OPC and Firm Boundaries in the AI-Native Era

Fast Reading

• The so-called OPC (One Person Company) refers to a new organizational form that takes AI tools as the core productivity, where an individual or a micro-team of no more than 10 people independently completes the full-process business activities from creativity, R&D to delivery and operation. Its core structure can be summarized as "1 + Small N + AI": one core founder, a small number of external partners collaborating on demand, and an AI tool system deeply embedded in the workflow.
• OPC challenges the underlying logic of management. Coase's firm boundary is no longer "where management cost equals transaction cost", but "where human-AI collaboration cost equals transaction cost"; Williamson's asset specificity is pushed to the extreme — the founder's cognition and judgment become non-outsourceable human assets, while AI undertakes all organizational work around this specific asset; the property right theory proposed by Hart et al. points out that residual control rights return to the individual founder.
• AI has lowered the threshold for starting a business, but at the same time raised the threshold for success to a new height. When everyone uses the same tools and calls the same APIs, the speed of product homogenization is ten times faster than before. Chinese OPCs face four major risks: hollowing out after subsidy withdrawal, track homogenization, AI dependency trap, and "being good at making products but poor at selling them". Tools iterate and models get updated, but the accumulation curve of deep cognition is far steeper than the slope of the technology curve.

In early 2024, OpenAI CEO Sam Altman threw out a bold prediction in a public dialogue: In the AI era, it is possible for one person to build a unicorn company valued at $1 billion.

At that time, this statement sounded more like the typical Silicon Valley-style optimistic manifesto. However, just two years later, this prediction is rapidly becoming a reality.

In May 2026, Anthropic quietly released a 36-page PDF document on its official blog — "Founder Playbook: Building an AI-Native Startup". This practical guide was led by Dario Amodei, co-founder and CEO of Anthropic.

At the beginning of the document, the author put forward a clear judgment: the growth path of startups is being redefined. Earlier, Amodei publicly predicted that the probability of the first "one-person unicorn" emerging was as high as 70%-80%, and the time window would most likely be in 2026.

The so-called OPC (One Person Company) refers to a new organizational form that takes AI tools as the core productivity, where an individual or a micro-team of no more than 10 people independently completes the full-process business activities from creativity, R&D to delivery and operation.

It should be noted that OPC is not a legal concept. As early as 2005, China's Company Law established the legal status of one-person limited liability companies. The new Company Law implemented in July 2024 further relaxed relevant restrictions.

The OPC discussed in this article does not focus on the legal sense of "one-person company", but on a completely new entrepreneurial paradigm. It represents that the entrepreneurial model has evolved from "using AI to assist entrepreneurship" to "using AI to reconstruct entrepreneurship". The core structure of this new paradigm can be summarized as "1 + Small N + AI": one core founder, a small number of external partners collaborating on demand, and an AI tool system deeply embedded in the workflow.

Among them, the founder is responsible for judgment and decision-making; AI undertakes execution and scaling; external collaborators are flexibly mobilized according to project needs. Compared with the traditional corporate organizational logic of "recruiting people first, then doing things", this is a completely different operating model.

01 How One Person Forms a Functional Team

From 2024 to 2025, large language models, AI agents, and automated workflows collectively crossed the usability threshold. In the design field, there are Midjourney and Stable Diffusion; in the programming field, there are GitHub Copilot and Cursor; in the marketing field, there are various copywriting and SEO tools; in the customer service field, there are AI dialogue systems; in the financial field, there are intelligent bookkeeping and tax filing tools. With the combination of these tools, for the first time, a single person can complete the workload that used to require a 5-10 person team at an extremely low cost.

What OPC brings is not a linear efficiency improvement, but a reconstruction of the workflow. The first step of traditional entrepreneurship is to form a team, because core functions require different people to undertake. The OPC model replaces this step with building a tool stack: the founder defines task boundaries, AI agents execute specific links, and humans make judgments and corrections at key nodes.

By 2025, mainstream AI agent platforms have been able to handle multi-step, cross-tool workflow orchestration. For example, they can automatically complete the full DevOps cycle from demand analysis, code generation, testing to deployment; or the full-link marketing from content planning, generation, multi-platform distribution to data tracking.

In 2026, Honghub, a platform focused on OPC and AI entrepreneurship, proposed an industry-benchmark indicator — Human-AI Cost Ratio (HACR). Its calculation logic is based on the average wage of employees in urban units in the software and information technology service industry published by the National Bureau of Statistics. The method is to compare the labor wage required for a one-person company to use AI tools to complete the same amount of work with its actual AI expenditure. The results show that for every 1 yuan invested in AI cost by a one-person company, it equivalently replaces about 72 yuan in labor expenditure. Experimental studies independently conducted by Harvard Business School and the Massachusetts Institute of Technology in 2024-2025 calculated the cost-saving ratio of AI tools replacing tasks such as manual programming, document processing, and data analysis, and the resulting magnitude is highly consistent with the calculation of Honghub.

It needs to be emphasized that the value of HACR equal to 72 describes equivalent substitution, not complete substitution. AI tools have greatly reduced the labor demand at the execution level, but links such as architecture design, quality control, and customer relations still rely on human judgment. The comparison of cost structures is the most intuitive: a traditional 5-person entrepreneurial team has an annual operating cost of about 500,000 yuan in China's first-tier cities; the annual fixed cost of an AI-driven OPC project can be reduced to less than 20,000 yuan. An OPC entrepreneur can support the cold start of the project from 0 to 1 with 3-6 months of personal savings, while a traditional entrepreneurial team usually requires at least 6-12 months of trial operation and a round of angel financing.

The profile of OPC entrepreneurs is significantly different from the public imagination. A survey by Honghub shows that 75% of OPC entrepreneurs do not have a technical background. Their occupational distribution is: 26% in operations, 25% in technology, 17.5% in product management, 16.1% as industry experts, and 8.4% as content creators.

After AI tools lower the technical implementation threshold, the ability to understand the industry and discover business pain points is more valuable than knowing how to write code. More than 90% of OPC entrepreneurs have an initial investment of less than $500, and the startup cost only requires a laptop and a few months of AI subscription fees. Low cost does not mean low investment — 57% of OPC entrepreneurs collaborate with AI for more than 8 hours a day. Researchers define this expenditure as "digital office rent". Founders exchange time for AI capabilities, and maintain the operating efficiency of the system through continuous human-computer interaction.

02 Rewriting the Underlying Logic of Management

From the perspective of management research, Anthropic's Founder Playbook is not only a tool operation guide, but also a challenge to the theory of firm boundaries, organizational design, and property rights theory.

In 1937, economist Ronald Coase raised a classic question in *The Nature of the Firm*: Since the market can trade everything, why do we need firms? His answer was: because using the market has costs. Finding partners, negotiating prices, signing contracts, and worrying about the unreliability of the other party — these "transaction costs" add up to potentially higher than internal management costs. The meaning of the existence of a firm is to replace market bargaining with internal "command-obedience". But there are also costs within the firm: when there are more people, coordination, supervision, and meetings are required, which is "management cost". The boundary of the firm falls at the point where "internal management cost equals market transaction cost".

In 2026, this boundary condition seems to be being rewritten. AI agents realize seamless interconnection between tools and systems through the MCP (Model Context Protocol), reducing the collaboration transaction cost of "cross-system and cross-department" to almost zero. For one-person companies, the formula has been updated: The boundary of the firm is no longer "where management cost equals transaction cost", but "where human-AI collaboration cost equals transaction cost". This makes the extremely streamlined "single-person organization" not only feasible, but also far faster in response speed than traditional hierarchical firms.

However, this picture is not so simple from the more refined perspective of institutional economics.

In his two representative works in 1975 and 1985, Oliver Williamson took Coase's framework a big step forward. He pointed out that the reason why transaction costs drive activities into the firm is not that "the market is not cheap", but a special variable — asset specificity. When a transaction requires both parties to invest in specific assets that are difficult to repurpose (specific equipment, specific skills, long-term relationships), market governance will fail due to opportunism and lock-in effects, and must be undertaken by the firm's vertical integration. According to this standard, OPC not only does not go beyond the logic of asset specificity, but pushes it to the extreme: the core asset of an OPC is the founder's own cognition, judgment, and industry insight — this is a kind of human asset specificity that is almost inseparable, non-outsourceable, and cannot be copied by the market. In traditional theory, this extreme specificity should give rise to a huge hierarchical organization to protect and utilize it (imagine how hospitals, law firms, and consulting firms build organizations around a few experts). However, OPC gives the opposite answer: Let AI undertake all the organizational work around this specific asset. AI agents no longer act as "another employee", but as a completely new governance form: it allows highly specific human capital to be used on a large scale without being wrapped in a traditional organization.

The property right theory proposed by Oliver Hart, Sanford Grossman, and John Moore (1986, 1990) raises another question: In a world where contracts are necessarily incomplete, who owns residual control rights? The answer is: the party that owns the key assets. In traditional firms, "key assets" are usually physical equipment, patents, brands, or code bases, and residual control rights are held by capital providers accordingly. But in OPC, the list of "key assets" has been rewritten. Code can be generated by AI (and often has no clear ownership), computing power is rented on demand, tools are subscribed monthly, and even marketing content comes from models — the traditional sense of "physical assets" has almost been emptied. The truly irreplaceable assets are the founder's own understanding of the scenario, judgment of user pain points, and response to edge cases. This means that residual control rights return to the individual founder. It also explains why the core competitiveness in the OPC era has fundamentally shifted: when AI undertakes most of the code writing, process automation, and basic research work, the founder's "problem definition ability" and "depth of industry insight" become the only scarce resources.

The founder's role has also transformed accordingly. Historically, founders spent most of their time on execution: writing code, managing personnel, and handling daily operations. In AI-native startups, the founder's role is no longer a single executor, but more of an orchestrator of agents — these professional AI assistants can read files, run commands, execute code, and even browse the web. The founder's attention shifts to higher-level work, focusing on higher-order tasks: generating ideas and guiding the system to put these ideas into practice. Therefore, in 2026, a person without an engineering background can also build production-grade software; and an entrepreneur with excellent technical skills but lack of business experience can also use AI to complete market analysis, financial modeling, and presentation material production. The technical threshold is dropping rapidly, but the cognitive threshold is continuously rising.

03 Global OPC Landscape: Explosion and Differentiation

A report released by Carta at the end of 2025 shows that the proportion of solo founders in the United States has climbed from 23.7% in 2019 to 36.3% in the first half of 2025. The stock of non-employer enterprises in the United States has reached 29.8 million, with a combined annual income of about $1.7 trillion, equivalent to nearly 6% of U.S. GDP. Data from China is also noteworthy: the stock of one-person companies has exceeded 16 million, and the number of new registrations in the first half of 2025 increased by 47% year-on-year.

In Silicon Valley, the closest empirical evidence to the "one-person unicorn" concept has emerged. In April 2026, *The New York Times* verified the financial data of Medvi, a GLP-1 telehealth platform: 41-year-old founder Matthew Gallagher used only $20,000 and more than a dozen AI tools to achieve revenue of $401 million in 2025, with a net profit margin of 16.2%, equivalent to about $65 million in profit. The company has only two full-time employees — Gallagher and his brother Elliot. In 2026, Medvi's revenue is expected to reach $1.8 billion, with daily revenue exceeding $3 million. In contrast, listed company Hims & Hers achieved revenue of $2.4 billion in 2025, but employed 2,442 employees, with a net profit margin of only 5.5%. Medvi's revenue per capita is about 200 times that of Hims.

However, the Medvi case also reveals the structural vulnerability of the OPC model. In February 2026, the FDA sent an official warning letter to Medvi, pointing out misleading product statements on its website; its clinical infrastructure partner OpenLoop Health disclosed a cybersecurity breach in January 2026 that exposed 1.6 million patient records; Medvi's AI customer service chatbot once "hallucinated" drug prices and fictionalized non-existent product lines. These costs are exactly what critics call the "single point of failure" — the OPC founder becomes the ultimate responsible person for every lawsuit, every compliance crisis, and every AI hallucination.

At the same time, China's OPC ecosystem presents a unique feature of high policy density. In 2026, more than 20 cities intensively introduced special policies, forming an almost scrambling institutional supply competition: Guangdong issued the country's first provincial-level OPC special policy, putting forward the goal of "hundreds of communities, thousands of enterprises, tens of thousands of talents"; Hangzhou released the country's first administrative normative document through the market supervision system, launching 12 initiatives such as "multiple licenses at one address", "workstation registration", and "sandbox supervision", allowing entrepreneurs to "start a company at a single desk"; Longhua District in Shenzhen pioneered the "Entrepreneurship Peace of Mind Insurance", and so on.

04 Hidden Reefs and Paradoxes

The revenue distribution of OPCs presents an extremely pyramidal structure. About 50% are in the exploration stage with a monthly income of less than 7,000 yuan; about 25% have entered the growth stage with a monthly income of 10,000 to 50,000 yuan; about 15% have reached the profitability stage with a monthly income of 50,000 to 100,000 yuan; the top 10% have an annual income of 1 million to 5 million yuan or even higher. This distribution conveys a key message: the mortality rate of OPCs is not low, and a considerable number of the 50% of entrepreneurs in the exploration stage will exit within about a year. But the revenue density achieved by the top 10% — one person, no team expansion costs, annual revenue of more than one million yuan — is almost impossible to achieve in the traditional entrepreneurial framework.

AI has lowered the threshold for starting a business, but at the same time raised the threshold for success to a new height. When everyone uses the same AI tools, calls the same APIs, and generates code and content of the same quality, the speed of product homogenization is ten times faster than before. The cycle from the emergence of an idea to its replication may only be a few weeks. In this environment, technical capabilities themselves are no longer a moat. There is only one real barrier: deep cognition of a specific niche scenario, a profound understanding of the real pain points of a specific group of people, and knowing what role AI should play in that scenario, rather than blindly applying tool templates.

The development of China's OPCs will also encounter its unique problems. First, there is the risk of hollowing out after the subsidy tide recedes. The core attraction of the vast majority of communities is cheap space and subsidy funds, not real orders and reusable service systems. The decision-making logic for entrepreneurs to settle in is to save money rather than make money. Second, the business tracks are highly homogenized. More than 90% of the communities list AI application development as the core recruitment track,