Strategic expert Zeng Ming: A lot of AIs are just doing tasks, not truly taking responsibility for the outcomes.

Imagine a scenario first.

You buy a car and drive it for five years. Over these five years, your driving skills become increasingly proficient, but the car remains the same one. It does not get better just because you drive it more. On the contrary, its parts wear out, power output declines, paint ages, and potential faults keep piling up. All the experience you have accumulated, such as how to brake on rainy days and how to take curves on mountain roads, stays in your mind and has nothing to do with this car.

For the past few hundred years, most of the world we are familiar with has worked this way: almost everything wears out the more you use it. Machines degrade, people get tired, and growing companies often become sluggish. In the words of Zeng Ming, founder of Zeng Ming Academy and former chief strategist of Alibaba Group, in the past, growth often meant consumption.

But now, a category of things works in the opposite way: the more you use it, the more capable it becomes.

In his new book *Intelligence: The Essence of Business, Organization, and Strategy in the AI Era*, Zeng Ming discusses exactly this transformation: as AI technology continues to develop at high speed, what fundamental changes will take place in the most critical fields of business, organization, and strategy?

This book is his concentrated reflection on the AI era following his previous works *Smart Business* and *Smart Strategy*. He attempts to deduce the underlying laws behind the future business world starting from first principles.

*Intelligence: The Essence of Business, Organization, and Strategy in the AI Era*

By Zeng Ming

June 2026

01 The Same Type of Car — Why Is Tesla Different

Let's go back to the car example.

The biggest difference between Tesla's autonomous driving and the ordinary car mentioned earlier is that the vehicle does not stop growing after being sold. Every mile each Tesla drives on real roads, including the complex road conditions, extreme weather, edge cases, and abnormal situations it encounters, can be converted into data, fed back to the system to train the model and optimize strategies. In other words, every time you drive the car, you are not just using it — you are also helping the system become smarter.

Zeng Ming calls this mechanism "intelligent compound interest." The more work a system does, the more scenarios it encounters, and the more feedback it obtains. This feedback continues to optimize the system and enhance its capabilities. The more capable the system becomes, the better it can handle more complex tasks, and the higher-quality feedback it can gain. Once this cycle starts running, capabilities will accumulate continuously just like compound interest.

In Zeng Ming's view, this is exactly one of the most critical underlying mechanisms in the AI era. The industrial era featured economies of scale, where larger scale meant lower costs; the internet era brought network effects, where more users made the network more valuable; in the AI era, the corresponding rule is intelligent compound interest: the more you use it, the stronger the system itself becomes. In the past, growth was often just an outcome. Now, growth itself can in turn reinforce capabilities.

02 "The More You Use It, the Stronger It Gets" Does Not Guarantee Success

However, the fact that intelligent compound interest can create value does not mean companies will definitely build competitive advantages. Zeng Ming specially reminds us that a system growing stronger and a company pulling ahead of its rivals are two completely different things.

For simple, standardized tasks that anyone can perform, all players will generally progress together. While you are getting stronger, your competitors are also growing, making it hard to widen the gap — competition will eventually return to the old path of price wars and subsidy races.

It is complex tasks that truly create a gap between competitors.

Complex tasks place great emphasis on long-accumulated experience. What pitfalls a system has fallen into, what mistakes it has made, and how it has weighed tradeoffs in difficult situations will all affect its subsequent performance. For a system, these are all experiences that cannot be directly replicated. When the capability gap becomes large enough to influence the decision of "who should be assigned this task," things will change: people will assign their most important and hardest tasks to the more reliable system. The more difficult the task, the more valuable the feedback it brings; the feedback in turn makes the system stronger; once it becomes more powerful, it will continue to attract more critical tasks.

Zeng Ming calls this snowballing advantage growth the "black hole effect": once a critical threshold is crossed, the leading advantage will keep expanding. You are not exactly competing against a single company — you are competing against the flow of the entire ecosystem. Since complex tasks rely heavily on experience accumulated step by step, latecomers, even if they obtain the same data, will find it extremely difficult to replicate the same capabilities without going through the exact same process. Therefore, in the AI era, taking a slight lead often means leading an entire stage.

03 Why Do Many Companies Not Become Stronger Even After Adopting AI?

Intelligent compound interest sounds very appealing, but when we return to reality, we find that there are not many systems that truly "grow stronger the more they are used." Many companies have already integrated AI, launched intelligent customer service, and implemented automated workflows, yet their organizational capabilities have not seen obvious improvement.

Where is the problem? Zeng Ming's answer is: much of the deployed AI is just performing tasks, not truly taking responsibility for outcomes.

Take AI customer service as an example. It usually performs well in handling standard inquiries such as order tracking, logistics status, and refund policies. But once it encounters rule conflicts, abnormal situations, or users communicating with strong emotions, the system is prone to errors. At this point, companies have to arrange human staff to monitor the process at all times, checking, correcting, or even taking over directly. It appears that AI is handling the tasks, but humans are the ones ultimately ensuring everything works properly.

Two problems arise accordingly. First, the scale cannot be expanded, as every task requires human supervision — the more tasks AI processes, the more people are needed to watch over it. More critically, AI cannot obtain complete and authentic feedback. What it sees is often results that have been filtered and modified by humans. It never gets the chance to fully understand how an error occurs, what the user's real reaction is, and how the final outcome comes to be. Once the closed loop of "action - outcome - feedback" is broken, intelligent compound interest cannot operate effectively.

Therefore, Zeng Ming believes that the truly critical turning point lies in whether AI can evolve from "humans backing it up" to "taking full responsibility for outcomes on its own." This is the hardest hurdle to overcome when building an intelligent system. Only when companies dare to delegate partial outcome accountability to the system can task scale be expanded; with expanded task scale, authentic feedback will continuously flow in; with that feedback, the flywheel of intelligent compound interest can truly start spinning.

This also reminds companies to re-evaluate their understanding of growth. Metrics like revenue, market share, and daily active users are shifting from being "causes" to being "outcomes." The truly important question is whether your system has formed a complete, self-sustaining closed loop of task execution.

04 From "Personalized Experiences for Thousands" to "Thousands of Experiences for One Person"

Zeng Ming divides the evolution of business into three stages. The industrial era was characterized by "one-size-fits-all" experiences, where supply was limited, and one product was sold to as many people as possible. Standardized goods and services such as Coca-Cola and McDonald's followed this logic. The internet era brought "personalized experiences for thousands," where platforms categorized users into different groups through recommendation systems, then pushed more relevant content, products, and services to them. In the AI era, Zeng Ming argues that business will further evolve toward "thousands of experiences for one person": business systems will no longer just judge "what type of person you are," but begin to understand "what exact problem you are trying to solve right now."

The key difference between categorizing users and focusing on their problems is that a person's behaviors are visible, but their true intentions are often hidden beneath the surface.

The book cites an example: a person repeatedly browses children's picture books. Based on behavioral data, the label is clear — this person is interested in children's books. But why are they browsing? They might be a new parent trying to cultivate their child's reading habits, someone selecting a gift for a friend's kid, or an entrepreneur researching the children's book market to start a business. The behaviors look almost identical, but the underlying problems they need to solve are completely different.

Traditional business models can only infer "roughly what type of person you are" and then push a bunch of picture books to you. For the first time, AI has the opportunity to truly understand the specific problem you need to solve through in-depth, repeated interactions. As stated in *Intelligence*, "whoever defines demand defines the market." When both demand and supply can be generated in real time, the focus of business competition will shift — it will no longer be about how many resources you own, but about whether you can organize resources to continuously meet real user needs.

05 A Deeper Transformation: Organizations and Strategy

The impact of AI on business models will eventually trigger two more fundamental transformations.

The first transformation relates to organizations. Zeng Ming raises a very practical question: why, even though AI has empowered every individual to become more capable, have organizations often failed to keep up, still relying on layers of meetings, repeated reporting, and cross-departmental wrangling? In his view, the corporate system we are familiar with today is essentially a product of the industrial era. Its goal is to allow a large group of people to efficiently execute predefined solutions. But compared to execution, what AI excels at most is continuously generating new solutions. It has also taken over many information sorting and coordination tasks that used to be handled by middle managers. As a result, the most critical capability of organizations has evolved from efficient execution to continuously forming new understandings. Future competition will very likely boil down to a contest of intelligence between different organizations.

The second transformation relates to strategy. Some companies that seem to have extremely clear strategies and thoroughly planned roadmaps are actually more likely to be caught off guard by sudden changes. The US online education platform Chegg is a typical example. It positioned itself in the student learning support market for a long time, with a clear user base and a stable subscription model — from a traditional strategic perspective, there were hardly any major flaws. But after generative AI emerged, the way students acquire knowledge underwent fundamental changes. Its original "content supply + subscription service" model lost its appeal in a very short period of time.

Chegg was not without a strategy, but its strategy was built on the premise that the way students acquire knowledge would not experience structural changes. Once that premise collapsed, even the most meticulously designed strategy would cease to be effective.

Conversely, many key development paths of OpenAI and Amazon AWS were not fully planned from the very beginning — they grew naturally through continuous action, feedback, and adjustment.

Therefore, Zeng Ming gives a new definition of strategy: strategy is not something you formulate first and then execute, but something generated through sustained action. Compared to having a good strategy in hand, truly outstanding companies are those that possess a mechanism capable of continuously generating good strategies.

06 We Are Entering More Than Just a New Technology Era

Beneath all these changes, Zeng Ming sees one unifying trend: the world is shifting from operating based on fixed structures to self-growing and self-creating through continuous generation.

From this, he draws a broader conclusion: AI represents the first time humanity is creating intelligence itself. All past business systems, organizational theories, and management practices implicitly assumed one premise: humans are the only thinking and creative entities in the world. The emergence of AI has for the first time shaken this foundational assumption.

Of course, Zeng Ming also reminds readers in the book that all these reflections are forward-looking deductions — the real world will always be more complex than any prediction. *Intelligence* does not provide a single definitive answer. Instead, it offers a set of reference coordinates to help us understand the chaotic changes unfolding before us. In an era where no one can accurately predict the future, understanding the underlying logic behind these changes is perhaps the very starting point for participating in the future.

As the final line of the book states: In the past, entrepreneurs searched for opportunities within an existing world; today, more and more people are beginning to participate in the generation of an entirely new world.

A Major New Work by Strategic Expert Zeng Ming

In-Depth Exploration of Business, Organization, and Strategy in the Intelligence Era

Grasp the Critical Trends of the Next Decade

This article is from the WeChat Official Account "CITIC Press", authored by A Xin, and published by 36Kr with authorization.