“Companies Will Disappear”: A Strategist’s Reasoning

Zeng Ming with Zhang Xiaojun · Episode 153

In September 2026, Zhang Xiaojun interviewed Zeng Ming, formerly Alibaba Group’s chief strategy officer and a participant in the founding of Cheung Kong Graduate School of Business and Hupan University. Among entrepreneurs, he is often known simply as the person who talks about strategy. The occasion was his new book, Intelligence: The Essence of Business, Organizations, and Strategy in the AI Era (an English rendering of the Chinese title).

A book prompted the conversation, but the discussion went well beyond it. Over more than two hours, Zeng offered 11 judgments, which he called non-consensus judgments.

The most striking sound like this:

“Jobs disappear, hierarchies disappear, and companies will eventually disappear. New forms of organization will emerge.”

“Anthropic and OpenAI will probably not be the major players or winners in the native-application stage.”

“For public infrastructure, only one business model works: oligopoly combined with strong government regulation.”

“Old platforms are unlikely to evolve into new platforms.”

“No one is safe in front of a giant wave. Whoever feels safe is the least safe.”

At first, these claims sound provocative, though not without reason. The more useful question is how Zeng reaches them, step by step.

This article examines the reasoning behind those judgments: how an interpretation of industrial history leads to a view of organizations in the AI era. If you are thinking about organizational transformation, treat it as a reading guide before listening to the full interview. Understanding how he interprets change matters more than remembering a handful of conclusions.

The discussion has four parts:

1. A historical view of industries: Where does Zeng think AI stands today?
2. From jobs to tasks: Why would changing the basic unit of an organization also change the company?
3. Generating strategy: How can people and AI train an organization’s “small model” together, allowing new insights to emerge?
4. The work matters more than the individual: Why might a genuine mission be necessary to move from being good to becoming exceptional?

We close with one practical exercise.

1. A historical view of industries: Where does AI stand?

People willing to make judgments this sweeping usually have a framework. Zeng built his by doing some painstaking homework: reviewing the PC internet, mobile internet, cloud computing, and the first, second, and third Industrial Revolutions. His conclusion is almost disarmingly simple: every major general-purpose technology passes through three stages.

First, society gradually accepts the technology, and it becomes infrastructure. Second, vast numbers of applications grow on that infrastructure, creating a period of experimentation and variety. Third, once enough applications exist, people recognize the need for a deeper foundation to run the new world. That is the native-application stage.

He uses the mobile internet as a reference. In his account, BlackBerry was an early product called a “smartphone” and held a dominant position for years. He cites a US market share still above 40% in 2010, against the iPhone’s 20-something percent. The reversal came after the iPhone 4. The iPhone, the App Store, and a new generation of native applications form the sequence in his explanation.

The mismatch between those stages is the important point. When he invested in mobile internet businesses, products such as Moji Weather quickly exceeded 100 million daily active users, he recalls. “But you soon discover that they do not represent the future.” In his terminology, native applications are the mainstream applications that address fundamental needs.

Applied to AI, his timing is explicit:

“2026 should mark the point at which the first stage is essentially complete.”

The marker is the token. Early in the year, he says, the term acquired broad recognition: Jensen Huang spoke of “token factories” at GTC, while Sam Altman described selling tokens as central to OpenAI. Zeng’s explanation is that a technology needs a standard unit of measurement before it can be used at scale and in standardized ways. We speak of a kilowatt-hour of electricity, a tonne of water, or a gigabyte of data. In this account, tokens make the use of intelligence measurable and priceable.

The same year, he argues, also marks the opening of the second stage. The wave around OpenClaw near Lunar New Year showed him “the limitless potential of an agent that can independently complete tasks.” He therefore identifies AI’s second stage as the agent stage: agents are the new applications of the AI era.

With this framework in place, the provocative opening claims become parts of a connected argument.

Why might leading model companies fail to become the eventual winners? In the infrastructure stage, he likens large models to oil refineries—and refineries rarely cross over into building cars. Why might old platforms fail to become new ones? Previous-generation platforms matched relatively simple information; AI-era platforms must match complex capabilities with complex needs. He sees a difference in the order of complexity. Why only one infrastructure business model? His argument is that public infrastructure tends toward oligopoly with strong government regulation, while the price of using intelligence settles at low levels, much like utilities.

These are structural judgments, rather than moral ones. He adds an even colder observation:

“Companies from the first stage struggle to survive into the second, and those from the second struggle to survive into the third. None of the previous era’s giants can confidently say it will make it across.”

Now the same framework turns toward organizations. That is the main focus of this article.

2. From jobs to tasks: Why would companies disappear?

The starting point is a different basic unit

“Companies will disappear” can sound like an emotional declaration. But Zeng’s explanation begins below the familiar debate about how many people AI will replace:

“The job is the company’s most basic operating unit. When we apply for work, we apply for a job. When a company wants to grow, it first defines a job, writes a job description, and recruits someone. Jobs bring reporting lines, hierarchies, and corresponding compensation. Everything is connected to the job. It is the basic unit of today’s organization.”

“In the AI era, however, the basic unit of the organization is the task: the thing that needs to be resolved.”

The point is not simply that jobs become meaningless. Jobs are the foundation of the entire personnel system. The job comes first, followed by its description, recruitment, reporting lines, hierarchy, and pay. Almost every management practice rests on that foundation.

This also helps explain why organizational reform often stalls. Changing appraisals, altering reporting lines, and merging departments can amount to repainting the same foundation. Replacing the basic unit changes the foundation itself.

What does an organization built around tasks look like? Zeng is specific:

“From day one, an AI-native organization will define the task, break it down, and allocate it. People and AI—silicon-based and carbon-based employees—will all collaborate around tasks.”

Why embedding AI in workflows is a transitional stage

One sentence in the interview deserves particular attention:

“People talk about AI having to be embedded in workflows. But that is a transitional product, because those workflows were designed for humans.”

When we talk about adding AI to a workflow, we assume the workflow already exists. People designed it first; AI becomes a new component inserted into one of its slots. If the basic unit is the task, however, the workflow is built around that task from the outset—with people and AI working together by design.

One approach adds a step to an existing process. The other starts with no inherited version of the process. The difference is far greater than simply using or not using AI.

Why hierarchy loses its rationale

Zeng’s explanation of hierarchy goes deeper than “greater efficiency means less management”:

“Hierarchy exists because we do not know who should have the final say, so we establish ranks and let the higher-ranking person decide. It is not based on judgment … in principle, whoever ranks higher gets the final say.”

In this reading, hierarchy is a substitute mechanism for reducing coordination costs when information is incomplete and judgment cannot readily be measured. It does not necessarily identify the person with the best judgment.

What replaces it?

“In the future, whoever has the stronger capability gets called upon.”

He offers a concrete picture:

“The major labs now publish tasks for people to claim. If you think you can do it, you take it on. If you can deliver the result, your work goes live. The company only needs to publish the tasks; you claim them.”

“Finding a job means finding a position” becomes “finding work means claiming a task.” That small linguistic change alters the premise of the incentive system: value depends on what you deliver rather than which box you occupy.

Context, not control: How does coordination work without commands?

If command stops being the main coordinating mechanism, something must take its place. Three ingredients appear throughout the interview, and they depend on one another.

First, high agency: motivation comes from within. Zeng puts it directly:

“At its core, it is self-direction. Why has ‘high agency’ become such a popular term in Silicon Valley over the past six months? It means taking charge of your own destiny. Nobody understands your growth better than you do. If you are growing and happy, keep going. If you are neither growing nor happy, move somewhere else.”

Second, contributions must be transparent. When tasks and results are visible, contributions become easier to identify:

“You took the task and delivered the result. That is very transparent, so you should receive the corresponding reward. Why are the transfer packages for top researchers so high? Because their contributions are transparent.”

Third, context must be shared. This is why “context, not control” has renewed relevance:

“Five years ago, when I discussed Netflix or even ByteDance, most people thought it was impressive but unrelated to them. Today everyone understands that you need to provide context. Give enough context, and it can make the right decisions.”

The alignment he describes concerns goals, judgments about the future, and the mental models behind those judgments, rather than prescribed actions. The dependency runs from shared context to transparent evaluation to sustainable self-direction. People need to understand the criteria before contributions can be assessed transparently; they need contributions to be visible for self-direction to work. Remove a layer and “self-motivation” risks becoming a slogan.

The culture changes with it. Zeng uses a sports-team analogy:

“Everyone is indispensable, but there are still stars, forwards, and defenders. The culture has to be open, transparent, and based on sharing.”

He also emphasizes co-creation: people recognize that they are creating something new together. That can substantially strengthen both their willingness and their ability to collaborate.

Two common misunderstandings

Misunderstanding one: The one-person company is the final destination. Zeng says no:

“A ‘one-person company’ means that many corporate functions have collapsed into one person. One person can now do what once required dozens. But what one person can do will always be limited. There will still have to be organization, division of labor, and cooperation.”

He treats the one-person company as transitional, but important. It reveals that a person using AI can achieve things that previously required an entire company and its management and coordination costs. His future picture is one in which each individual becomes a broadly capable “one-person company,” and those individuals then cooperate.

The deeper implication is that the smallest unit of capability has grown. What once took several people can now be done by one person with AI. When that unit changes, organizational forms need to be redesigned.

Misunderstanding two: Technology simply kills the company. Zeng describes the company as an accompanying institution:

“The company arose with the industrial era. Before it, there were workshops. The industrial era’s greatest institutional innovation was the company … It is a complementary institution. If the fundamental economics of the industrial era are challenged, its organizational form will also be challenged.”

This is his historical framing of the modern company. He connects it to a contemporary cultural observation: why have dismissive terms for older authority figures become more common? In his explanation, respect for age once rested on the value of experience and the continuity of established norms. Today, he argues:

“Experience that can be converted into knowledge no longer has value—it has been absorbed by large models. Smart young people who use AI well can instantly become experts in almost any field.”

From that premise, he asks what remains of the value of the previous 20 years of experience. The claim is uncomfortable, but his intention is to explain a mechanism. His advice to older people is blunt: either retire comfortably or become young again.

He is not sentimental about the disappearance of companies:

“There is nothing to regret. More interesting things will emerge. How many people are happy going to work at a company today?”

3. Generating strategy: How do people and AI train an organization’s “small model”?

This is the densest part of the argument.

Zeng proposes moving from “strategic planning” to “strategy generation.” The wording deliberately echoes image and video generation. The emphasis shifts toward building an environment capable of producing answers.

Organizational design therefore becomes the design of an environment in which insights can emerge:

“What we really need to design is a system, an environment, that allows strategic insights to emerge—a system in which strategy is continuously generated. Strategy eventually becomes a natural decision: at the right time, the right person makes the right decision, and the company moves forward.”

The organization’s small model: Insights emerge from collaboration

He then offers perhaps the most vivid metaphor of the interview:

“The future organization is a workflow, a flow of tasks—a task-oriented system. People and AI connect to it, and emergence is at its core. Just as with a large model, people and AI are effectively training an organization’s small model together.”

The metaphor brings several ideas together. Such a “model” needs context as input, real feedback to guide improvement, and repeated iteration. It cannot be completed in a single planning exercise. The test is whether it produces something new:

“Are insights emerging that nobody previously thought of? Is the group’s collective intelligence evolving? That is the ultimate goal.”

What, then, do humans contribute?

“Humans cannot process the full volume of information the way AI can. But people have moments of creativity and the ability to see connections that do not yet exist.”

These capabilities are not interchangeable. The goal of organizational design is to connect both to the same flow of tasks.

Look ten years ahead, think through three, act for one

Zeng describes the strategic method as “look ten years ahead, think through three years, and act for one year.” People often remember the ten-year view and the one-year action plan because both resemble familiar industrial-era habits. Yet he locates creativity in the middle:

“The real creativity lies in looking three years ahead.”

In his account of the industrial era, the ten-year picture was relatively predictable: annual growth of 5% or 8%, few new entrants, and a future that could largely be extrapolated. Current-year planning became linear projection. Strategy became routine enough for large companies to outsource much of it to consultants such as McKinsey.

Today, he argues, linear extrapolation cannot explain the 80-fold year-on-year growth he cites for Anthropic. Yet looking only ten years ahead still does not tell you where to begin. The task is to form a plausible three-year picture and identify two or three milestones, creating productive tension between the short, medium, and long term:

“Without the tension of a three-year view, there is no environment in which strategy can happen. Seeking an optimum in three-dimensional space is entirely different from seeking one on a two-dimensional plane.”

He gives a counterexample: an application startup with rapidly rising annual recurring revenue, or ARR. He asked what would happen after two more years of growth, what the industry’s eventual shape might be, whether ARR had a ceiling, and how soon it would arrive. The founders calculated that the ceiling was not high and moved quickly toward something new. His warning is that ARR can encourage an illusion of growth: it extrapolates today’s run rate over 12 months.

He is similarly direct about objectives and key results, or OKRs. Despite high expectations, they have often been abandoned or turned into modified KPIs. His judgment is:

“Only in the AI era may it become possible to make OKRs work.”

The reasoning returns to the organization’s small model. A system able to follow natural-language goals and evolving activity could notice drift and offer feedback: the organization is moving away from its OKRs, and more specific trends—even ones that are not numerical—suggest where attention is needed.

4. The work matters more than the individual: Why mission matters

The preceding arguments explain how an organization might be built. This one asks whether it can keep working over time. The answer turns toward people.

Zeng distinguishes two paths.

Being good involves a process of growth with continuing positive feedback, which fits how most people operate. Becoming exceptional is different:

“Greatness is recognized in hindsight. You overcome negative feedback again and again, when nobody agrees with you, and eventually a major success shows that you saw the future before others did.”

He observes that exceptional people often do not see themselves as fundamentally different. They credit other people and good fortune for their success. That might sound like politeness, but he argues it reflects something deeper:

“The work is bigger than the individual. Your ego has to be small to go that far. The era makes you. If all you want is to prove how impressive you are, eventually the world will show you that you are less impressive than you imagined.”

A large ego needs continual validation and therefore short-term feedback. It keeps a person inside a positive-feedback loop. The cost is reduced openness to reality: information gets filtered through prior beliefs before it can challenge them.

This connects directly to the earlier organizational argument. Once command lines recede, coordination relies on self-direction, transparency, and shared context. All three require people to accept that their own judgment may be wrong. Someone who depends on short-term validation may struggle to hear other perspectives, even when context is openly shared.

“Altruism, empathy, and a gradually diminishing ego are probably indispensable to becoming exceptional.”

Zeng identifies four drivers of organizations: opportunity, strategy, vision, and mission. Reaching the strategy-driven stage already makes an organization very good, he says. But a mission may be necessary for it to become exceptional. The difference is authenticity:

“The mission has to come from the heart—something you truly believe in and fully embrace, not something you invent to put on the wall and motivate other people.”

How do you tell where you stand? His final question is personal:

“What kind of person would you be satisfied to have become ten years from now?”

He finds that the question often stops people in their tracks. The answers reveal different motivations. “I care deeply about this work and want it to succeed” puts the work first. “I am curious” also points beyond self-validation. “I want to build a ten-billion-dollar company,” in his interpretation, is usually an expression of ego and a need for external recognition:

“They are strongly influenced by external measures rather than driven from within, so each funding round’s valuation can unsettle them.”

Bringing the reasoning together

The provocative claims at the beginning are different cross-sections of one argument:

The basic unit shifts from jobs to tasks → hierarchy loses its rationale → self-direction, transparency, and shared context take over coordination → the goal shifts from completing plans to allowing insights to emerge → sustained progress depends on organizations in which the work matters more than individual egos.

Very little of this argument concerns technical detail. Its underlying point is that when the basic unit changes, organization itself has to be redesigned.

The other opening claims—about model companies, infrastructure economics, and whether platforms can cross generations—require their own full industrial argument. Here, they serve as background.

One thing to try this week

Choose something your team is already working on and try describing only the task, without naming a job title.

Write one sentence stating the problem to be resolved. Then add two lines: what must be delivered, and what would count as acceptance. You may find that the acceptance criteria are the hardest part. They may also be the most important.

A question to take back to your organization

Perhaps Zeng’s strongest warning is this:

“No one is safe in front of a giant wave. Whoever feels safe is the least safe.”

If tasks really replace jobs as the basic unit, which task in your organization should be the first to be broken down and redesigned?

Sources and attribution

• This article is based on the official abridged transcript, Talking Industrial History with Zeng Ming: My Non-Consensus Judgments, published by the WeChat account “语言即世界 language is world” on September 7, 2026. It corresponds to episode 153 of Zhang Xiaojun Jùn | Business Interviews, released September 3, 2026, with a running time of 2 hours 34 minutes. The full podcast is the authoritative record. The book discussed is 曾鸣《智能：AI 时代的商业、组织与战略的本质》; its English title above is a descriptive translation, not an asserted official edition title.
• Judgments 1–5 provide background; the focus is on the organizational and workflow arguments in judgments 6, 7, and 8. The views and judgments belong to Zeng Ming; the questions belong to Zhang Xiaojun. Quotations here are our translations of the Chinese passages quoted in the source article, not an official English transcript. The remaining text is our retelling, ordering, and interpretation.
• This is an independent interpretation. There is no collaboration with or authorization from the guest, interviewer, or publisher. Illustrations were created in-house rather than taken from the interview or third parties.