Something significant is being removed from organisations, and almost nobody is measuring it correctly. The public narrative about AI and employment focuses on two points of contact: the automation of entry-level, task-intensive work, and the augmentation of executive decision-making. Both are real. Neither is where the most structurally consequential change is occurring.
The most consequential change is in the middle.
Mid-level professional roles — managers, senior analysts, team leads, department coordinators — have been declining faster than either end of the organisational hierarchy in every major workforce dataset from the past 18 months. A 2025 McKinsey Global Institute analysis of S&P 500 hiring patterns found that mid-tier professional postings fell 18% over 12 months, while executive and entry-level postings remained roughly flat. IBM’s 2024 restructuring explicitly targeted middle management layers. Meta’s “Year of Efficiency” removed approximately 11,000 employees, disproportionately from mid-level roles. The pattern is consistent across sectors.
The standard interpretation of this data is economic: middle management is expensive, its decision-making functions can be partially automated, and AI-enabled executives can manage wider spans of control. This interpretation is accurate as far as it goes. It does not go very far.
What middle managers actually do is not primarily decision-making. Decision-making is the most visible part of their role — and, as it happens, the part most susceptible to automation.
What middle managers actually do inside organisations is not primarily decision-making. Decision-making is the most visible part of their role and, as it happens, the part most susceptible to automation. But the less visible functions are more structurally important, and they are harder to replace with software.
The first function is institutional memory. Middle managers are the people who remember why policies were designed the way they were. They know that the procurement process has a particular step because of a supplier dispute four years ago. They know that the marketing team and the engineering team have a communication protocol because of a failed product launch three years ago. They hold the context that is not in any document, because it was transmitted informally over time. When they leave, that context leaves with them.
The second function is translation. Strategy does not implement itself. The gap between a board-level directive and a team-level action is filled by people who understand both the strategic intent and the operational reality. Middle managers translate upward and downward: they compress what frontline teams are experiencing into language legible to leadership, and they decompress what leadership intends into specific actions frontline teams can execute.
The third function is mentorship. Junior employees learn how to operate in a professional context not from handbooks or onboarding processes but from proximity to more experienced colleagues. They learn what to prioritise by watching someone prioritise. They learn how to communicate in a given organisational culture by absorbing it through interaction. Middle managers are the primary vector of this transmission.
Research from MIT’s Sloan School, published in late 2024, found that organisations that had reduced middle management layers by more than 30% in the preceding three years showed measurable increases in project failure rates, employee turnover among junior staff, and time-to-competency for new hires. The productivity gains from reduced headcount were partially offset — and in some cases exceeded — by these costs, which are harder to measure and therefore less visible in the data that drives restructuring decisions.
This is not an argument against AI adoption, or even against organisational delayering in principle. Some middle management layers were genuinely redundant. The argument is more precise: that the knowledge transmission, translation, and mentorship functions that middle managers perform are distinct from their decision-making functions, and that they do not automatically migrate upward to executives or downward to AI tools when the middle tier is removed.
What organisations that are restructuring aggressively are betting is that these functions are either unimportant or replaceable. The evidence available suggests that bet is likely to prove wrong, and that its costs will become visible over time horizons of three to five years — longer than the quarterly windows in which the restructuring decision was optimised.
