Original · Unique Research · 2026-06-03
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. Registration, survival, cost, efficiency, market and financing figures are source or speaker claims, not independently audited findings. Company, personal and work titles are transliterated where official English forms remain unverified. The source is dated June 3, 2026.
AI Industry Observer
6 Million People Registered One-Person Companies—Fewer Than One in Ten May Survive
Is OPC a dividend for everyone, or an arsenal for the few?
"
In 2025, China saw 6 million newly registered one-person companies (OPCs), accounting for one-tenth of all enterprise registrations.
Social media is full of OPC success stories with six-figure monthly incomes, and full of slogans like "In the AI era, one person is an army."
But no one tells you the other set of numbers: of those 6 million, 90% may quietly disappear within the first year.
So is OPC a dividend for everyone, or an arsenal for the few?
At a recent roundtable, four founders with very different backgrounds gave their answers. Li Biao, founder of PainHunt and a 14-year engineering veteran, has agentized the entire pipeline from idea to launch by himself. Zeng Min (Dennis), CEO of WUI.AI, a Yale MBA and former TikTok product lead, runs a four-person team building custom video systems. Zhang Pinpin, founder of Yongbao Zhixu and a 20-year cross-industry veteran, has only two people on social security but can mobilize a flexible network of 150 people. Cheng Hui, chief scientist at ORBOT and a professor at Sun Yat-sen University, is exploring whether robotics hardware can also take the lightweight startup route.
01. Side A of the Lever: One Person Really Can Take Down a Team
What is the OPC dividend? Li Biao's answer is just two words: leverage.
The leverage he describes has two layers. The first is amplifying your professional ability. You already know code, product and marketing; in the past you had to do every step yourself, but now you can let Agents run the whole pipeline for you. The second layer is more interesting: amplifying areas you are interested in but not good at. In the past you were a pure engineer, miles away from customers; now AI can handle sales scripts, market analysis and even promotional materials for you.
Li Biao himself is the most extreme example of this leverage. His overseas product PainHunt was completed entirely by one person—from writing the first line of code to payment launch and promotion. When delivering To B projects in China, after the client discusses requirements, the entire chain of coding, design, testing, regression and deployment is run by Agents. He only needs to do one thing: chat with the client.
Managing 10 Agents and managing 1,000 makes no difference to him. Because what he does is define the system's Pipeline and its internal positive and negative feedback logic. Once defined, the system executes for him at scale.
This efficiency amplification is even more intuitive on the cost side. Zhang Pinpin did the math: in the past, hiring a technical co-founder—without painting pie-in-the-sky pictures—meant a monthly salary of at least 30,000 to 40,000 yuan. Now he opens two US$200 Claude plans and can cover a technical co-founder's workload. Rent is saved, headcount is saved, taxes are lower. One person plus AI pushes the ratio of productivity to cost structure to a number unimaginable in the past.
Zeng Min explained the economic principle behind this amplification from a more macro perspective. He cited a term called the "Jevons paradox": when the efficiency of using a resource increases dramatically, people assume consumption will decrease, but in fact total consumption surges instead. Because efficiency gains make scenarios that previously had no economic value suddenly viable.
Large models have lowered the cost of analysis and research by one to two orders of magnitude, AI generation has lowered the cost of content creation by one to two orders of magnitude, and Coding Agents have lowered the cost of software development by one to two orders of magnitude. With all three combined, a large number of market opportunities that in the past were too small, too fragmented and too uneconomical for anyone to touch have suddenly become cakes that an OPC can eat alone.
Big companies will not go after these cakes. They are too cumbersome, their decision chains too long, and their organizational DNA determines that they are only suited to eating big ones.
This raises a key question: the size and shape of the cake determine what kind of person can eat it.
02. Side B of the Lever: Those Who Cannot Define the Pipeline Will Be Flung Off by the Lever
The four guests at the roundtable share one common trait: they all have more than ten years of industry accumulation.
Li Biao worked at Megvii for nearly seven years, doing everything from raw data and engineering to algorithms and scenario delivery. Zeng Min spent more than a decade doing AI products at Meituan, Didi and TikTok, with a computer-science background plus a Yale MBA. Zhang Pinpin spent 20 years moving between 4A advertising agencies and Tencent, always keeping his own company. Cheng Hui is a professor at the School of Computer Science at Sun Yat-sen University and head of its robotics lab, joking that at school he is CEO, CTO, CFO and student-handling factotum all in one.
Their success in doing OPC with AI is like Michael Jordan wearing a good pair of running shoes. The shoes are indeed good, but what really works is Jordan's legs.
This brutal fact must be faced squarely: the one-person company in the AI era is not a shelter for ordinary people, but an arsenal for high-net-worth, compound-talented elites.
Li Biao says managing 10 Agents is the same as managing 1,000, but the prerequisite is that he has 14 years of muscle memory, knows how to decompose complex business into a Pipeline, and knows how to design the system's positive and negative feedback. For someone who cannot define a Pipeline, they do not even have the ability to send a high-quality instruction to an Agent.
Zeng Min says the core of OPC is a company: profitability does not depend entirely on individual creativity, but is led by a core person who completes the full cycle of requirements definition, MVP building, channel setup and commercial monetization. Today's Coding Agents already have very strong internal consistency—if you explain it clearly, they can build it for you. But "explaining it clearly" is precisely what AI cannot do. The Founder is responsible for aligning with the external world: ensuring that what is built has users and can make money.
The system's ultimate performance is limited by "the communication bandwidth between humans and Agents." Those with high bandwidth produce explosive output; those with low bandwidth, no matter how strong AI is, only have an expensive toy.
Ordinary people who lack industry know-how, lack commercial judgment, and have only entered the field because of layoffs or following the trend—their outcome after entering OPC is not to become super individuals, but to become all-weather, high-load, self-employed cyber-underclass handicraftsmen. AI has lowered the barrier to production, but at the same time has pulled the intensity of competition to an unprecedented level.
Someone might ask: if OPC is destined to only make small money and eat the crumbs big companies do not want, what is the essential difference from a traditional sole proprietorship?
Zhang Pinpin's answer is very down-to-earth. He says opportunities to get rich quick are rare, or they are all written in the criminal code—those are not under consideration. He advises people doing OPC to start from "certainty": on day one, either get a high-proportion advance payment, or make money as soon as you launch. Find niche essential needs that big giants overlook, such as embroidery pattern-making software or second-hand guitar appraisal. The transaction volume across the whole industry is not large, so giants will not compete, but if you have resources and expertise, it is certain money.
This is indeed "high-end street vending." But the biggest difference between this kind of vending and a traditional street stall is: close to 100% profit margin and near-zero marginal cost. No employees, no office, no labor disputes. A sole proprietorship with a clear ceiling but extremely high annual net profit has far higher survival certainty in the current cycle than a startup carrying tens of millions in financing and bleeding cash every day.
Whether investors invest in OPC no longer matters. OPC, this new species, does not need investors' money from day one. With infinite agility and terrifying cash-flow rates, it deconstructs the traditional myth of scale.
03. The Endgame of the Lever: Who Will Be the Last Person Replaced?
This is the most unsettling question of the entire roundtable.
Li Biao says he has "already become a factotum," with the entire chain agentized. Zeng Min says the bottleneck of the system is the communication bandwidth between humans and Agents.
So what if one day Agents evolve to the point where they no longer need humans to define the Pipeline? What if they can themselves capture requirements, write code, launch, collect money and self-iterate? Will the OPC founder himself become the last living mouth to be replaced?
The first half of this line of reasoning is optimistic. Methodology can be copied, but "the retina for finding the next pain point" cannot.
After Li Biao's PainHunt was productized, anyone can use it to capture data from 20 platforms and test products with probability models. This is like after Photoshop became widespread, everyone gained photo-retouching ability, but top designers did not lose their jobs. Tool homogenization only forces competition back to the ultimate origin: your sense of smell for real-world pain points, your intuition for catching a customer's emotional fluctuation at the dinner table—these are things an Agent cannot summarize from Reddit data.
Cheng Hui added another dimension from the hardware perspective. She says that for products with warmth like robots, the requirement for developers, beyond professional skills, has another keyword: interesting. Only interesting people can make interesting products, which raises the bar for people higher and more diverse.
This observation aligns with Zeng Min's classification of four OPC models. Zeng Min divides successfully operating OPCs into four types: independent developer, influencer-traffic, domain expert, and business-process re-architect. The core competitiveness of each does not lie in coding ability, but in a deep understanding of specific people and specific scenarios.
AI can write code, but it does not know what kind of property appraisal report a divorce lawyer in Texas needs. AI can make videos, but it does not know what kind of paper support a medical-science popularization account needs for a top-tier hospital doctor to be willing to lend their name to a collaboration. These judgment abilities are the last card in the OPC founder's hand.
But the second half of this line of reasoning is cold.
The Agent capability curve is rising steeply. The "alignment with the external world" that Founders are responsible for today—sooner or later large models will do it no worse than humans. On that day, what the OPC founder truly has left may be only one card: he is the only legal-person entity of debt and interest, and the only interface that can build carbon-based trust with human customers. When humans hand over their life savings or millions of dollars, they would rather believe a flesh-and-blood body that sweats, gets anxious and can bear legal responsibility. This irrational preference based on species instinct is, in the short term, a logical dead end for AI.
But the words "in the short term" are the most heart-palpitating part.
The endgame of OPC is a survivor elimination tournament. When Agents have competed all execution power to the limit, if a Founder cannot outrun AI's intellectual iteration speed, he will become the most obstructive part in the entire company. To be the helmsman of an OPC, you must ensure you are forever the variable that cannot be logicized by code.
The longer the lever, the more critical the weight of the person at the fulcrum. AI is frantically lengthening the lever arm; what you must do is not be grateful for the lever's existence, but nail yourself to the fulcrum and not get flung off.
Are you the person standing at the fulcrum, or the one about to be flung off?
More Details from the Conversation
Guests:
PainHunt Founder — Li Biao (Bill)
WUI.AI CEO — Zeng Min (Dennis)
Yongbao Zhixu Founder — Zhang Pinpin
ORBOT Chief Scientist — Cheng Hui
Moderator: Unique Research Partner — Xue Qian (Amber)
I. Guest Introductions and Team-Size Grilling
Xue Qian (Amber): The theme of this Panel is the OPC (One Person Company) dividend, and this is the session I have been most looking forward to today. In the AI industry, both domestically and internationally, everyone is talking about OPC. At the beginning of the year I saw a figure: in 2025, 6 million one-person companies were registered in China, accounting for one-tenth of all enterprise registrations.
The sources of the OPC explosion are very diverse: there are top technical talents leaving big companies who find that going solo yields richer returns, and there are individuals who have achieved results in the going-overseas track. Although some question whether its business model can scale, for now the OPC dividend far outweighs the challenges. Today we have invited four guests from different fields—some are OPCs themselves, some serve OPCs. Let us first ask everyone to introduce themselves, and be honest: how many people are on your team right now?
Li Biao (Bill): I will introduce myself in two parts. Personally, I have been in the IT industry for 14 years, and have been doing AI for seven or eight years. In the AI 1.0 era, before everyone was competing on Agents, I worked at Megvii for nearly 7 years, doing everything from raw data and engineering to algorithms and scenario delivery—a typical engineering background.
Currently I have two companies. One was established in Beijing in 2024, undertaking commercial delivery of AI Agents for domestic enterprise office, education and privatization scenarios. The other was launched in the United States last year, and this year's main product is called PainHunt. It is an overseas intelligence platform that has captured data from more than 20 platforms, mainly in North America, including Reddit, X and the App Store. Through a large matrix of Agents running the full process in the backend, it can analyze whether a certain overseas direction is worth pursuing, and release signals of "where users are holding money and looking for solutions." In the future we will also use it to incubate our own new products.
Xue Qian (Amber): So how many people does your company have now?
Li Biao (Bill): For the PainHunt product, it is actually just me. From idea, promotion, launch to payment closure, I handle everything alone. Now, with a strong engineering foundation and Agent matrix, domestic project delivery—from coding, design, testing to launch—can all be agentized to run the full process. I have basically become a factotum; I only need to chat with clients and take requirements, and I do not need to write the rest myself.
Zeng Min (Dennis): Hello everyone, I am Dennis. I mainly work as a product manager (PM) in the AI field, with an undergraduate degree in computer science and later an MBA from Yale. I have worked at Meituan, Laiye (RPA), Didi and WeChat, mainly using AI to improve process intelligence in complex scenarios. I left in 2024 to build a video production system.
We now mainly build customized video systems for OPCs or small teams. One category is creators: our Agents can capture and store the account's constraints and DNA, automatically translate requirements into intermediate states and generate videos. The other is a new attempt we are making: an "AI marketing employee" system for small companies. Let the Agent interact with users in advance, figure out the audience and purpose, and directly deliver a video digital employee who understands the business.
Xue Qian (Amber): How many people are on this product's team currently?
Zeng Min (Dennis): We have always been 4 people, and now the new project is mainly being explored by me. I have not written industrial-grade code since graduation, but I found that Claude can greatly improve efficiency. To make the business more agile, the team needs to become smaller. In vertical scenarios that big companies overlook, ideally one person can independently complete the closed loop of requirements identification, MVP building, delivery and collection. The fewer people needed to complete the closed loop, the better, so we are compressing the team.
Zhang Pinpin: I have 20 years of career experience, with the main line being digital media marketing—I have worked at 4A companies and Tencent, and have also been a film and television drama producer and artist incubator. Beyond the main line, I have also done restaurants and bed-and-breakfasts. Even when I was at big companies, as long as there was no compliance conflict, I have been running my own personal company for about 8 years.
Currently the core business is "enterprise services for AI-generated professional content." For example, we produce medical-science popularization videos, using medical popularization content supported by professional papers, combined with a doctor's digital human avatar to speak; we also collaborate with top photography experts to develop e-commerce product photos backed by expert knowledge. These are things general tools cannot do.
Our company has only 2 people paying social security. But I have calculated that the people the company can connect to for doing work follows "Dunbar's number" logic, within 150 people. The ratio of people doing work, flexible workers and sales channels is roughly 2:1—nearly 100 people can be assigned tasks to do work, and another 50 people are responsible for selling.
Cheng Hui: Hello everyone, I am Cheng Hui. ORBOT is a startup robotics company in the Guangzhou Higher Education Mega Center. In the traditional concept, robotics R&D investment is very heavy because it involves multi-disciplinary, complex software and hardware systems. With the help of AI tools and supply chains, I want to see whether robotics R&D can move toward smaller-scale agile development.
At the same time, my other identity is a professor at the School of Computer Science at Sun Yat-sen University and head of the robotics lab. At school we often joke that we are "research sole proprietors": building teams, supervising students, being CEO, CTO and CFO factotum all falls on the professor. From this perspective, I am also an OPC entrepreneur.
II. Defining the Dividend: The Greatest Convenience of the OPC Model
Xue Qian (Amber): Thank you. All four guests fit the characteristics of OPC very well: they can iterate very quickly from the first product to the second; even without writing code, they have extremely strong commercial acumen and resource-integration ability. Everyone is now talking about the OPC dividend. In your view, what exactly does the dividend refer to? And what is the biggest difference from the traditional model?
Li Biao (Bill): I often talk about a word called "leverage." I think the dividend at this stage is approximately equal to leverage. AI's amplification of ability manifests in two directions:
Amplifying your own profession: you know code, marketing or sales; in the past you needed to oversee the entire process, but now you can use AI to multiply the steps you know can deliver results.
Making up for shortcomings: if you are very interested but do not have time to research, you can use AI to handle it. In the past, technical people were too far from customers, and the biggest headache when starting a business was sales. Now you can use AI to handle many things you do not know.
Once powered on and connected to Tokens, it breaks through the leverage of time. For me, managing 10 Agents is the same as managing 1,000; I only need to define the workflow (Pipeline) and the system's positive and negative feedback. I become the "God" in the system—after defining it, let it scale you up. This is where the OPC dividend lies.
Zeng Min (Dennis): Following up on what Bill said, OPC is essentially about leverage. It needs to be clarified that super-creators like MediaStorm or MrBeast are super individuals, but not necessarily "one-person companies."
Large models have lowered the cost of analysis and research by one or two orders of magnitude, generative models have lowered the cost of content creation by one or two orders of magnitude, and Coding Agents have lowered development costs by one or two orders of magnitude. According to the "Jevons paradox," after costs fall, narrow, vertical scenarios that previously had no economic value become viable. Big companies are too cumbersome, their business models cannot cover these opportunities, and small teams can agilely eat their own piece of the cake.
The core of OPC is that the Founder is responsible for aligning with the external world (Align), ensuring that what is built has users and can make money—this is something AI cannot do. After alignment, sync the state to the Agent and let it implement. The system's final performance is limited by "the communication bandwidth between humans and Agents"; whoever does feedback well has extremely low trial-and-error costs and explosive output.
Zhang Pinpin: I think there are four levels:
Cost dividend: rent is eliminated, taxes are lower. In the past, hiring a technical co-founder cost at least 30,000 to 40,000 yuan a month; now buying two US$200 Claude plans yourself can handle his workload.
Talent resource dividend: in the past it was hard to collaborate with experts; now experts are liberated from organizations, and I can flexibly collaborate deeply with professional producers and salespeople.
Competition dividend: everyone is a new company, so opportunities are fairer. A few days ago I went to Shanghai for a pitch competition, and my opponent was a big company that had raised hundreds of millions in investment—but now I have a chance to compete at the same table, and clients are willing to trust a professional OPC.
Productization and scaling dividend: in the past, restaurants and bed-and-breakfasts could not scale. Under the traditional model, a product takes half a year from planning to development, and once it fails PMF it fails; now I directly enter three directions, build and launch the next day, and may directly have revenue.
Cheng Hui: With the help of AI capabilities, team size can indeed shrink a lot. In the past, senior architects now almost do not need to write code themselves, and R&D efficiency has greatly improved.
For intelligent hardware, if considered from the product-technology side, you need a very professional super individual. He does not need to build the whole machine—maybe just a sensor—but he must understand the professional direction, know how to effectively organize a small number of people to seek external expert collaboration, and needs strong organizational ability and insight.
Another opportunity is that robotics and AI used to be completely different tracks, but now they have merged into the new "embodied intelligence" track. How to conveniently deploy these systems and do remote operations as simply as installing software is also a good market opportunity for small teams.
III. Niche Choice: Hardware Opportunities and Going-Overseas Decisions
Xue Qian (Amber): Hardware robotics is itself a heavy and long track. Professor Cheng Hui, does OPC have an opportunity to find a unique niche in the hardware industry or robotics track? What kind of fusion-type OPC do you need as a Partner?
Cheng Hui: It depends on the use case. If used in industrial manufacturing, it is still heavy. But if used in the service industry to provide emotional value, the requirement for developers, beyond professional skills, is another one: to be "very interesting." Only interesting people can make interesting products, which raises the bar for human diversity higher.
Xue Qian (Amber): I see that your current robot design is very cute, and it also leans toward needing interesting people to make it, right?
Cheng Hui: Yes, we believe a robot is a partner with a soul and warmth. We consider the combination of technology and art; the product can be not only To B but also To C, and we hope that in the future robots can serve as partners to better connect people with each other.
Xue Qian (Amber): Understood. Let us ask Li Biao and Dennis—both of your products were directly targeting overseas from the start. Why did you make this choice? How did you think about finding opportunity points?
Li Biao (Bill): This still depends on specific circumstances. First, no matter how competitive the market is, some niche markets of a few million are still doable. Whether to go overseas depends on your own situation—for example, if you are in Shenzhen and have foreign-trade resources, do it in Shenzhen; if you understand human relationships in China, you can do well without understanding technology.
Why do we do overseas? First, because the domestic market is too competitive; second, because overseas users have better payment awareness and habits. As an entrepreneur, go where the money is, where the market is, and where you are good at what you do. Domestic To B is labor-intensive—it can make money but it is hard to make big money. We follow a probability model: the market is unpredictable, but I can increase probability through extremely large-scale product testing. PainHunt is what helps us test products in batches through AI, like SHEIN testing clothing—what works gets more investment, turning software into something like an industrial manufacturing model.
Xue Qian (Amber): Dennis, you previously had long experience at big companies and TikTok, and now WUI.AI is also a small team going overseas. Have you encountered any pitfalls as an OPC team going overseas? Anything to share?
Zeng Min (Dennis): Why go overseas is quite natural for me, because I have lived in the United States for five years and have certain comparative advantages. At the time there was a consensus that many dark-horse companies that succeeded in going overseas were actually founded by Chinese people—it was a kind of following the trend or taking a calculated risk.
But this is absolutely not the only way out. In the AI era, OPC still must obey basic business laws: where exactly is the demand? From what I have seen, successful OPC models basically fall into four types:
Independent developer (Hacker): one person launches dozens of low-barrier products (such as calorie tracking) at key time points to gain growth.
Influencer with built-in traffic: look at where your audience is, and create differentiation.
Domain expert: for example, practicing law overseas, specializing in producing a certain type of house report—it was hard to do in the past, but now you can develop one yourself with a Coding Agent.
Business-process re-architect: decompose customer requirements into workflows and deliver at low cost. Overseas payment ability is strong, but this requires you to be able to enter that market. The core is to do it based on your own comparative advantage.
Xue Qian (Amber): Are your current customers B-end or C-end?
Zeng Min (Dennis): We now mainly target long- and short-video creators on YouTube. YouTube has strong advertising monetization ability, so you just need to make good content. But this group of users has limited payment ability themselves, so now we want to combine Coding Agent with content generation to deliver "digital employees" for their specific marketing scenarios, so that pricing can rise from tens of dollars to hundreds or thousands of dollars—this is the new direction we are exploring.
IV. Cross-Boundary Project Initiation: Underlying Logic and Certainty
Xue Qian (Amber): Zhang Pinpin's projects are very cross-boundary. If you were to initiate a new product now, what would the underlying business logic look like?
Zhang Pinpin: My perspective may not cover all OPCs, but it roughly falls into three situations:
Genius level: rely on a good idea to build something and wait to be acquired by big capital. But now copying a product is too fast, and first-mover advantage is not enough for capital to invest—these get-rich-quick opportunities are rare and not under consideration.
Reality level (supporting yourself): earn money for daily life. I recommend making standardized products and thinking from the dimension of "certainty"—on day one, either get a high-proportion advance payment, or make money as soon as you launch. Go find "niche essential needs" that big giants overlook. For example, embroidery pattern-making software can be completely built with an AI Coder; second-hand guitar appraisal—the whole industry's transaction volume is not large, so giants will not compete. With resources and expertise, this is earning certain money.
Interest level: during this year's Spring Festival, I experienced "dreams coming true." I know absolutely nothing about Rust, but using Claude I stayed up for five days and five nights and wrote an input method that is only 8 megabytes; now I can also use AI to generate drawings for hardware and have factories produce them quickly. Under these circumstances, invest wherever your interest lies—experience and engineering methods can accumulate as models iterate. The most important thing is to start doing; only with feedback can you iterate.
Xue Qian (Amber): Thank you very much to the four guests for your sincere sharing. It feels like today's OPCs are a group of very capable and passionate people. That concludes today's Panel. Thank you everyone!