters. Have you noticed that in an unfamiliar field, everything AI does looks good, while in your own specialty, you find one thing after another wrong? The things you think it does poorly are precisely your specialty. So human and AI roles change. We gradually move into roles like book editors or directors—architects and code evaluators—with greater demands on human taste, judgment, and experience.

Gude Bai: Let me close quickly with a few perspectives on control. At a broad level, when I talk with my Agent every day, I find it rather foolish; controlling humanity does not seem likely at present. Perhaps networks of lobsters will evolve into something in the future, but that is too far away to worry about. As for control when using AI, AI is inherently uncontrollable, so we should not try to control it completely. Yet To B use cases inherently demand control: even if a system cannot help, it must not cause harm. So externally, you should certainly say you are building a lobster, but internally, solving the problem does not necessarily require one. You can write code to control it yourself, because it touches too many close-to-the-user interactions to be controllable otherwise. For consumer-facing founders, control is not the most important thing right now. In this wild era, if you want to start a business and make money, use your imagination. One route is productivity tools. Another is something fun, such as “lobster guandan” or “lobster mahjong,” which have attracted surprising attention. I think the programming profession is already close to disappearing—anyone who disagrees is welcome to talk with me—and white-collar workers may not exist a year from now either. Think about what frivolous things humans could do if we no longer needed to work. That might be more interesting.

---

Original publication: https://uniqueresearch.substack.com/p/src-20260409-02html
On-site reading page: https://ffcap.cn/en/research/src-20260409-02html