The arrival of AI has made it very difficult for middle managers at big tech companies to get by.

Over the past decade, middle management has been a relatively stable position in large tech enterprises. Their routine work includes passing on information, coordinating resources, and tracking project progress. Acting as a critical link between senior leadership and frontline teams, they could hold their positions securely as long as they did not take the wrong side in internal factional conflicts and avoided major mistakes. 

Since the rise of AI, this position is no longer easy to "coast through". ByteDance has updated its leadership principles, emphasizing that managers must "return to the front line"; Tencent has piloted the project responsibility system, where management roles such as team leaders and directors flow along with projects; JD has cut two tiers of management, leaving the remaining personnel to oversee a much wider scope of work. Amazon has taken more direct measures: in its latest round of layoffs, over 78% of the disclosed eliminated positions are concentrated at levels L5 to L7. 

When tasks like relaying messages, supervising employees, and following formal procedures no longer generate value, how can personnel in the middle tier prove their worth? 

We talked to five current or former middle managers from large tech enterprises, three of whom have a background in technology or technical management, one has experience in consumer and internet industry management, and one works in product operation. 

Some of them have seen their team's output double, yet they are more exhausted than before, as all the time saved is spent on verification and taking ultimate responsibility for work. Some do not believe in "full-process AI transformation", but are forced to maintain those metrics that "look very AI-aligned". Some have calculated their own work, listing all the parts that can be taken over by AI and are being removed by the company one by one, so as to allocate their energy more reasonably. There are also people who watch their bosses bypass them and put more trust in their subordinates who are more adept at using AI. 

The five people are in different situations, but they share the same feeling: in the past, getting this position meant success, but now they have to prove their value every single day. 

01. The performance of middle managers now largely depends on how well they use AI 

Huoshan | Former middle and senior manager of Procter & Gamble, Shanda, Alibaba and Tencent systems 

I am a veteran middle manager with more than 20 years of experience. After graduating from Fudan University, I joined Ogilvy & Mather, and stepped into the middle management tier after switching to Procter & Gamble. Later, I was swept into Shanda by the internet boom, and then worked at Alibaba and Tencent, once leading a team of more than 200 people. 

Normally, with such a resume, I could wait for promotion as my tenure accumulates. But in the past two years, as AI has become increasingly mature and large tech enterprises have changed their assessment standards for middle managers, I clearly feel that people who "want to get into middle management", "coast through their middle management tenure" and "cannot get promoted beyond middle management" are all finding it harder to muddle along. 

Those who "want to get into middle management" used to rely on following the right people and simply obeying instructions. When your "boss" got promoted, you would take his position. In the past, as long as you could ensure that orders from superiors were delivered to subordinates, pass on your boss's tasks and track the progress, you would not make mistakes. Now, AI workflows have emerged: AI can break down tasks, track progress 24/7, and make timely adjustments, making the old working mode no longer feasible. 

Those who "coast through their middle management tenure" used to get promoted based on their length of service and seniority. Without solid backers, they would rely on being politically savvy or even "squeezing" their subordinates to keep their positions. Now that AI can even replace part of the work of frontline employees, if you cannot master AI, how can you "keep muddling along"? 

Those who "cannot get promoted beyond middle management", like me, always lack the last bit of momentum. Now with the emergence of AI, if I fail to respond properly, AI may become a shortcut for my junior subordinate (the lower-level manager) to overtake me, let alone get further promotion. 

At the same time, the salary structure is also changing. The salary of middle managers is not necessarily cut directly, but adjusted in another form. For example, the premise of getting a total of 100,000 yuan including basic salary, performance pay and bonus is to achieve a performance of 1 million yuan. Now with the introduction of AI, work efficiency and production capacity have been improved, so the required performance target has been raised to 1.5 million yuan. If you cannot meet the target, your performance pay and bonus will drop by a large proportion. You have to earn that 100,000 yuan by your own efforts. 

In my observation, there are three types of middle managers' "survival modes" in the AI wave. 

The first type is complacent and refuses to make progress. They are resistant to AI in their hearts, because their strengths lie in building relationships and leading teams. When companies widely deploy digital employees and AI-based standardized processes, their good at human-based management will become outdated. Most of these people are marginalized, under the nice name of "job rotation", which essentially means waiting to be eliminated. 

The second type has the desire to learn, but no motivation to take action. They have elderly family members to support and young children to raise, so the heavy family burden makes them want to learn. But outside work, they also need to spend time on family and interpersonal relationships, and dare not bet their limited time and energy on an uncertain new technology. They are the most anxious and distressed group: they neither dare to embrace changes, nor can they completely give up and lie flat, so they can only choose to be ostriches. 

The third type, like me, always maintains a sense of crisis and is used to embracing changes. I came into contact with AI three years ago, and have experienced technologies from large language models to digital humans, computer vision, and knowledge distillation. I was also one of the first non-technical background middle managers in the company to actively try to implement AI applications. 

But this does not mean I can rest easy. After all, AI is evolving extremely fast, most of the AI application knowledge I learned three years ago has been eliminated last year. 

Now many large tech companies are reducing headcount for middle management positions, which does not mean they do not need middle managers anymore, but that they no longer need "middle managers who cannot do practical work". Bosses require you to be hands-on, get directly involved in frontline work, understand business, master AI, and lead the team to create incremental value. 

But for middle managers, AI is actually an opportunity. 

For example, if a middle manager, young employees, subordinates and leaders all embrace AI at the same time, the middle manager has the background and connections that young people do not have, the resources and experience that subordinates do not have, and the practical operation experience that leaders do not have...... 

In essence, AI is just a tool. Whether it becomes an opportunity or a threat to replace you depends on whether you can keep up with the pace of development. 

02. AI has not threatened my position, but it has increased my workload 

Zeng Xiaojian | 31 years old, based in Singapore, AI Overseas Technical Lead 

I have served as a technical manager in the company for more than two years, mainly in charge of businesses related to large language models, multimodal AI and AI engineering. 

After AI became a global hit, the first change I felt was that the team's working pace accelerated. In the past, modifying a version of PPT, sorting out technical documents or writing patent materials usually took half a day to one day. Now that the superiors know AI can be used, their expectation becomes "deliver it immediately". The first draft that used to take half a day to finish now needs to be produced in more than ten minutes, and the team's output efficiency of documents, plans and codes once increased by more than twice. 

But problems soon emerged. During that period, I thought the team was in a "false prosperity period": materials, plans and codes were generated continuously, everyone seemed to be busier and faster, but the number of deliverables that could be launched or handed over to customers did not increase accordingly. There are three most typical situations: documents with complete structure and professional terms turn out to have wrong data after verification; codes that can run in the demonstration environment cannot stand the test in production scenarios; some people deliver the content generated by AI after simple sorting, without really understanding it. The time saved by AI in the initial stage is eventually spent on reworking. 

Superiors are pressing for speed, but the results delivered by subordinates need to be rechecked, and I am the one who has to take the final responsibility. Later, we added a double verification process: one person is responsible for generation and preliminary modification, and the other conducts independent inspection. For content related to customer delivery and key technical routes, we also need to carry out code review, automated testing and small-scale trial operation. The assessment standards have also changed: in the past, we mainly focused on how many tasks were completed, but now we also need to consider effective delivery, verification pass rate and business value. The ability to identify AI errors in time has become more important than generating content. 

But the biggest change is actually my own role. 

In the past, I spent most of my time splitting requirements, coordinating resources, conducting technical reviews and tracking project progress. When the superior gave a general direction, I would break it down into specific tasks and assign them to others, and my core responsibility was to manage the team well and deliver work on time. Now I also need to participate in a lot of frontline technical work. After assigning some tasks to subordinates, I find they are not familiar with the business or not proficient in using AI, and the results are even worse than if I directly call the model to generate the first draft. Many tasks now turn into that I first use AI to generate a relatively complete version, and then hand it over to the team for verification and iteration. 

In the past, I only checked whether the tasks were completed, but now I also need to judge whether the data source is reliable, whether the results are verifiable, and whether there are missing risks. The focus of management has shifted from assignment and tracking to judgment and taking final responsibility, and the frontline workload I undertake has increased by at least half compared with the past. 

The organizational structure is also changing. The company's business and orders are growing, and the number of team members is also increasing, but the number of management personnel has decreased. The original organization has been split into three smaller working groups, which no longer have multiple layers of leaders respectively, but are managed uniformly by one leader. One manager needs to cover a wider business scope and process more information. 

Seeing such changes, I sometimes wonder whether the middle management position is less stable than before. In the past, the company evaluated you based on how many people you led, how many meetings you held, and how many processes you promoted. Now bosses care more about why the work cannot be automated, why so many people are needed, and whether the plan can generate revenue or reduce costs. The living space for middle managers who can only assign tasks, track progress and forward information will definitely get smaller and smaller. 

AI will eliminate some management positions, but it will not eliminate management itself. Facing potential threats, what I can do is to keep learning, maintain an open mind, and continuously improve my cognition, judgment, business understanding, technical depth and AI application ability. 

03. The AI efficiency improvement required by my boss is sometimes achieved by my extra work 

Dong Ning | 35 years old, based in Beijing, middle manager and technical lead of a large internet enterprise 

I used to work at Baidu and reached the M3 level, and now I am a middle manager at a large internet enterprise in Beijing, focusing on technical fields. 

Although I am a middle manager, my daily work is more like splitting requirements and providing technical guidance. Most of my subordinates report to me in a dotted line mode, and I rarely do pure management work. 

Precisely because I am in a technical position, I feel the impact of this AI wave earlier and deeper than other middle managers. At first, I did not blindly believe in "full-process AI transformation", and did not require my subordinates to consume a certain number of tokens. I paid more attention to output: AI is just a tool, as long as the work is done well. 

At that time, if someone in the team spent 1000 yuan on tokens in a week, I would talk to him about what he used the tokens for and what results he got. 

Now the situation is reversed. If someone spends less than 2000 yuan on tokens in a week, I will have a talk with him. 

The pressure comes from the top leadership. The boss only wants results, and the results must match the narrative of "AI efficiency improvement". Some indicators do not directly reflect output, such as the proportion of code written by AI and the growth of token consumption, but you have to present these data in the report. 

My own workload has also increased a lot. Sometimes I have to do extra work to make the report look better. To be honest, I feel very conflicted: I clearly pay more attention to real output, but now I have to spend energy maintaining those metrics that "look very AI-aligned". 

Increasing workload is one aspect, and the more complex reporting relationship is another. The company has newly established AI-related departments, so more people report to me both in solid line and dotted line modes, and I also need to report to more superiors. The originally orderly process has become more chaotic in the AI era. 

In the past, people joked that middle managers are routers, but now I feel I am both a router and a server, with countless lines plugged into me. 

Despite the chaos, I do not think "de-middle management" can really eliminate all middle managers. This topic has been widely discussed in our circle recently: cutting job titles, compressing management tiers, and merging teams. But according to my observation, most of these measures are just cosmetic changes: the job title is removed, the person remains the same, and the reporting relationship does not change. 

After all, large tech enterprises cannot do without a hierarchical rank system. For an organization with tens of thousands of employees, management tiers are necessary to transmit information and share responsibilities. 

Of course, the actual work content has really changed. In the technical field, AI has been able to handle a considerable part of the requirements, and the output rhythm of subordinates has accelerated significantly. 

If you ask whether I am anxious, the answer is no.

I do not worry that "it is getting harder for middle managers to survive", nor do I worry that I will be replaced by AI. The reason is very simple: if the day of being eliminated really comes, no one can escape, no matter you are a junior employee, middle manager or senior leader, and the rank itself cannot bring you a sense of security. 

AI speeds up a lot of work, and also makes the problems that could be covered by