The One-Person Unicorn: Building Lean Businesses with AI

The one-person unicorn company in the AI era Li Zhifei, founder of the Chinese AI company 出門問問 (Chumen Wenwen), ran an experiment to see how powerful AI coding had become and whether it could build a complex system. Li was deeply involved in the previous generation of AI. He earned a computer-science PhD at Johns Hopkins, specializing in machine translation and natural language processing (NLP). He initially led Google’s offline mobile-translation system and later returned to China to start a company. The result astonished him. In just two days, AI helped him build a new, feature-rich office-collaboration platform with 40,000 lines of code. At his most productive at Google, Li could write 300 lines of algorithmic code a day. He had not abandoned technology during his entrepreneurial years, but today’s AI left him deeply impressed. He said he was nothing compared with AI and that its coding ability now significantly exceeded that of ordinary programmers in the market. Afterward, he uploaded the code to GitHub for evaluation. His collaboration with AI drew enthusiastic responses. Some said such a complex product would require dozens of people and months of work. On learning that AI had produced it so quickly, they exclaimed, “This is crazy.” With AI, a professional can independently build in a very short time a complex system that would take an entire technical team months, and can perform tasks previously requiring people more intelligently and autonomously. Li’s small experiment offered preliminary evidence that one-person unicorns could emerge in the AI-agent era. “One-person unicorn” is a term newly coined by Sam Altman. He introduced the idea in a November 2024 interview with YC CEO Garry Tan: “One person plus 10,000 GPUs will build companies worth billions of dollars.” --- Exponential gains in AI-model performance allow individuals to replace traditional startup-team functions by accessing large-scale computing resources, such as a 10,000-GPU cluster, and automate the entire process of product development, operations, and growth. This new AI-era startup model will overturn the previous entrepreneurial paradigm. Under the one-person-unicorn model, entrepreneurs can bypass traditional corporate decision hierarchies and use AI to respond instantly to technological iterations, shortening decision and execution cycles. Traditional startups are constrained by quarterly or annual planning cycles. The cost structure of creating a unicorn will also change. As GPU-compute rental costs continue falling, one person can potentially operate a global service at very low marginal cost. ---- An entirely new entrepreneurial era Exactly when a one-person unicorn appears matters less than the fact that one person directing AI can write 100,000 lines of code in three days. That is no less revolutionary than humanity’s transition from horse-drawn carriages to steam trains. A new era has arrived. ---- More people only get in one another’s way; AI can maximize each person’s capabilities Zhu Zheqing previously led Meta AI’s applied reinforcement-learning team. In October 2024, he and three colleagues founded Pokee AI with the ambition of using AI agents to automate every workflow. Pokee AI’s name comes from the idea of a “little pocket,” referring to lightweight models with strong decision-making capabilities. Its technical approach is distinctive: reinforcement-learning (RL) models, rather than large language models (LLMs), form the agent’s decision-making core. In July 2025, Pokee AI raised $12 million in seed funding with a team of only six. Zhu said it would not expand headcount in the near term. As a native AI-era entrepreneur, he firmly believes that more people merely obstruct one another and that AI can fully maximize each person’s capabilities. AI-application startups characteristically have small teams: as few as four or five people, and at most a dozen or twenty. Even after fundraising, they no longer rush to expand as startups once did. Agent startups of various kinds may become the first AI-native organizations. In such organizations, a small number of human employees guide and supervise multiple AI agents carrying out complex tasks. One view expressed at a closed-door Sequoia Capital seminar was that existing hierarchical, static structures would give way to more flexible, adaptive networks of dynamic collaboration. This implies a shift in management: managers must learn not only to manage people but also to work with agents possessing some autonomy. Governance faces new challenges as well—understanding algorithmic uncertainty while establishing controllable, auditable mechanisms for human–AI accountability. ----- Entrepreneurship is becoming leaner The initial cost of an AI-agent startup is now genuinely low: a monthly subscription to an AI model can be enough. Starting a company is therefore no longer the high-risk undertaking it once was, requiring extensive hiring and heavy upfront spending without knowing for a long time whether the idea could work. After speaking with investors, many AI startup teams conclude that they do not currently need much funding. Building a small, excellent business is not particularly difficult at present: in B2B, identify the right use case, let AI handle implementation and iteration, and secure a few customers to generate stable revenue. This group of AI entrepreneurs is expanding, jokingly called the AI era’s “self-employed operators.” They have no ambition to become large and no desire to raise money. They quickly implement an idea; if it makes no money, they try another. If it does, that does not prevent them from looking for additional opportunities. The sharp decline in perceived startup costs reflects several technological innovations working together. In Sequoia Capital’s view, AI applications are beginning to take off because all the key conditions—compute, networks, data, distributed architectures, and talent—are now in place. Open-source technology, pay-as-you-go cloud services, and AI tools, such as the $20-a-month professional version of the AI coding tool Cursor, let entrepreneurs experiment at extremely low cost and concentrate on innovation. ------ Opportunities for innovation are everywhere. Yi Bo, founder of the AI company 宜創科技 (Yichuang Technology), says the core entrepreneurial task is to bring AI productivity to every industry. Every industry and every use case deserves to be rebuilt with AI. Entrepreneurship is becoming leaner, with more opportunities than ever. Some use AI to streamline legal-document generation for lawyers; others design virtual presenters to sell products in e-commerce; still others earn money from AI fortune-telling. Entrepreneurs are looking for gold wherever large companies consider the work beneath them or find it inconvenient. ----- The age of highly capable generalists Future startup organizations will be extremely flat and small. Those who remain will be highly capable generalists directing teams of AI agents. “Such people are easy to understand: generalists who can both define problems and solve them,” says Li Zhifei. He believes that people of mediocre ability, and even today’s popular specialists, will be replaced by AI. The future belongs to these highly capable individuals. The people leading one-person unicorns will invariably be multifaceted generalists: technically proficient, knowledgeable about products and markets, and able to integrate resources. Their rise will also transform traditional employment relationships, with freelancers and entrepreneurs becoming the majority of society. Of course, many people have encountered AI-driven layoffs before having a chance to become such individuals. Programmers developing AI—those closest to it and most familiar with the technology—may be the first it replaces. Chen Yiran, a professor in Duke University’s Department of Electrical and Computer Engineering, says employment prospects for computer-science students at ordinary US universities are deteriorating. Knowing a little programming is significantly less likely to secure a job at a large technology company than it once was. Following the experiment, Li did not wield the layoff axe, but began reorganizing product development and engineering at Chumen Wenwen, introducing the full-stack-engineer role. Previously, development involved seven or eight job categories, including front-end designers, back-end designers, and product managers. Now there are only two: a system designer with full responsibility for creative design and decisions, and a full-stack engineer responsible for directing AI’s engineering execution. Li says: “At a software company like ours, AI can already do 80% of the work. If we do not change, we will be disrupted by AI.” He once thought the technology-giant model had become impossible to disrupt because startups could not match its concentration of talent, technology, or capital. Now he sees an opening again. ---- In the United States, legal services, sales, and coding are among the hottest areas for agent startups. Chen’s impression is that US agent startups are embedding themselves more deeply in industry, extending into chip design and manufacturing-process optimization. China leans more toward consumer entertainment. “This relates to investment logic. Hard technology takes longer, and exit mechanisms are more complicated, so Chinese capital is more willing to invest in consumer markets. Many Chinese agent startups are busy expanding overseas because they believe foreign users are more willing to pay. Cal AI uses AI to identify food and record calories. Two of its three founding developers are high-school students aged just 17. It first launched in April last year, earned $1 million within four months, and subsequently averaged $1.12 million a month. ----- The agent-startup field is crowded. Who will become the next business giant? The key is speed. It used to be normal to spend six months developing a product. Now, after six months, you have already missed the moment. AI develops so rapidly and unpredictably that no one knows what it will look like in three months. A leap in a foundation model’s capabilities might swallow your entire product. “Catching the right window in AI’s development matters more than an isolated technical breakthrough. Entrepreneurs must watch structural changes, gateways to user traffic, and organizational innovation.” Today’s AI startups severely test a team’s understanding of customer needs and AI itself. He believes founders must be very clear about how their products use an AI-native approach and what problems they solve. Sequoia Capital partner Pat Grady advises AI entrepreneurs to determine whether they have produced genuine changes in user behavior, checking product usage, engagement, and retention. Simply accumulating data is meaningless; only a data flywheel that drives business metrics has value. Grady emphasizes that, as in the cloud and mobile eras, the companies that ultimately win will be AI-application-layer players that genuinely stay close to users and build use cases. #Caijing201515 ----- Robin: “Entrepreneurs are not the only ones who need to keep up with a new era. Venture capital professionals must also keep their finger on its pulse, or they will soon be left behind.”