Don't Treat AI as a Wishing Well: Organizational Change Is a "Squid Game"

In organizational transformation, many leaders claim to embrace AI but in reality treat it merely as a "wishing well" with superficial actions; enterprises are fully aware that the transformation to an agentic organization is an inevitable trend, yet they refuse to take substantive steps forward.

Mu Sheng, a renowned domestic management scholar deeply immersed in organizational transformation, a continuous observer of AI trends, founder of Mu Sheng Consulting, and Doctor of Business Administration from the Guanghua School of Management at Peking University, points out that the root cause of stalled transformation lies not in technology or cost, but in diluted power and human relationship constraints. AI technology will not automatically drive organizational evolution. The practical path is to start with pilots in business scenarios and build a "small front-end + robust middle-back office" architecture. The good news is that once you embark on this journey, the transformation gains irreversible momentum — there is no turning back once the engine is started.

The following is the second part, the "Organization Chapter".

Business Magazine (hereafter referred to as "Business"): Your theory of platform-based organizations has been widely disseminated, and most Chinese enterprises are familiar with it. Since platform-based organizations serve as a springboard for transforming into agentic organizations, why don't large tech companies that are well-versed in technological trends actively promote this shift? We rarely hear of successful organizational transformation cases from major tech firms.

Dr. Mu Sheng (hereafter referred to as "Mu"): A significant number of large tech companies have either attended my courses on platform-based organizations or read my monographs on the subject. After each course, participants are usually filled with enthusiasm and eager to take action.

Talents at large tech firms have a deep understanding of business, so they can clearly see the potential benefits this organizational model can bring. Especially among the elite, there is a strong desire for organizational transformation. However, this enthusiasm rarely translates into practical implementation. The reason is simple: organizational transformation is a top-down initiative driven by the chief leader. No matter how passionate employees are, their efforts will not yield results if the top leader avoids the transformation at all costs.

For enterprises with decent performance, leaders believe their current organizational model works well and see no need for change. For enterprises with declining performance, leaders argue that they should focus on improving business results first, as there is no time left for organizational transformation. As business performance fluctuates, leaders can always find excuses to avoid pushing forward with organizational change.

Business: What reasons do leaders have to oppose organizational transformation? In the AI era, organizational evolution is an inevitable trend, and agentic organizations are characterized by lower costs, reduced risks, and higher efficiency — isn't this exactly what leaders want?

Mu: Leaders resist organizational transformation for several main reasons.

The first type of leader lacks expertise in organizational design. They understand the advantages of a platform-based organization but do not know how to implement it. Some even mistakenly believe that a divisional structure is equivalent to a platform-based organization, claiming that "divisions operate on my platform" — but in reality, each division is still a large hierarchical pyramid. Without a deep understanding of organizational logic, it is naturally very difficult to carry out genuine transformation.

The second type of leader fears losing power. When business units and the platform operate autonomously, dynamically allocating resources and delivering incentives automatically, leaders will no longer need to micromanage daily operations. Some leaders strongly oppose this model, no matter how efficient or profitable the organization becomes. They crave the sense of control that comes from acting as arbiters when the "big company disease" plagues the organization.

The third type of leader is unwilling to navigate complex interpersonal relationships. Senior executives in the traditional pyramid structure are often close associates of the leader. If responsibilities, rights, and interests are redistributed, some of them will lose their privileges. Facing opposition from these familiar faces, leaders find it difficult to push through reforms unilaterally, especially in a society that values personal connections.

When I communicated with Mr. Zhang Ruimin of Haier 10 years ago, he believed that "the fear of losing power" was the biggest obstacle. After witnessing organizational transformations in multiple enterprises, I deeply admire his accurate judgment.

Business: What kind of leaders will make a firm decision to push forward organizational transformation? If the top leader is determined to transform, does that guarantee success?

Mu: There are three types of leaders who are truly determined to drive organizational transformation.

The first type is driven by "aspiration and will". These leaders have strong strategic determination and an extremely high level of organizational insight. They can foresee the "final form of the organization" and proactively and unswervingly push forward transformation even before the enterprise encounters a crisis. They do not care about criticism from ordinary people, nor do they take "one-time successful transformation" for granted. Instead, they are willing to pay the necessary costs during the process to achieve the final victory. They view transformation failures as part of the journey, and even temporary setbacks are seen as progress that brings them closer to ultimate success.

With firm insight, resolute action, sufficient patience, and the tolerance for mistakes provided when the enterprise is thriving, such organizations will only move forward more smoothly on the path of transformation, and success is inevitable.

The second type is driven by the "cliff edge crisis". These enterprises are on the verge of collapse but still have a slim chance of survival. At this point, transformation becomes a race against time, a narrow bridge that must be crossed carefully. Only by making no mistakes at every step can they move away from the cliff and return to safety. The question is: if they did not understand these principles in the past, will the increased pressure of survival make them suddenly enlightened?

The third type is driven by the "deathbed scenario". For these enterprises, transformation is more like a "placebo". The leader sees the inevitable failure but refuses to accept it, or tries "everything possible to save a dying horse", hoping to create a glimmer of opportunity for a turnaround and preserve some dignity.

Most leaders fall into the latter two categories when they choose to pursue organizational transformation. Those rare entrepreneurs who drive transformation out of "aspiration and will" are almost the godfather-level figures in China's business community.

Business: A platform-based organization integrated with AI becomes an "agentic organization". This architecture raises a question: AI enhances individual capabilities, which seems to reduce hierarchical levels; but at the same time, data, models, and computing power resources are likely to be concentrated at the headquarters or platform departments. Under the impact of AI, will large tech companies' organizations become flatter, or will a new form of centralization emerge?

Mu: Whether it is a platform-based organization or an agentic organization, the organizational structure must be multi-centered and flat. Small front-end business units act as individual nodes in the network, connecting various functional departments to form "end-to-end" delivery for customers. This is exactly the effect of AI enhancing individual capabilities: organizations can operate like special forces teams instead of relying on massive traditional armies.

However, the platform is shared, and resources must be concentrated in the "platform" composed of middle-back office functional departments — which may breed more bureaucracy. This is exactly the dilemma encountered in past "large middle-back office" construction. But in an agentic organization, platform resources are not allocated by the middle-back office, but "flow" through an internal market mechanism. Under this resource allocation model, connections are established via APIs, based on fixed economic contracts, without needing to cater to the preferences of middle-back office "officials". As long as the front-end unit offers a reasonable price, it can obtain services and pay according to actual results.

If a "platform" falls into bureaucracy, it does not deserve to be called a platform. Such an enterprise is hardly different from a traditional pyramid organization, and at best can only be described as a "pyramid structure embedded with AI tools".

Business: Many enterprises now count token consumption as a metric, but this indicator only shows that employees are using tools, not that organizational capabilities have been upgraded. How can managers determine whether a department is genuinely undergoing AI transformation or just achieving "high tool usage"? What more effective AI transformation metrics should enterprises establish?

Mu: Token consumption is one of the metrics used to assess employees' AI usage when enterprises promote "AI transformation". It is similar to how white-collar workers' workload was measured by the number of keystrokes in the past: it works temporarily but has little long-term value.

To determine whether a department is genuinely undergoing "AI transformation", we need to look at the maturity of the agents it produces. First, check the proportion of manual work — if human intervention is too high, the agent's functionality will be limited. Second, track the number of times the agent is invoked, the depth of its usage, and its actual performance results — these are the key metrics.

For example, if the front-end stores of a chain enterprise use the "pricing agent" from the middle-back office to find the optimal balance between production volume and price to maximize revenue, the final incremental revenue is the core performance indicator. If the "pricing agent" performs poorly, it proves that the large language model is not mature enough, and the front-end stores will turn to external service providers. This switch can be done with one click via an API, making the transition very easy.

Business: Will AI introduce new variables to the organizational transformations driven by the different motivations we discussed earlier? Will it increase the success rate of organizational transformation?

Mu: An agentic organization is a platform-based organization fully integrated with AI technology, and it does increase the success rate of transformation. Once an enterprise starts building large language models across various functional areas, it will greatly reduce the number of human employees, retaining only elites who focus on model design, training, supervision, and contingency handling. Functional collaboration within the enterprise will be completed through API connections. On this basis, if a market-oriented incentive mechanism is established with clear economic contracts, the agentic organization can be realized step by step.

This process is irreversible. Once leaders press the start button, there is no turning back. In the past, when enterprises built large middle-back offices, they required these departments to provide services for the front-end to call on demand, but the large middle-back offices still hoarded resources. Now, API connections are completely emotionless: the front-end calls services based on price and quality, and if the service is poor or too expensive, it can switch to external suppliers with one click. Just like the "pricing agent" example mentioned earlier, if it performs poorly, the front-end business units will abandon it. In this scenario, "pricing" is no longer a "command" from a functional department, but a "service" provided by the platform.

In the past, such organizational transformation required the cooperation of the middle-back office. But now, once you are on this track, you must move forward, otherwise you will fall off the cliff. In the past, when implementing platform-based organization plans, you needed to support and guide the transformation. Now, the system will run forward on its own. In the past, even after full digital transformation was rolled out, some departments could still operate outside the system. Now, the "Noah's Ark" has limited berths — if a department does not board, it will have to bear the consequences on its own.

Interestingly, enterprises cannot afford to reject this major technological trend of AI: if you do not use it while others do, you will inevitably become slower, heavier, and weaker than your competitors. Just like the offline retail industry in the past: once some enterprises started to launch e-commerce businesses, others had to follow, otherwise they could not compete at all. Today's enterprise "AI transformation" is like an arms race — once one enterprise starts the process, the entire industry has to follow suit.

Business: Large internet companies have inherent technical advantages in building agentic organizations, and they also need to promote organizational transformation amid the intense competitive pressure. Have you observed any similar practices among internet enterprises?

Mu: Internationally, Microsoft has proposed the "Frontier Enterprise" blueprint and piloted it first in its internal IT department, envisioning that the organization will evolve into a state of "human-led, AI-operated". To achieve this, they have built a complex technical architecture, and deployed "Copilot Cowork" as digital colleagues at the application layer, maximizing human-AI collaboration and embedding AI capabilities into every corner of the organization.

Meta is trying to completely reshape the organization with an "AI-native" mindset. Mark Zuckerberg is developing a "CEO agent" to cut through hierarchical barriers and access information directly, building an ultra-flat new organization. They have developed internal tools such as Metamate, even created employee groups for individual agents to interact with each other, and used gamification mechanisms to incentivize employees to use AI. This practice that touches on the incentive system is a notable progress.

Since 2025, Amazon has deployed "thousands of agents" across all business lines, set behavioral boundaries for agents through the Amazon Bedrock AgentCore platform, implemented the "model factory" strategy, launched the Nova Act model that can autonomously execute tasks, and reconstructed its advertising system using AI.

In addition, Google is building the "Agentiverse" ecosystem, Salesforce has deployed the Agentforce digital workforce platform, and IBM Research is piloting teams composed of multiple agents to handle complex projects. These are all attempts by enterprises in relevant directions.

Domestically, the team behind Tencent's desktop agent tool WorkBuddy operates as a flat organization, where its business modules and code are transparently shared with all employees. ByteDance has also promoted the implementation of multi-agent frameworks: its Seed team developed the M3-Agent multi-modal agent framework and is advancing agent deployment across multiple business lines. Huawei has also released the enterprise-level agent platform ModelArts Versatile to promote human-AI collaboration.

These practices are noteworthy, but they are mostly limited to "developing AI native to functional departments, building technical platforms, and facilitating human-AI collaboration". In essence, large tech companies have made these transformations based on their precise grasp of technological trends and firm investment, but they have not yet delved into the fine-grained reform of organizational structures and incentive systems. While the slogans all point to "transforming into agentic organizations", genuine "organizational transformation" has not yet started.

Business: These practices lead us to a hypothesis — if enterprises fully adopt AI technology, will they automatically transform into agentic organizations? If not, what do they need to do? In other words, what is the first step to transforming into an agentic organization?

Mu: The practices mentioned above are mostly technical explorations and have not yet entered the stage of genuine organizational transformation. Once they touch real "transformation", the obstacles we discussed earlier will still emerge. Therefore, we cannot assume that AI technology will automatically drive organizational evolution.

Many enterprise leaders recognize the power of AI technology, but they lack sufficient understanding of AI and the incentive mechanisms required for organizational transformation. They tend to treat AI as a "wishing well", constantly pouring in their unrealistic fantasies.

Leaders do not need to be AI technical experts, but they must have AI literacy and a deep understanding of organizational design principles. To transform into an agentic organization, the first step is to select a pilot scenario and open up a complete business workflow. This mainly involves two tasks: first, building the initial framework of the three-platform architecture, and second, introducing market-oriented incentives.

For the first task, we need to build a "small front-end + robust middle-back office" architecture based on a specific business scenario. On one hand, we form small business units that connect to all functions required to close the business loop, where functions are accessed in the form of "functional BP + functional agent". On the other hand, we turn necessary functions into agents that can provide services via APIs. The purpose of this step is to leverage the power of organizational transformation to force necessary functions into the business loop simultaneously, preventing any department from "staying on the shore" and operating with old bureaucratic habits.

For the second task, we need to center around business units and establish a mechanism where all members co-create value, share risks, and distribute benefits. Members of the business unit, even those in the middle-back office, can obtain a share of excess profits through performance betting and subscription. The purpose of this step is to use benefit alignment to eliminate the retreat options for functional departments participating in the transformation.