The AI Leaders Are Only Refineries, Not the Native-App Winners
AI has entered the agent stage: the leaders of stage one are only refineries, not the big winners of native applications; new platforms do not come from old platforms; organizations have to shift from jobs to tasks; and in the end what people compete on is the creativity to make something out of nothing.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Building agents today is like building websites in 1992
Zeng Ming divides the industrialization of a general-purpose technology into three stages: first it becomes social infrastructure, then applications explode, and finally native applications appear. In 2026 the token became the consensus unit of measure, which shows that the first stage is mature; the agent event around Chinese New Year (popularly known as the lobster) marks AI's formal entry into the agent stage. His analogy is the website-building movement of 1992: back then people built websites to share information, today they build agents to share capability. The browser has not appeared yet, and whoever can establish the standard for agents and seize the user's point of entry may be holding the biggest opportunity of the next two years.
— Zeng MingA ten-trillion-dollar company will appear, but not these two
Zeng Ming says plainly that OpenAI and Anthropic are remarkable technical and commercial achievements, but business history shows that the companies that break out in the first wave are most likely not the big winners of the native-application stage. Yahoo reached a market value of $120 billion five years after it was founded, and AOL's valuation reached $220 billion; in the end neither became the next era's Google. Model companies look more like AI cloud companies, refineries that produce base oil, while the real application value belongs to the chemical plants and the car companies — and cars were not built by the oil companies. He believes a ten-trillion-dollar company will certainly emerge, but not necessarily either of the two now in the lead. Model companies will end up as a mature business of oligopoly plus heavy government regulation.
— Zeng MingThe unit of an AI-native organization is the task, not the job
The company corresponds to the hierarchy of the industrial age; AI disrupts the industrial age, so the company as an institution will wither away. The basic unit of an AI-native organization is not the job but the task: people are organized not along reporting lines but by ‘what problem needs solving’, and both people and AI coordinate around tasks. Well-designed jobs and layers will therefore disappear, and traditional middle management will die out too; in the future a founder's first act is to define the organization's network of tasks. Silicon Valley's NewLab is an embryonic form of this new kind of organization, and OpenAI itself is more like a Lab than a traditional company. The culture of the new organization is more like a sports team, emphasizing transparency, sharing and co-creation rather than commands and performance reviews.
— Zeng MingGreatness can only be judged in hindsight; excellence is visible on the spot
Zeng Ming holds that ‘excellent is not the same as great’. Excellence is growth through continuous positive feedback, while greatness requires overcoming negative feedback again and again, holding on when nobody agrees with you, and finally proving through enormous success that you were the era's prophet — which is why ‘greatness is judged in hindsight’. The excellent very easily fall into a loop of proving themselves, which pushes their strategy toward the conservative. Great people have a small ego; they tend to credit their success to the era and to their partners, and they are driven by mission rather than by outside valuation. When sizing up a founder he asks, ‘ten years from now, what version of yourself would satisfy you’ — and if the answer is ‘to have built a $10 billion company’, that is mainly ego drive, not an internal mission. Genuinely great founders are usually altruistic and empathetic.
— Zeng MingStrategy in the AI era cannot be planned, only generated
In the AI era, strategy is not planned but generated. Zeng Ming proposes a ‘strategy generation system’: the organization builds an environment and a network in which strategic insight emerges naturally. Because decisions come more frequently and the quality bar is higher, a CEO cannot plan by linear extrapolation, and cannot outsource strategy to McKinsey either — McKinsey's rise was a product of the slow maturation of the industrial age. His method is ‘look ten years out, think three years out, execute one year’, and the core is ‘think three years out’: you need to see at least two or three milestones ahead, so that near-term action and the mid-term picture pull against each other. Building the organization means guaranteeing enough context rather than control, so that the right people make the right decisions at the right time.
— Zeng MingThe right analogy for robots is appliances, not automobiles
Zeng Ming holds that robotics is still in its period of strategic exploration, nowhere near convergence. Of the two current routes — build a generalized brain first, or close the loop on a scenario first — both are logically sound, and there is no way to judge in advance which moves faster. By historical analogy, the automobile industry had several thousand companies between 1900 and 1920, and only entered scale production once Ford built the assembly line in 1913; robotics has not yet had its Model T moment, and whoever first genuinely sells 10,000 robots will have done something remarkable. But the better analogy for robots is appliances rather than automobiles: after electricity was invented, countless categories appeared — refrigerators, washing machines, air conditioners — and robots interacting with the physical world will have N large scenarios, with household, companionship and industrial each capable of producing its own native giant.
— Zeng MingThe incumbent that feels safe is the least safe of all
Zeng Ming says that in front of a giant wave nobody is safe, and feeling safe is exactly what makes you least safe. In business history no era's enterprises have ever crossed smoothly into the next era; IBM and Microsoft are the only two possible exceptions, and both are full of uncertainty. AI is a productivity revolution that disrupts the industrial revolution, not a continuous innovation in the style of mobile internet, so the old giants' technology reserves and organizational cultures are both negative assets. Google can survive as an AI cloud company but will not necessarily win the consumer point of entry; ByteDance has an AI cloud opportunity, but Doubao is not the application of the future; Tencent's social relationships will be restructured; Alibaba is not naturally safe either. The new platform will most likely be created by new companies.
— Zeng MingHuman value lies in making something from nothing, not writing songs or painting
Borrowing Drucker's framework, Zeng Ming defines the AI era as the era of creativity: any knowledge work that can be handled in structured form will be taken over by AI, and people are on one hand forced and on the other hand finally liberated to develop new potential. That potential is called creativity — not writing songs or painting, but the ability to define complex problems originally, to make something out of nothing. AI has already taken all existing knowledge; human value lies in creating what does not yet exist. That is why he describes his own mission as ‘being a happy researcher’, and why he believes civilization's next step will reorganize itself around creativity, with even the education system forced to shift from ‘pouring in knowledge’ to ‘open exploration’, because young people already know the old path leads nowhere.
— Zeng MingIn their own words · checked verbatim
The most successful companies are always anti-consensus, but being anti-consensus does not necessarily mean success.
最成功的企业一定是反共识的,但反共识不一定成功。
Zeng Ming7:19
It may not survive to the end. It will very likely go on doing quite well, but it is most likely not a big player, not a big winner, in the native-application stage.
它不一定能活到最后。它很可能会往后活得不错,但它大概率不是原生应用阶段的大玩家,大赢家。
Zeng Ming27:49
Companies from the first stage rarely survive into the second stage, and companies from the second stage rarely survive into the third.
第一阶段的企业很难活到第二阶段,第二阶段的企业很难活到第三阶段。
Zeng Ming29:50
So companies will disappear, companies will die out. There is nothing to regret about that. Something more fun will come along later — how many people going to work at a company today are happy? They are not happy.
所以公司会消失,公司会消亡。这没有什么可以遗憾的呀。将来有更好玩的,你说多少人现在去公司上班是开心的?不开心。
Zeng Ming1:15:33
My technology is this good, I created this much value, so why might my company not be worth much? Because of a basic law of economics: as long as there is undifferentiated supply, you cannot capture high profits, and you are a company that isn't worth much.
我的技术这么牛,我创造了这么大的价值,为什么我公司可能不值钱?因为经济学的基本规律,只要有同质化的供给,你就不可能获取高额利润,你就是一家不怎么值钱的公司。
Zeng Ming1:28:01
Greatness is judged in hindsight, because you overcame negative feedback again and again, and when nobody agreed with you, you finally proved through enormous success that you were the earliest prophet of the era
卓越是事后论定的,是因为你克服了一次一次的负反馈,让所有人都不认可你的时候,你最后会以巨大的成功证明,你是时代最早的那个先知
Zeng Ming1:31:32
I have always felt that in front of a giant wave nobody is safe; whoever feels safe is the least safe of all
我从来觉得在巨浪面前没有人是安全的,谁觉得安全,谁就是最不安全的
Zeng Ming2:02:39
Figures
| K3 model parameter count | 2.8T | 35:58 |
| Years since OpenAI was founded (as of recording) | 11 years | 2:17:47 |
Glossary
- Token factory
- Treating AI compute as infrastructure that produces tokens, a sign that intelligence has entered a stage of standardized pricing.
- Agentic OS
- A system that dispatches every agent directly from intent to complete a task; Zeng Ming places it in the third stage and considers it unbuildable at present.
- New Lab
- A new organizational form for those dissatisfied with the company as an institution, staffed mainly by researchers; OpenAI is regarded as the earliest example.
- High agency
- A Silicon Valley buzzword for holding your own fate in your hands and seeking out tasks and growth yourself, replacing hierarchical command.
- Context not control
- A management principle: give people enough shared context and information that they make the right decisions on their own.
How to listen
CEOs currently setting AI strategy, executives at established incumbents hunting for a second curve, AI investors, and technical founders torn between building models and building applications.
The rapid-fire Q&A and the book list at the end are skippable (after 2:29).