The reorg landed on a Friday and the new chart went out Monday. A director of engineering who had been running four teams came back from the weekend running nine. The memo announcing it used the word streamlined twice. Nobody took work away from her, and nobody gave her a bigger title. Two layers of managers underneath her had been removed and their people pushed up, so what she inherited was the check-ins and the small daily interpretations of what leadership actually wanted, all of it now landing on the same calendar she had the week before. Six months on she was still doing the job. She was doing it from a distance that had turned it into a different job.

What the numbers actually say

Gallup has a name for the thing that used to be true about people like her. They call it the engagement premium. Managers, in Gallup's long-running global data, run consistently more engaged than the people they lead, and for years the gap held wide. In 2022, manager engagement globally sat at 31 percent against 20 percent for individual contributors.1 By 2025 managers were at 22 and individual contributors were at 19.1 Nine points came off the manager number in three years while the people below them barely moved. The premium did not shrink by degrees. It very nearly stopped existing.

Gallup sells manager development programs, which is worth saying out loud before leaning on their data. It does not make the count wrong. The State of the Global Workplace study is the largest ongoing measurement of employee experience anywhere, and this finding runs against the easier story a vendor in that business might prefer, which would put the problem in the individual contributors.

One caveat matters for anyone reading this from an American office. Gallup reports that engagement in the U.S. and Canada held steady in 2025 and remained among the highest of any region.1 The steepest single-year manager decline showed up in South Asia, down eight points, where Gallup notes the share of workers who are managers fell over the same stretch, which is what organizational flattening looks like once it reaches a dataset. Gallup's own language stays careful there. Flattening "may be a factor."1 So read this less as a bulletin that your managers are collapsing and more as a reading of where the pressure originates, because the mechanism has nothing regional about it.

The mechanism

That mechanism is span of control. Gallup's research on team size finds that manager engagement can decline as spans widen.1 Put it in plainer terms. The thing a manager does that cannot be automated, delegated, or absorbed into a dashboard is pay attention to one person at a time. It is the only part of the job with a hard ceiling built into it. You can push more strategy through a manager, more reporting, more tooling. Push more people through the same finite attention and the parts of the job that run on attention are the parts that go.

Flattening removes the layer. It does not remove the work the layer was doing. That work goes somewhere, and where it goes is into the attention of whoever is still standing there.

Why this is an agility problem

Most of the flattening I have watched get justified over the last decade was justified in the language of speed. Fewer layers, faster decisions, less padding between the person doing the work and the person answerable for it. The logic holds up, and I have made that argument myself in rooms where it was the right one. What gets skipped is that removing a layer changes what the remaining layer can do. Gallup's own note deserves quoting straight: organizational flattening "should not just be a transaction," and if you do it, it has to account for manager and team engagement along with everything else managers already carry.1

The Enos Progression runs from compliance to commitment to ownership. Compliance is cheap and it scales beautifully. You can get compliance out of a manager who has nineteen people and forty spare minutes, because compliance only asks that people know the rule. Commitment costs somebody the time to make a case that could fail. Ownership costs more than that, because it means handing real authority to a named person and then staying near enough to see what they do with it. Both of those run on attention. Widen a span past the point where a manager has attention to spend, and the organization does not come out of the trade flatter and faster. It comes out a compliance machine with agile vocabulary painted on the side.

The part that reaches the AI slide

Here is where all of this arrives at the thing every executive team is currently funding. Gallup's Q1 2026 U.S. workforce survey found that setting technical integration aside, the strongest predictor of whether an employee uses AI at all is whether their own direct manager actively champions it.1 Gallup's CEO put it flatly in the report's introduction: “Even the most sophisticated neural network cannot overcome an indifferent team leader.”1 So an organization can flatten its way to a manager with no room left to champion anything, spend eight figures on a capability whose adoption turns on that manager caring, then commission a study about why the return never arrived.

What a leader can check this week

The diagnostic here is not a survey. Ask a manager two levels below you how many people report to them, and then ask what the number was three years ago. If it grew, ask what came off their plate when it did. That last question is the one that produces the silence. When nothing came off, the flattening was an accounting exercise and nothing more. The layer is still there in the work. It has just relocated into somebody's evenings.

The director with nine teams is not detached because she stopped caring about any of it. She is detached because the version of the job she was good at ran on a proximity the new chart made arithmetically impossible, and nobody in the room where that chart got drawn was tracking proximity as a cost. It never appeared on the slide. It appears now, three years later, as a number that slid from thirty-one to twenty-two while everyone kept their eyes on the individual contributors. What is the span of control on your own second layer, and when did anyone last ask what it cost to widen it?