Huang Renxun has about60executives who report directly. He has no second-in-command and no one-on-one meetings. He said:"

According to traditional management, the upper limit of the number of people a leader can effectively manage is3to8. Any more and it will be easy to get out of control. But Huang Renxun relied on this extreme flattening to reach the top three in global market value.

text="">7500employees have been significantly reduced to about 1,500people. The outside world generally predicted that the platform would collapse. As a result, Twitter (now X) not only survived, but product iterations continued.

These two cases reveal a tacit question in modern enterprises: What role does the middle layer in the pyramid structure play?

Middle layer: the relay station of information

The truth may be a bit cruel: the hierarchical management system we have used for two thousand years has never been the optimal solution for organizational evolution. It is just a helpless solution subject to human physiological defects.

Whether it is Qin's county system, the nested command system of the Roman army, or the Prussian army in 18The staff officers specially established during the reconstruction of the century, and the hierarchical design of all large-scale organizations in human history, essentially solve the same problem: how to accurately flow information among large-scale people.

The core function of middle managers is only one thing: to translate decisions from above into instructions that can be executed below, and to aggregate situations from below into information that can be understood by above. They are not tools of power, they are information protocols.

When Ren Zhengfei was in Huawei's early expansion, he spent a lot of energy building the business unit system and formulating collaboration processes. The same was true for the underlying logic——How to get tens of thousands of people to collaborate so that information is not distorted during transmission.

Looking back at Twitter's layoffs now, what Musk is doing is logically clear: He is verifying which positions are actually creating products and which positions are doing information delivery and internal coordination.

When Musk transcends these information relay stations and lets engineers face the code directly, the machine is still running. But Twitter’s approach was brutal. After large-scale layoffs, many really useful coordination functions also disappeared, and some problems with the platform were subsequently exposed.

This shows that simply laying off people cannot solve the fundamental problem, and an alternative mechanism is needed.

Three experiments by major Chinese manufacturers

Before the emergence of alternative mechanisms, fiercely competitive Chinese Internet companies had long been aware of the losses in the information flow process, and tried to use extreme systems to push human collaborative efficiency to the limit.

ByteDance is trying to break down information barriers with extremely high transparency. Feishu, an internal collaboration tool, gives employees enough background information, reduces orders and approvals, and reduces reliance on middle-level command. However, as the scale expanded, Byte later moved towards a divisional structure divided by business lines, and the hierarchy quietly returned. Pure information transparency is not enough.

Pinduoduo has streamlined to the extreme level and is strongly results-oriented, allowing Pinduoduo to achieve efficiency figures that are unrivaled in the e-commerce field. But the path is nearly impossible to replicate. It relies on extreme talent density and cultural pressure, a price most companies cannot afford.

Haier's Zhang Ruimin in2005Year 8Ten thousand employees were divided into4000Multiple micro-entrepreneurial units. Each unit is directly responsible for users, eliminating a large number of intermediate management positions. But 10More than 20,000 employees left during the year because they could not adapt to the highly self-driven model. The larger the scale, the higher the coordination cost. It depends purely on system reorganization, which reaches its peak.

These three paths confirm a reality: in purely human organizations, the pursuit of extreme flatness often faces the dilemma of loss of control or high pressure. Remove the middle layer and the information flow is easily broken. Relying on systems and culture to improve collaborative efficiency, what human mechanisms can do has basically reached its peak.

Manus:78Individual,1.25USD

Alternative mechanisms that can truly break this ceiling are emerging.

In 2025, a family started in China and later moved to SingaporeAICompany Manus gave the ultimate answer. They only rely on a team of 78 people to achieve success in 9Within a month, it achieved an annual revenue of approximately 1.25billion US dollars. 2025End of the year, MetaAcquired it for more than 20USD.

In Manus, there is no traditional approval process and no project manager to coordinate in the middle. Their system consists of multiple AIprograms working together——One is responsible for dismantling the task, one is responsible for execution, and the other is responsible for testing the results. AI completed all the coordination work independently, and this78Individuals focus on processing parts that are temporarily out of reach of the system: intuitive judgments, value boundaries, and the first step in a new scenario.

This is not a lucky startup. This is proof of where efficiency can go when a loophole is plugged. Small teams leverage huge commercial value, which is exactly the qualitative change in efficiency brought about by AI after taking over the coordination power.

AI: Take over the coordination layer infrastructure

ManusThis is not an exception, this is happening on a larger scale.

American financial technology company Block(owns well-known payment applications Square and Cash App) Jack·Dorsey (Jack Dorsey) made a shocking structural adjustment at the beginning of this year. The company has reduced its employees from over 10,000 to less than 6,000, mainly streamlining middle management, product manager and analyst positions.

The person in charge wrote in a letter to shareholders: This is to use smart tools and smaller, flatter teams to rebuild a new way of working. The market reacted positively to this, with the stock price rising sharply in a single day.

But what deserves more attention than these layoffs is what they are building. In order for AI to truly take over the middle layer, two things are needed.

One is a real-time picture inside the company. BlockIs remote working, decision-making, coding, design, all work leaves digital traces. By reading this data, the system can grasp the internal status of the entire company in real time: who is doing what, where things are stuck, and where the resources are. In the past, this required managers to meet to synchronize; now, AI automatically completes the alignment of internal states.

The other is the customer truth model. The core responsibility of traditional middle-level managers and product managers is to understand user needs and translate them into product directions. However, interviews and questionnaires are naturally biased, and what people say and what they do are often inconsistent. BlockBy processing millions of real transaction data, capital flows become behavioral facts. Changes in merchant cash flow, user migration, and consumption structure shifts can all be captured by the system and combined into decision-making inputs.

These two layers of capabilities are superimposed, and the information integration and decision-making coordination functions originally undertaken by the middle layer are completely replaced. The organization began to shrink into three types of roles:

In-depth experts, work directly in the complete context provided by the system; Special project leader, given a specific problem, have the right to mobilize all required resources within a limited time, and this role is disbanded after the problem is solved; Experts who lead people, not only do practical work, but also cultivate the people around you.

A large number of progress alignment meetings disappeared because the system has completed information synchronization under the hood. The long-standing form of middle management is beginning to be rewritten.

But the premise of all this is: whether the company has high enough density and real enough digital data to allow the system to construct these two"Reality Map". Without this foundation, AI is just a more efficient tool, not the organization itself.

The disappearance of the middle class is not because they don’t work hard

Huang Renxun used60Nvidia, which has 30,000 people, maintains the decision-making speed of startups; Twitter laid off80%The platform continues to operate as usual; text="">Leave half of the coordination work to the system;Manus Use78People have achieved the output of a medium-sized company.

These things are not coincidences, nor are they special cases of individual geniuses, but manifestations of the same trend in different places.

For two thousand years, the hierarchy has existed for only one reason: humans were the only information relay mechanism available at that time. This premise has now changed.

Middle managers disappear, not because they don’t work hard, but because of the thing they rely on for survival——Collect, aggregate, transmit, and translate information——has been made faster, more accurate and cheaper by the system.

Two thousand years have come to an end here.