Written by: Fang Dao
Fang Dao & AI｜Industry Observation
The discussion about "the disappearance of middle management" is essentially not a question of job survival, but rather a systematic replacement of a certain type of function within an organization.
When AI Agents begin to possess the capabilities of task decomposition, process execution, and result feedback, the information processing and SOP execution that once relied on human labor within enterprises are being transformed into capabilities that can be automatically completed by systems. What is being compressed is never a specific level, but rather "that part of work that can be written into a process."
The reason traditional corporate structures needed middle management was not management itself, but the high cost of information processing. The strategies of the upper layers needed to be translated into specific tasks, and the execution from the front line needed to be aggregated into structured feedback—this back-and-forth formed the foundation for the existence of middle management.
But when information can be generated and synchronized in real time by the system, and when tasks can be automatically decomposed and tracked by Agents, the necessity of this intermediate link begins to decline. Organizations no longer rely on "people to transmit information" but begin to rely on "systems to run processes."
This change shares the same logic as the replacement of front-line positions: front-line positions are replaced because their work heavily relies on standard processes; the part being compressed in middle management also revolves around processes—breaking down processes, transmitting processes, and monitoring processes. When the processes themselves are taken over by the system, the differences between middle management and front-line positions in this dimension begin to converge. What is being replaced is not the position, but the work structure.
Therefore, the ongoing change in organizational structure is not simply "flattening," but a kind of "de-intermediation." Information no longer needs to be transmitted level by level, and execution no longer relies on manual supervision. The hierarchical logic within enterprises is shifting from "information asymmetry" to "capability distribution." In this new structure, what is retained is no longer the position, but the capability itself.
Middle management is beginning to be re-differentiated into two types of roles: one type relies on process operation, and its value decreases as system capabilities improve; the other type can define goals, organize resources, and take responsibility for results, and its value is instead amplified. The core of the former is to ensure the process runs without errors, while the core of the latter is to make the system work under uncertainty.
This is also why "being able to use AI" is frequently mentioned, but a more accurate statement would be: the ability to embed AI into business processes and generate stable output is becoming the new threshold. AI itself does not automatically create value; it needs to be orchestrated, constrained, and integrated into the organization's operational system. Whoever can complete this orchestration occupies the new structural position.
If we push this change one step further, we can see a deeper transformation taking place in the form of enterprises: a company is no longer just a human organization, but is gradually evolving into a "capability orchestration system." In this system, some capabilities come from humans, some from models, and what is truly scarce is the ability to integrate the two into a sustainable operational structure.
Middle management will not disappear, but the "pricing logic of middle management" has already changed. Past value came from information asymmetry, while future value comes from system control power. After AI compresses the value of information, what truly remains in the organization will only be those nodes that can define rules, orchestrate systems, and take responsibility for results. In this sense, enterprises no longer need more "transmitters," but more "system nodes."
