A director I know went from seven direct reports to fourteen over about nine months. Nobody sat her down to discuss it. Two peers left and were not backfilled, a layer above her was removed in a restructuring, and the org chart redrew itself around her one quiet decision at a time. Her title stayed the same. Her compensation stayed the same. Her calendar turned into a wall of thirty-minute blocks with no space between them. When she finally raised it with her own boss, the answer she got was that everyone was carrying more right now.
Her story is no longer unusual, and the numbers now make it hard to treat as an anecdote. Something structural is happening to the shape of organizations. It has a name, and most companies are doing it without measuring the cost. If you manage people, or manage people who manage people, this may be the single biggest change to your actual job in the past two years, and it arrived without an announcement.
The Numbers Are Not Subtle
Start with the simplest measure there is: how many people report to one manager. That figure, span of control, sat at 8.1 in 2013. By 2024 it had reached 10.9. By 2025 it was 12.1. That is a fifty percent increase in a little over a decade, and the curve steepened at the end rather than flattening out. Among small and medium-sized businesses the change was sharper still, with average spans doubling from three direct reports to six between 2019 and 2025.
The averages hide something important. Gallup’s data show the median American manager oversees six people, while the mean has climbed to 12.1. When the median and the mean split that far apart, it means a growing minority of managers are carrying enormous teams and pulling the average up behind them. Roughly a fifth of managers now have between ten and twenty-four direct reports, and about thirteen percent oversee twenty-five or more. In organizations that have deployed agentic AI at scale, MIT Sloan’s 2026 research found spans stretching as high as fifteen inside some divisions.
The layer cuts show up in the layoff data too. Middle management’s share of total layoffs climbed from twenty percent in 2019 to thirty-two percent in 2023, a sixty percent jump. This was not incidental. Companies looking to reduce cost went after the layer that coordinates, not the layer that produces, and they did it on purpose. Korn Ferry surveyed fifteen thousand professionals worldwide and found that forty-one percent said their employer had trimmed middle management layers in the prior year.
Why Companies Decided the Layer Was Optional
The argument for cutting is not stupid, which is part of why it spread so fast. A large share of what a middle manager does in a given week is coordination: collecting status, reconciling two teams’ timelines, chasing an approval, and writing the update that rolls into the update that rolls into the deck. That work is real, and it is also the kind of work that software has been creeping up on for twenty years. AI agents can now pull status from the systems where work actually lives, draft the rollup, flag the dependency that slipped, and route the approval. In back-office functions like finance, legal, and human resources, that automation is already in production rather than in a pilot.
Gartner put a number on the ambition. Through 2026, the firm projected that one in five organizations would use AI to flatten structure, eliminating more than half of their existing middle management positions. Fortune covered the same shift in June under the heading of AI agents flattening corporate hierarchies, and the phrase that stuck to the whole phenomenon was the great flattening.
Here is where the logic gets thin. Coordination is part of the manager’s job, not the whole of it. The rest is judgment about people you know well enough to read: the decision about who gets the stretch assignment, the conversation where someone finally admits they are lost, the political cover that lets a team take a risk. None of that is status reporting, and none of it has been automated. When you remove the layer that was doing both kinds of work, the automatable part gets absorbed by software and the human part gets absorbed by whoever is left standing.
What Breaks First
The first thing to break is not performance. It is attention. And attention degrades before any dashboard notices.
Gallup ran a meta-analysis across more than two hundred thousand manager-led teams to find out whether an ideal team size exists. The honest finding was that no universal number holds up, because manager quality and context matter more than the ratio. But one pattern in the same body of research deserves a spot on every senior leader’s wall: manager engagement itself peaks somewhere around eight or nine direct reports and declines as the span widens past that. The manager is the first casualty of their own expanded team, and a disengaged manager is a poor instrument for engaging anybody else.
Gartner found that seventy-five percent of chief human resources officers believe their managers are overwhelmed. That is not a fringe view from a few burned-out people. That is three quarters of the executives who own the people function saying, on the record, that the operating model they are running does not fit the humans running it.
Employees feel the gap even when they cannot name it. In the Korn Ferry data, forty-three percent of people said their leaders were not aligned with one another, and thirty-seven percent said the reduction in managers had left them feeling directionless. Those two findings belong together. The layer that used to translate strategy into a specific answer for a specific person is thinner now, so more people are working from their own interpretation of a slide they saw in an all-hands.
Writing in Forbes at the end of July, Jaka Lindic framed the problem in one sentence worth sitting with: as organizations reduce middle management layers, the leaders who remain are being asked to operate teams beyond their capacity. Capacity is the word that matters. Nobody in this story lacks skill. They lack hours.
The Coordination Work Did Not Vanish
The assumption underneath most flattening decisions is that the coordination work disappears along with the coordinator. It does not. Some of it goes to AI agents, which is the part that works. Some of it goes upward to senior leaders, who inherit day-to-day operational load on top of a strategic role they were already struggling to protect. And a lot of it goes sideways and downward onto individual contributors, who now spend their afternoons in the meetings their manager used to attend on their behalf.
You can see this in the mismatch between the layoff savings and the productivity numbers. Organizations booked the salary savings immediately and precisely. The absorbed work landed diffusely, in fifteen-minute increments, across dozens of calendars where nobody measures it. That is why the cut looks clean in the finance model and messy in the building.
The practical test is simple. Take one process your removed layer used to own and trace it end to end. Ask who does each step now. If the answer for most steps is a person rather than a system, the flattening moved work rather than eliminated it, and the manager effectiveness problem you are seeing is arithmetic, not attitude.
Leading Well at Fourteen When You Were Built for Seven
Assume nobody is giving your layer back. What actually helps?
That means the answer is not simply to work harder. It is to redesign the manager’s operating system around fewer live conversations, clearer decision rights, and more deliberate use of attention.
Stop distributing your attention evenly. At seven reports, you could give everyone a weekly thirty and roughly keep up. At fourteen, an even split gives everybody half of what they used to get, which means the person who needed forty minutes this week gets fifteen and the person who needed five also gets fifteen. Differentiate openly. Tell your team you are going to spend more time with whoever is in the hardest situation this month, and that the allocation will move. People accept uneven attention when the logic is stated out loud. They resent it when it feels arbitrary.
Push decisions down with named boundaries rather than vague encouragement. Telling a team to take more ownership does nothing. Telling a specific person that they decide anything under a stated dollar figure, or anything that does not change the delivery date, actually moves work off your desk. So does telling them to proceed without asking. Ambiguity about decision rights is what generates the approval traffic that eats a widened span alive.
Make written updates the default and treat meetings as the exception. This is the one place where AI agents genuinely reduce load rather than shifting it. A short written status that an agent can assemble from your existing systems removes the meeting whose only purpose was information transfer. Protect the meetings that involve disagreement, ambiguity, or someone’s career, because those are the ones that fail in writing.
Track development explicitly, since it is the first thing a wide span silently cuts. When a manager runs out of hours, the coaching conversation is what gets dropped, because nothing breaks this quarter when it does. Put a name and a date on it. If you have not had a growth conversation with someone in ninety days, that is a fact you can observe, and observable facts are easier to act on than good intentions.
The Pipeline Problem Arriving in About Three Years
There is a slower risk buried in this that almost nobody has budgeted for. Fortune flagged it in June, and it is the part of the great flattening that will be hardest to reverse.
The junior manager role was never only about coordination. It was the apprenticeship where people learned to run a budget, deliver bad news, sit in a staffing debate, and decide something without enough information. If you remove the rung, you keep the savings now and discover in three to five years that you have nobody ready for the rung above it. The same logic applies to early-career individual contributor work that AI now absorbs. The tedious first two years were how people built the judgment you want to promote later.
Some organizations are handling this deliberately by retraining displaced managers into strategic and client-facing roles instead of cutting them, and by building explicit development paths where the apprenticeship used to happen on its own. That costs money in the year you do it and pays back in a year when someone else holds the budget, which is exactly why it keeps getting postponed.
Where This Leaves You
If you run a team that has grown without anyone acknowledging it, the useful first move is to make the change visible. Write down what your span of control was two years ago and what it is today. Write down which responsibilities arrived with the added people. Bring that to your own manager as a resourcing question rather than a complaint, because a number invites a decision and a feeling invites reassurance.
If you sit above the managers doing this, the question to ask before the next layer comes out is where the coordination work is going to land, by name. Not whether the structure will be leaner, but which specific person absorbs each specific piece. The companies that come through this in decent shape will answer that question in advance. The rest will find out the expensive way, somewhere around the second year, when their best managers leave and the bench behind them turns out to be empty.