The Great Flattening: How AI Agents Are Quietly Gutting Middle Management
Forty-one percent. That is how many employees, in a Korn Ferry survey of 15,000 professionals worldwide, say their company trimmed management layers in 2025. One out of every two managers you passed in the hall last year may have no counterpart today. Welcome to what researchers are calling the Great Flattening — and it is happening faster than most organizations planned for.
The coordination argument that is falling apart
Middle management has always justified itself through coordination. Status updates, task assignments, information filtering up the chain, strategy translated into daily work. Not vision, not leadership. Logistics. That is exactly where AI agents are sharpest.
Companies like Meta, IBM, and Coinbase have begun deploying automated systems that handle reporting, workflow orchestration, and cross-team communication. IBM consulting, which employs roughly 150,000 human consultants, is now building what senior vice president Mohamed Ali calls "digital workers" directly into project teams. He expects entirely new management structures to emerge from it. McKinsey senior partner Alexis Krivkovich puts the shift plainly: AI gives leaders "more of a superhuman capacity to manage across bigger scopes." One person can now cover what previously required three or four middle managers acting as information conduits.
The numbers behind the flattening
Manager headcount at public companies fell 6.1% between 2022 and 2025. The agent software market sits at around $11 billion for 2026 according to Grand View Research, growing at 40% annually. KPMG estimated agentic AI could drive a 30% increase in workforce efficiency and a 25% drop in operational costs by 2027. Companies in their research report earning $3.50 for every $1 invested in agentic AI.
Gartner projected that 20% of organizations would use AI to flatten structures by 2026, eliminating more than half of current middle management positions. The data suggests this is less a forecast and more a description of what is already happening.
What nobody is saying loudly enough
There is a problem that is not getting much airtime: where do senior leaders come from in five years?
Shaun Warman, who coined the Great Flattening label, raised this directly. "Eliminating their middle managers in 2026 are simultaneously dismantling the apparatus that produces senior leaders in 2028," he warned. Middle management is where people learn to handle ambiguity, deliver bad news, develop junior employees, and align a team around a shared goal. AI can track tasks. It cannot teach those things.
Korn Ferry data bears this out. Nearly half of senior executives now report struggling to manage everything that landed on them after layers were removed. And 37% of employees say the absence of managers has left them feeling directionless. The savings are real. The capability costs surface later.
Flat vs. flattened: a distinction that matters
Max Martina, president of Cambridge Leadership Associates, draws a useful line. Flat organizations are designed from the start with minimal management layers, clear roles, and autonomous teams. Flattened organizations are traditional hierarchies that had layers removed, often without redesigning the underlying work. The second group tends to end up with the problems of both models and the advantages of neither.
The companies handling this best are treating it as a redesign, not a cost cut. They are asking which human functions AI cannot absorb — judgment, mentorship, emotional context, escalation calls — and who owns those functions after the restructuring. Researchers identify the new management competencies as: orchestrating human-AI collaboration, coaching at depth rather than at scale, translating ambiguous strategy into specific work for individual contributors, and knowing which decisions need a human hand.
Where things stand
The most significant organizational changes are still concentrated among technology-forward companies. Most organizations are in early-stage AI adoption without the infrastructure to automate coordination at scale. But the direction is clear enough. The safe middle management career — the layer that moved information and collected a salary for doing so — is ending. What remains is the harder work that could not be automated in the first place. Whether organizations are actually building the people to do it is the more difficult question, and so far, most are not asking it loudly.