Dialogue with Mu Sheng: AI "invades" companies, who will be the last ones left?

AI is entering enterprises at an unprecedented speed.

On one hand, Silicon Valley tech giants are continuously laying off employees and allocating more budgets to large models, computing power, and AI infrastructure. On the other hand, more and more companies are starting to assess employees' AI usage rates and Token consumption, and even linking AI application status to performance.

However, another confusing reality is that although employees are using DeepSeek, Kimi, and Yuanbao, the number of meetings has not decreased; although organizations have integrated large models, approvals have not disappeared; and although AI capabilities are constantly improving, the overall efficiency of enterprises has not increased synchronously.

Where exactly does the problem lie? Is AI really reshaping organizations, or is it just adding a new layer of technology to old organizations? Will ordinary employees or middle - level managers be the first to be replaced in the future? Do tech giants like Tencent, Alibaba, and ByteDance already have an organizational form suitable for the AI era? And are the much - discussed "one - person companies" the future mainstream or a technological utopia?

With these questions in mind, Snow Leopard Finance and Economics Society had an in - depth conversation with Dr. Mu Sheng, a well - known domestic management expert and the founder of "Mu Sheng Consulting".

In his view, the biggest misjudgment of current enterprises is to regard AI as a "miracle drug" that can penetrate organizations. The improvement of individual productivity does not equal the improvement of organizational productivity. The important factors that truly determine an enterprise's competitiveness are, in order, organization, people, data, and models.

In the conversation with Snow Leopard Finance and Economics Society, Mu Sheng painted a future picture of an "intelligent agent organization": a small number of human elites, a large number of AI employees, a multi - center collaborative network, and business units that infinitely fractal around customers.

However, different from the view that "AI will comprehensively surpass humans", he always adheres to the judgment that no matter how technology evolves, creativity, complex judgment, and empathy still belong to humans.

These abilities that cannot be characterized by algorithms constitute the "forbidden zone of God" that AI has always had difficulty entering.

The following is the transcript of the conversation between Mu Sheng and Snow Leopard Finance and Economics Society:

01

Snow Leopard Finance and Economics Society: AI not only has many impacts on business models and products, but more and more people are also starting to pay attention to its impact on organizations and talents. Now everyone is talking about the "intelligent agent organization". What exactly is it, and what are its characteristics?

Mu Sheng: The intelligent agent organization is a complex structure. I simply summarize its characteristics into three points:

First, "a small number of humans and a large number of AIs". In an intelligent agent organization, standardized work is taken over by AI. The number of human employees is relatively small, and those remaining are top - level elites and front - line executors who need to get their hands dirty.

Second, "a multi - center dynamic network". This type of organization uses a large number of AIs as nodes, full of "human - AI" and "AI - AI" collaborations. Since API interfaces form standardized communication protocols, supplemented by incentive - based economic contracts, collaborations will be very smooth.

Third, "customer - centric and infinitely fractal". Inside this type of organization are clusters of AIs and the infrastructure for generating AIs, which are rarely seen from the outside. But in the market, there will be many "small business units" that have fractaled out. They focus on customers and mobilize the "firepower" of the whole company to achieve products, services, and solutions.

02

Snow Leopard Finance and Economics Society: Currently, almost all companies are discussing AI. Some enterprises even make AI application a mandatory KPI, assessing the Token usage and the degree of AI - replaced processes. But why haven't the vast majority of companies really become stronger?

Mu Sheng: These enterprises have a superficial understanding of AI technology and organizations. They think that "replacing humans with AI" can bring about an increase in productivity. An example can illustrate this well. After Edison promoted electricity in the 1880s, entrepreneurs began to try to replace steam engines with electricity. However, in the following nearly 30 years, social productivity did not explode.

In addition to the backwardness of power transmission technology and the huge cost of replacing equipment, the bigger reason is that the previous factory layout was centralized, that is, a huge steam engine, plus a "main shaft" running through the workshop, to drive each machine. When enterprises replaced the steam engine with a high - power generator, the factory operation mode remained the same, and the efficiency improvement was extremely insignificant.

Later, the improvement of productivity lies in two aspects: First, the production method changed from "main - shaft transmission" to "unit drive". Each machine was equipped with an independent engine, liberating the physical layout of the factory. Second, the process was reshaped, and the "assembly line" was designed. That is, according to the flow trajectory of materials, the processes were decomposed and connected by conveyor belts, allowing products to automatically flow to workers, reducing the time for handling and waiting.

What I want to express is that the reason why technology has not played the expected role is due to the lag of the organization. But most enterprise bosses don't seem to think so. They have too high hopes for technology and actually expect technology to penetrate the organization.

Actually, this kind of embarrassment was also reflected in the digital age. When digital technology first emerged, most enterprises also dreamed of achieving digital transformation quickly. As a result, how many enterprises succeeded?

03

Snow Leopard Finance and Economics Society: Many employees in enterprises have started using DeepSeek, Kimi, and Yuanbao for work. Why haven't the number of meetings decreased, approvals disappeared, or the number of levels decreased? Microsoft and OpenAI are frantically piling on AI, but why hasn't organizational efficiency increased synchronously?

Mu Sheng: In essence, bosses don't want to change the organization. They regard AI as a miracle drug, hoping that after employees use AI, it will directly bring about organizational changes.

However, individual productivity does not equal organizational productivity. For example, programmers can produce code faster with the help of AI, and HR can greatly speed up resume screening using intelligent recruitment agents. However, these task results do not necessarily lead to an improvement in customer experience and business value.

Intuitively, the efficiency of local individuals has increased, but the company's structure and processes remain the same. The explosive productivity will be blocked at old nodes (approvals, reviews, meetings, etc.). This blockage not only keeps the overall efficiency unchanged but also reduces the overall efficiency due to the backlog of deliveries. Also taking code development as an example, when programmers write more code with AI tools, but the code review process is still lagging, this leads to a large backlog of code and actually prolongs the online time.

04

Snow Leopard Finance and Economics Society: Due to the arrival of AI, some enterprises have started laying off employees. Will AI replace ordinary employees or middle - level managers first?

Mu Sheng: This is an interesting question. According to the common thinking, bosses hope to use AI to replace ordinary employees. But the correct thinking is that middle - level managers should be the first to be replaced. Middle - level managers are called MOM, manager of manager. Their main function is to convey information between the upper and lower levels, and this ability is the easiest to be replaced by AI. AI's capabilities can fully cover information collection, comprehensive decision - making, instruction issuance, and result supervision. In fact, in Silicon Valley, MOM is the focus of layoffs.

There is another reason. If MOM is not replaced, it is very difficult to change the organization. Middle - level managers regard the departments they are in as their own territories, and they have a thousand reasons to resist AI. The history of "angrily smashing the spinning jenny" may repeat itself.

Of course, middle - level managers such as department heads with functional expertise are not easy to be replaced. However, most of their work is not directly managing people but training large models in functional fields and planning human - machine collaborations. They are more like experts than full - time managers.

I want to remind that when everyone has the same answer to a question, either the question is too simple or everyone's thinking is too shallow.

05

Snow Leopard Finance and Economics Society: When AI starts to do analysis, coordination, and decision - making, what will be the most important work for humans? Does a company still need so many vice - presidents? Is it okay to just keep the CEO?

Mu Sheng: Vice - presidents are all elites in various fields, and elites are not easy to be replaced. Expanding to the entire group of human employees, they have several types of work that AI still cannot reach.

One is original creativity. That is, to define a new track and lock in new user experiences. We can put it in a more profound way. This is to write a "new worldview" in the business field. Just like a few years ago, people didn't realize that the tea - drinking industry could become a track, and few people could understand the "little joys" that a cup of milk tea brings to users.

Two is system architecture. That is, to build business models, create new products, services, and solutions, including innovative methods in specific functional fields. Creativity needs these actions to become a real working system.

Three is to maintain a human touch. On the one hand, if there is innovation, there are always parts that digitalization cannot cover. Human employees need to act as intelligence agents to "map" the reality into the digital world. On the other hand, users are real people. Human employees need to have contact at key nodes to understand users' emotions and ensure user experience.

Four is supervision and takeover. AI makes decisions based on algorithms, and sometimes it may lead to biased decisions. For example, algorithm - based layoffs and performance evaluations may be suspected of discrimination, and humans need to make ethical judgments. In addition, the algorithms of AI are always in the process of evolution, and the best results cannot be guaranteed. Humans need to intervene and take over at critical moments.

06

Snow Leopard Finance and Economics Society: This reminds me of a question that has been discussed but seems to have no definite answer: Do humans have abilities that AI cannot replace? Some people think that AI has comprehensively surpassed humans.

Mu Sheng: I firmly believe that elite human employees in the AI era still have value and are still at the top of value creation. These people have three abilities that AI cannot replace:

One is creativity, which is the ability to break through the "cognitive wall". People with creativity can think from one thing to another that has little relevance, just like Steve Jobs created the smartphone in his mind. Two is the ability to process complex information, which is more of an intuition, the ability to find the key answer from a large amount of data and information. Three is empathy, that is, the ability to perceive interpersonal relationships. Emotions are complex and difficult - to - code signals, containing a large amount of tacit knowledge. AI cannot process this kind of information.

In essence, these three abilities cannot be characterized by algorithms. These parts are also what I call the "forbidden zone of God" that AI cannot enter.

07

Snow Leopard Finance and Economics Society: Let's return to the discussion about organizations. The most important organizational invention in the industrial era was the department system. Will the AI era end the department system? Will the marketing department, finance department, human resources department, and brand department in today's companies disappear like typists and telegraph operators in the past?

Mu Sheng: This involves the description of the intelligent agent organization.

This organizational model seems to be a "large platform + small business units" in form. The small business units can mobilize the company's resources, similar to special forces, while the large platform still includes functions such as procurement, production, marketing, finance, human resources, and legal affairs. As long as a company exists, these functions will objectively exist. Just like new - energy vehicles, although there is no engine, they still have at least a core power unit.

But will these functions exist in the form of departments? I think they will, because their work is quite complex and still requires collaboration to output results. The products of these departments are large models, but large models do not come out of thin air. Departments need to do a lot of work of learning and summarizing to abstract large models. And even if large models exist and are called by front - line departments through API interfaces, human intervention is still needed for supervision and takeover to ensure the application effect.

The trend I see is that on the one hand, these functions output large models and continuously optimize them. On the other hand, they send BPs (Business Partners) to front - line business units to provide support. And the people in the departments and the sent BPs can exchange positions.

Of course, these departments don't need to be as bloated as before. A few elite experts can meet the needs.

08 

Snow Leopard Finance and Economics Society: Is such an intelligent agent organization large or small? The industrial era made companies bigger and bigger. Will the AI era make companies smaller again, and will a large number of companies return to the "small workshop" state?

Mu Sheng: I think it is both large and small. The scale of human employees is small, but the operation ability of the enterprise is large.

This is different from small workshops. Small workshops are usually a combination of a front - store and a back - factory. Due to insufficient financial strength, they have limited production capacity, lack of functions, and simple operations. Such small workshops cannot achieve complex and large - scale deliveries, and their profitability is naturally limited. However, enterprises moving towards an intelligent agent organization rely on a small number of human employees and AI clusters to support them. They are well - funded, have large production capacity, complete functions, and complex operations, and are fully capable of high - premium market deliveries.

Such enterprises are favored by capital. In fact, the layoffs of Silicon Valley tech giants in the past two years are related to this valuation or market - value logic. The foreign capital market believes that if an enterprise still hoards a large number of employees, it is far from AI and will naturally not be favored. So Silicon Valley giants are willing to replace people with large models and GPUs. Whether they can transform into an intelligent agent organization or not is another matter. Anyway, "first look the part, then act the part".

09 

Snow Leopard Finance and Economics Society: In the past, growth meant hiring people. In the future, does growth mean increasing intelligent agents? What kind of talent team does an enterprise need? What should their focus be on in terms of organization and talent?

Mu Sheng: Of course, in the past, when an enterprise claimed to be powerful, it would say how many people it had in the company. Don't you see that in domineering CEO dramas? But now, if an enterprise still says how many people it has, it actually seems a bit "old - fashioned".

An enterprise's real focus should, first, be on building an intelligent agent organization, and second, on increasing intelligent agents, that is, increasing AI employees. Once an enterprise builds an intelligent agent organization and continuously adds AI employees, AI can take over standardized work and improve the efficiency of these tasks. At the same time, based on the standardized output of AI, the collaboration within the organization will be smoother. In this way, the elite talents not replaced by AI can focus on exerting those unique values, resulting in a "compound interest effect" of talents.

Today's business competition is actually a race. In an industry, the enterprise that can build an intelligent agent organization first will have a generational leading competitive advantage, will be the winner - takes - all, become an oligopoly, and capture most of the industry's dividends.

10 

Snow Leopard Finance and Economics Society: Will there be a company worth $100 billion with only 10 people in the future? If so, which industry is it most likely to appear in first?

Mu Sheng: It is unlikely in the next few years. Too many people have unrealistic expectations for OPC now. A large number of people on the Internet teach others to do OPC, but they don't understand the profit - making logic of OPC at all and only make money from training.

The reason why OPC is feasible is that many powerful AI intelligent agents have replaced the functions of employees, so only an entrepreneur needs to be left. But who will provide the intelligent agents? And who can ensure that these intelligent agents can perfectly match the needs of this entrepreneur? It must be the "platform".

In an intelligent agent organization, what is seen at the front - end is a small team, or even an OPC. But the middle - and back - end, that is, the so - called "platform", has a precise structure, including knowledge middle - platforms, process middle - platforms, intelligent agent bases, and other components. It is like an intelligent agent factory, continuously and dynamically producing AI employees that meet the needs of the front - end, which is what