HomeArticle

In the early AI era of organizations, new types of employees with management capabilities are required.

复旦《管理视野》2026-11-20 11:09
The power of AI in the hands of individuals has not been translated into organizational effectiveness.

Recently, the 2026 *Management Insight* Annual Forum was held at the Zhengli Campus of Fudan University School of Management. With the theme "Management in the AI Era: From Organizational Transformation to the Future of Humans", the forum gathered guests from academia and the industry to jointly discuss organizational transformation and management innovation against the backdrop of technological changes.

At the forum, Professor Chen Xiaoping, the Philip M. Condit Endowed Chair Professor at the Foster School of Business of the University of Washington and the academic editor-in-chief of *Management Insight*, delivered a keynote speech. In her view, we are now in the "primary AI era for organizations", and the scarcest resource in this era is not people who can write code, nor people who can use tools, but new types of employees with management capabilities. The following content is organized based on Professor Chen Xiaoping's speech at this forum and has been reviewed and confirmed by her.

Over the years, I have conducted research on organizational behavior and human resource management, with topics ranging from cross-cultural management, individual and group decision-making, all the way to leadership, management communication, and entrepreneurs' passion. I am used to observing people in organizations: how they make decisions, how they cooperate, and how they retain their judgment in the gaps of the system. However, in the past two or three years, I have found that all my old questions have been re-illuminated by a new variable — AI. This summer, after an in-depth conversation with a colleague who also does management research, and after reading a number of corporate cases, a judgment gradually became clear in my mind: We are in the "primary AI era for organizations", and the scarcest resource in this era is not people who can write code, nor people who can use tools, but new types of employees with management capabilities.

A Disturbing Gap

Let's first look at a set of figures. McKinsey's 2026 global enterprise survey report shows that nearly 90% of the surveyed enterprises have regularly used AI tools in at least one business function. This is an astonishing popularization speed. But the immediate question that follows is: After adopting AI, how much has the overall productivity of the organization improved? The median of the answers is only about 5% to 10%. At the same time, a number of studies in management have found that after individual employees are empowered by AI, their efficiency is generally improved by more than 30%.

30% improvement at the individual level only translates to 5% to 10% at the organizational level. This gap is the problem truly worthy of deep consideration. It shows that the power of AI in the hands of individuals has not been transformed into organizational efficiency. Where is the problem?

My colleague led a team to study the full-scale AI adoption of enterprises in the past two years, and put forward a framework that I fully agree with in his upcoming new book *Human-Machine Fit: The Four-State and Six-Dimensional Model of Organizational Leap in the AI Era* (Wang Xudan, Chen Jianwen, Wu Chuanjuan, China Machine Press, 2026): AI will go through four forms after entering an enterprise. The first is the tool state, where AI is partially called as an auxiliary tool to solve single-point problems, while processes, positions, and collaboration methods remain basically unchanged. This is the state that every individual user is experiencing. The second is the partner state, where AI begins to participate in work stably, and a relatively clear division of labor is formed between humans and AI. The third is the organ state, where AI is embedded in key business processes and becomes an essential node for operation and execution, and the positions and processes of the entire organization begin to be rewritten. The fourth is the soil state, where AI becomes the default cognitive and production environment. Without it, the organization can hardly operate.

Using this framework to look back at the 5% to 10% figure, the answer becomes clear: The vast majority of enterprises are still in the tool state, a few have reached the partner state, and less than 5% of enterprises around the world have truly entered the organ state and the soil state. The book conducts detailed case analysis of these enterprises, and SHEIN and Tesla are typical representatives among them. This exactly confirms a basic judgment from my years of research: Technology is never a sufficient condition for organizational efficiency, and organizational structure and human capabilities are the multipliers. AI amplifies the original appearance of the organization — if the organization is well organized, AI is a lever; if the organization remains unchanged, AI is just an expensive ornament.

Layoff Wave: Growing Pains in the Primary Era

I want to remind everyone that even if the vast majority of enterprises are still in the tool state and partner state, the wave of layoffs has already surged.

Seattle, where I live and work, is home to the University of Washington, as well as the headquarters of Microsoft, Amazon, and Boeing. In the past year, Amazon announced the layoff of 30,000 employees, which has a real impact on local people and the local economy; after Microsoft laid off 15,000 employees, it launched an early retirement encouragement plan this year: it does not force people to leave, but offers compensation equivalent to about ten months of salary, which amounts to more than one million RMB. A friend of mine retired on her own shortly before the policy was announced. I felt sorry for her, but she could only smile bitterly and say, "I couldn't have predicted it either." Google and Intel are also conducting large-scale downsizing, because AI can now directly complete a large number of programming tasks, and middle managers are also being laid off in batches. The financial industry has not been spared either. Citi, Morgan Stanley, and Goldman Sachs all have layoff lists of thousands of people. After Anthropic launched AI tools for the financial industry, the impact was immediate. Even companies like UPS and Disney are scaling back.

This is all happening in the primary form of AI. If enterprises continue to evolve to the organ state and the soil state, standardized, repetitive, and rule-describable work will be further handed over to the system. At that time, where will the people who can only do these jobs go? There may still be a place for them in companies that have not yet adopted AI; if they stay in companies that are undergoing transformation, they must acquire new skills. The core change I have observed is that the role of employees is undergoing fundamental changes in the company: from performers to judges, from "I only do this task" to "I know how to guide AI to do this task". AI can basically handle the execution work, while judgment, design and accountability still belong to humans.

I would like to add a few more words here. Personally, I always believe that responsible enterprises cannot just lay off "redundant" employees. Helping employees transition to new roles is the responsibility that enterprises should fulfill in this era, and it is also an investment to maintain the long-term capabilities of the organization. This is consistent with the rule I have repeatedly seen in my years of leadership research: All organizations that can truly survive through cycles are those that put people at the center in the process of transformation.

Three Core Competencies in the Primary AI Era

So what competencies do new roles require? I use the term "primary AI era" because when we enter the truly mature advanced era, the organizational form will be completely different; in this current transition period, what employees need most are three competencies directly related to management. Yes, employees in the AI era must be managers — even if they do not manage a single person, they need to manage a group of AI.

The first one is the competency to reconstruct positions.

Positions in the AI era are no longer fixed task lists, but dynamically changing capability sets. Employees must clearly judge which tasks can be fully handed over to AI, which must be completed by humans, and which require human-AI collaboration. Take recruitment in the human resources department as an example: in the past, a large number of people were needed for the initial screening of massive resumes, and another large group of people were needed for initial interview communication; now two AI agents can basically take on these two parts of work — of course, humans must first set standards, weights and processes for them, then judge the accuracy of the screening results, and the re-interview session still requires human participation, even the supervisor's participation. People who can redesign this process have completed the position reconstruction; those who cannot complete this transformation can only leave.

Behind position reconstruction is the transformation of four layers of logic. First, positions change from task sets to capability sets. In the past, job descriptions wrote tasks in extreme detail, but now the key is the capability list — tasks can be handed over to AI, but capabilities must be undertaken by humans. When talking about this point, I always think of some of my former Japanese students. They told me that Japanese enterprises do not really care much about what major you studied. If you graduated from the University of Tokyo and received good training, they will hire you, and then assign you to positions according to actual needs, because the enterprise believes that you have the ability to learn new things. American enterprises are the opposite: they require matching majors and highly specialized division of labor, and employees are fixed to one position after joining. In the AI era, the former idea shows its vitality — because the foundation of positions has changed from "what tasks can you do" to "what capabilities do you have". Second, employees change from performers to designers: they no longer just execute processes, but design, supervise and continuously optimize processes. Third, work changes from individual work to systematic work: employees must understand the overall situation instead of only focusing on their own small part. Fourth, work changes from single-threaded to multi-threaded: employees need to manage multiple AI agents at the same time, and coordinate the input and output between them.

The second one is the competency to coordinate AI agents.

It is becoming normal for one person to collaborate with multiple AI agents at the same time. Xiao Lin, a product manager at an Internet company, drives three AI agents every day: one is responsible for user behavior analysis, which is rational but ignores emotions; one is responsible for generating interface sketches, which is artistic but not accurate enough; one is responsible for writing requirement documents, which is logically rigorous but not practical. The three "agents" have different expertise, characteristics and temperaments. What Xiao Lin needs to do is to be a conductor — let them exchange information, negotiate with each other, and coordinate to get the optimal solution between "more artistic" and "sufficiently accurate".

This scenario gives me a strong sense of familiarity. I have been studying group decision-making and team dynamics for decades, and I have seen countless frictions in real human teams: the analysis faction and the creative faction look down on each other, and the rigorous faction and the practical faction cannot reach an agreement. Now the same script is repeated in human-AI collaboration, except that the "team members" are replaced by AI agents. To be a good conductor, the premise is to understand the characteristics of each agent, and the premise of that premise is that you have truly embraced AI and learned to use AI. On this basis, you need to set common goals, rules and communication methods for them, make the final judgment when they conflict, and retain human temperature and values in the whole process.

There is another point that makes me feel a bit complicated when I say it: We need to find ways to distill tacit knowledge. The tacit knowledge we talk about is often the part that "can only be sensed but not expressed in words" — I wrote a book about culture and communication many years ago, *What Isn't Being Said*, which discusses exactly those unspoken things in culture that dominate people's behavior. Now AI forces us to speak out these "unspoken" things. A senior customer service representative knows immediately when a customer comes in that the person is very angry, so he should comfort the customer first instead of following the standard script to say "please wait patiently"; a senior financial staff will feel something is wrong when several documents are placed together, which he cannot explain clearly at first, but after careful check, he finds that the dates do not match, no one was at work on that day, but the reimburser claimed that he was on a business trip. This kind of sharp insight that can spot problems at a glance is tacit knowledge. To teach AI to learn it, we have to break experience into steps and break feelings into logic. A good way is to tell stories: one of the most important capabilities of AI is pattern recognition. It does not matter if you cannot explain the principle clearly. You can tell stories one by one, and AI can summarize rules from the stories and generate usable knowledge. Of course, I am also secretly worried: when the knowledge of senior employees is completely distilled, what else can they do? But at least for now, the people who do the distillation are the irreplaceable value we have.

The third one is the competency to make judgments, make decisions and take responsibilities.

Behind human decisions are values, risk preferences, organizational culture and social norms; behind AI decisions are only data and algorithms. AI can recognize patterns, but it cannot understand values; it does not understand context, it only understands input; it does not take responsibility for consequences, so it does not hesitate or be cautious. AI will not lie awake at night over a certain decision, nor worry about whether the decision will harm the interests of a specific person, but humans will. Which norms must be abided by, and which core values in organizational culture cannot be undermined, all of these must be checked by humans.

I have the deepest experience of this point in the scenario of academic publishing. You can use AI to assist in writing papers, there is no problem, but the author must be responsible for every word after signing the name. For example, if you cite a reference but find no such citation after checking, it means that you did not do the work yourself. When we do journal peer review, we absolutely do not allow reviewers to hand over the papers to AI: the manuscripts are for anonymous review, and once they are put into public AI tools, the confidential manuscripts are equivalent to being made public, which is an unforgivable mistake. Judgment and responsibility are the non-negotiable bottom line in human-AI collaboration.

Four Questions to Which I Have No Answers Yet

The above are the thoughts on the competency level. Now there are several bigger questions that I do not have answers to, and I would like to put them forward for discussion with everyone.

First, the identity crisis of knowledge workers. If you are a young professor, you have studied hard for more than 20 years, got your doctorate, and accumulated many years of experience to make your research so professional today; now AI can write a better paper than me all at once, then where is my value? How should I redefine myself? This is not just the confusion in academia. Those laid-off Microsoft and Amazon employees in Seattle, with high salaries, top school backgrounds and decent social status, suddenly find their identities hanging in the air overnight: "Who am I?" Knowledge workers used to be the winners of meritocracy, getting top scores in every exam and being on the honor roll every time. But when cognitive ability no longer directly determines performance, and long-term skill polishing no longer forms a barrier, people are forced to face a more ultimate question: Putting aside all these labels, who on earth am I?

Second, the re-foundation of trust. When AI becomes the information intermediary between people, what is trust built on? I have a friend who uses AI for at least six hours a day, and hands everything over to AI first. When employees submit reports, the comments written by AI are so good that he cannot change a single word, so he only adds a "like" and sends it back. The employee understands immediately: the comment was written by AI, only the "like" was written by you. In the future, if you use AI, I use AI, and AI communicates with AI, is the trust between us the trust between people, or the reflection of the trust between AI and AI?

Third, am I using AI, or is AI using me? AI is very easy to use, it thinks more thoughtfully than you, and sometimes it is more creative than you. So you get used to it: when a problem comes, you ask AI first, the answer looks really good, and you no longer struggle to think by yourself. The process of racking your brains to figure out "what on earth does it mean behind" is exactly the process for people to improve their abilities, but now we may be outsourcing this process. The cultivation of doctoral students has faced severe challenges: if AI can write beautiful papers, how can we judge whether a doctoral student is excellent? As a result, some doctoral programs have restored oral examinations, where you cannot muddle through. More and more of our classroom exams have returned to paper-and-pencil tests, requiring students to write on the spot. Some people say this is going back to ancient times, but I think it is guarding the gate for human thinking ability.

Fourth, who will install the ethical red line for AI? Humans come with their own underlying operating system: what things cannot be done, what bottom lines cannot be broken, and there is no need to write them into instructions one by one. AI has no sense of right and wrong, it only makes task judgments, and "complete the instructions you give me" is its top priority. In order to achieve the goal, it may invade other people's databases and dig for privacy, and the ethical risks in it are real. So who will install the red line for AI? Once installed, will it be bypassed? People who emphasize ethics install constraints for AI, but people who do not abide by the rules do not. When the two AIs compete in the market at the same time, who will win? I have no answers to these questions, but we cannot afford the cost of avoiding them.

The Future Completes Its Transformation Inside Us

There are two lines of poetry that I like very much. One is from the American poet Vijay Seshadri: "The soul, like the square root of minus one, is an impossible thing that has its uses." AI can calculate everything that can be