The most repeated prediction about the great flattening puts a machine at the center of it. Gartner forecasts that by the end of 2026, 20 percent of organizations will use AI to flatten their structures, eliminating more than half of current middle-management positions in those firms, a figure cited both by Deloitte (2025) and in Fortune's essay on the coming extinction of the middle manager (Fortune, 2025). The story underneath the number is that AI democratizes the knowledge that once made a manager valuable, shifting worth from knowing to executing (Fortune, 2025).
That story is half right, and the missing half is where the interesting argument lives. If a manager's job was mostly to be a router for information, AI is a genuine threat, because a good model routes information faster and at lower cost. But the manager who was only a router was already the weakest version of the role. AI does not so much end middle management as expose which managers were doing something a model cannot.
What AI actually removes
Take the specific claim seriously. Knowledge brokerage, the manager as the person who knows which policy applies, who to ask, what the process is, is exactly the sort of task a well-fed language model handles. Deloitte's finding that managers spend nearly 40 percent of their time on firefighting and administration describes a large surface of automatable work (Deloitte, 2025). Some of that surface will not survive, and it should not. Nobody was developed as a leader by reformatting a status report.
What the automation narrative skips is the other side of Deloitte's number. The same managers spend only 13 percent of their time developing their people, and 36 percent say they feel unprepared for the people-management part of the role (Deloitte, 2025). That is not a knowledge problem, and it is not a task an execution engine absorbs. Coaching a struggling colleague, holding a team's context through a messy quarter, deciding which of three plausible priorities actually gets done: this is judgment applied to specific people under uncertainty. AI can inform those calls. It does not make them, and it does not carry the accountability for getting them wrong.
Flat did not need AI
The strongest evidence that AI is not the real driver of flattening is that the most studied flat organization runs on almost none of it. Buurtzorg coordinates more than 900 self-managing teams and roughly 14,700 staff with two directors and no middle managers, and it did so years before generative AI existed (Martela and Nandram, 2025). The lever was not technology. It was team design: small units with defined roles, decision rights pushed to the frontline, and a thin layer of coaches who hold no formal authority (Martela and Nandram, 2025). If a company can go flat without AI, then AI was never the thing making flatness possible. It is an accelerant applied to an organizational choice that stands on its own.
That reframes the Gartner forecast. Twenty percent of organizations may indeed use AI to justify cutting half their middle layer (Fortune, 2025). Whether that produces a Buurtzorg or a mess depends entirely on whether the coordination those managers held gets redesigned or simply deleted. AI lowers the cost of the knowledge work. It does nothing, on its own, for the judgment and development work, which is why McKinsey's survey found two-thirds of leaders already calling their organizations overly complex (McKinsey, 2023). Automate the routing and leave the judgment orphaned, and complexity does not fall. It moves.
It is worth being precise about what the forecast is and is not. Gartner's figure is a prediction, not a measured outcome, and it describes intent, that a fifth of organizations will use AI to flatten, more than it describes results (Fortune, 2025). Predictions about how many firms will try something are cheap. The harder and unanswered question is how many will succeed, and success here is not defined by how many boxes disappear. It is defined by whether the work those boxes did still gets done. On that measure the technology is neutral. It hands a company a faster router and leaves the organizational design entirely to the humans, who remain free to do it well or badly.
The exposure
The useful way to read the next two years is as an audit rather than a purge. AI will strip the knowledge-brokerage function out of the manager's job and lay bare what remains. For some managers, very little remains, and their roles were fragile before any model arrived. For others, the automation clears away the 40 percent of firefighting and returns time to the 13 percent that was always the point (Deloitte, 2025). The technology does not decide which manager you are. It reveals it.
The practical implication for anyone running a team is to stop competing with the model on knowledge and start doing the work the model cannot touch. Ricardo Semler's version of self-management made the same bet decades early: value lives in trust, judgment, and the development of capable people, not in controlling access to information. AI is about to prove him right by removing the information moat entirely. The managers who survive the flattening will be the ones who were never really selling knowledge in the first place.
Sources
- Fortune (2025). Surviving the great flattening: the coming extinction of the middle manager. Reports Gartner's prediction that 20% of organizations will use AI to flatten and cut more than half of middle-management roles by end 2026, and the knowledge-to-execution thesis this piece contests. https://fortune.com/2025/09/19/surviving-great-flattening-coming-extinction-of-middle-manager-layoffs/
- Deloitte (2025). 2025 Global Human Capital Trends. Managers spend ~40% firefighting versus 13% developing people and 36% feel unprepared to manage people, separating the automatable knowledge work from the human judgment AI cannot absorb. https://www.deloitte.com/us/en/insights/topics/talent/human-capital-trends/2025.html
- Martela and Nandram (2025). Buurtzorg: scaling up an organization with hundreds of self-managing teams but no middle managers. Buurtzorg runs flat via team design, not technology, showing AI is an accelerant rather than the cause of flattening. https://link.springer.com/article/10.1007/s41469-024-00184-y
- McKinsey (2023). The State of Organizations 2023. Two-thirds of leaders call their organizations overly complex, showing that automating routing without redesigning judgment work moves complexity rather than removing it. https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-state-of-organizations-2023