By: Kumar Dattatreyan

One in five. That's the share of organizations Gartner forecast, at its October 2024 symposium, would use AI through 2026 to flatten their structure and eliminate more than half of their middle-management positions. The number now anchors the loudest AI story in the leadership press. Fortune's June 2026 reporting adds a survey to it: Korn Ferry polled 15,000 professionals, and about 41 percent said their company had trimmed management layers in the past year.

The headlines credit AI. A leader made the decision. Removing a management layer is a choice about structure, and AI shows up afterward as the explanation. Leaders pick the layer because its cost sits on the spreadsheet and the work it does sits nowhere.

That gap between the cost you can see and the work you can't is the argument of this post. A cut doesn't teach you how to run a flatter organization. It tests whether your organization was ever ready to be flat, and you learn the answer by watching what breaks.

Is AI replacing middle managers?

Some of them, yes, and the strongest version of that case deserves a fair hearing first. Fortune's reporting lays it out: agents now handle coordination and status reporting, the work that once justified a layer of managers who relayed information up and down. If the relay no longer carries anything, paying for the layer is waste. Chris Daily made a version of the same point on episode 58, when he said AI was already shrinking content teams to one or two experienced writers and that some jobs were already going away. The camp adds that organizations waiting until they feel ready will lose ground to rivals rebuilt around agents from the start.

Part of that holds. Some layers existed only to move information. AI moves information well, so those layers are redundant. Cut them.

The weak point is the word "some." Gartner forecast one in five organizations, not five in five. Fortune notes that the biggest changes cluster in technology-forward companies with sophisticated digital infrastructure and that most organizations are still early in AI adoption. And Bret Greenstein of West Monroe, quoted in the same piece, warns against treating AI as a cost-cutting tool and says companies should first automate routine work so people can move to higher-value tasks. Even the reporting that carries the flattening story attaches a sequence: prepare the work, then change the structure. That sequence is the readiness argument, and the other side is making it. None of this says wait a year. The checks later in this post can run alongside the decision and need no reorganization to start.

What a management layer holds that the org chart can't show

A layer does two jobs. The visible one is carrying status and approvals between the people who do the work and the people who fund it. The quiet one is holding coordination nobody wrote down: someone in that layer knows which two teams will collide over a shared dependency and which approvals matter, and someone else can tell a senior stakeholder no without damaging the relationship. None of this appears on a chart, because it lives in informal trust relationships that formed long before anyone drew the boxes.

A sudden structural change exposes where decisions really ran. After a cut, the org chart tells you who reports to whom. Watching the work for a quarter tells you what the layer was for.

Consider two hypothetical companies that remove the same layer in the same week and issue the same press release crediting AI. In the first, decision rights were explicit and the layer's routing work was written down, so an agent picked up the routing and the work kept moving. In the second, the layer was the only place coordination lived. Decisions stall and two teams build conflicting versions of the same thing. A quarter later a leader describes the wreckage as an AI transition. Both companies are made up for illustration. The cuts were identical and the readiness wasn't.

AI makes a convenient explanation for the second company. It arrived alongside the cuts and it can't object.

Readiness first: Vijay Luthra's rule

On episode 139, Vijay Luthra described advice he'd given a client weighing a very big technology transformation: "don't not do it, but do it last. Do everything else first." He gave the reason in the same breath. There's "very little point in undertaking an ambitious technology transformation if your organization isn't ready for it." He asks whether the organization is ready for the new technology and reports that "Invariably, the answer is no."

A flatten is a technology transformation in all but name. It changes who decides and how information moves, and it leans on AI to carry the load. Run it without the readiness work and you get what Luthra says his own clients eventually report: they make the mistakes, then come back asking for help to rethink the whole thing.

Luthra doesn't predict that AI leaves jobs untouched. He said this revolution probably won't create as many jobs as earlier ones did. What readiness decides is whether the cuts that do happen come from a design or from a guess.

Tool first, problem second

Chris Daily's complaint on episode 58 was that AI had become a label on everything. His example was that "my light switch has AI in it." His correction was a sequence: "we should be looking for problems to solve," and reach for AI only if it's the right tool for the problem you find.

Organizations that cut a layer and credit AI run the sequence backward. They hold the tool and go looking for a structure to apply it to, and the management layer gets picked because it's the cost they can see. Daily had a line for what follows: "we're just speeding up how quickly we produce garbage." He meant bad content. It applies to org design too, because cutting a layer whose work was never defined and then adding AI opens the gap faster.

Daily's other point connects to readiness from the individual side. He said "it's less important to understand AI right now" than to ask how people and companies will adapt to the changes under way. Adaptation at the scale of a company is readiness by another name.

The constraint probably isn't the layer

Evan Leybourn, who founded the Business Agility Institute, explains the Theory of Constraints with a car production line. Wheels take four minutes to install and engines take five, so no improvement to the wheel team changes the five-minute output. Fix the engine team and the constraint moves to the wheels. His corollary for organizations: "An organization can only be as agile as its least agile function."

He puts the odds plainly: "Every system, every organization has a constraining factor to their agility, and it's probably not technology anymore." In the Business Agility Report data, he said, organizations struggle most with governance, meaning funding models and approval queues. His examples run from a nine-month wait for a budget change to a four-month wait for a hiring ticket.

Apply that to a flatten. If the least agile function is a nine-month budget approval, removing a manager leaves it untouched and removes the person who was routing around it. The constraint stays where it was, now with less capacity around it to absorb the load. Leybourn's line explains why: "Your agility is constrained by the system, the organization that it operates within." The application to management layers is mine. Leybourn made the point about technology and governance.

The slow part: decision rights and defaults

The org chart is the easy half to change. In a large complex organization the hard part of any transformation is the political system around the work, and a flatten doesn't skip it. People have to agree that fewer layers means decisions get made closer to the work, and they have to agree to make decision rights explicit and move coordination into the open. That agreement takes longer than a headcount decision, which is why it gets skipped. Our earlier post on why reorgs fail at the coordination layer covers the same ground from the reorg side.

Defaults come next. A flat organization works where the coordination you want is the default behavior, built into how people are rewarded and how decisions route. Design the incentive so the outcome you want happens without anyone protecting the old structure out of habit. Remove a layer and change nothing about those defaults, and the old coordination doesn't reappear on its own. Luthra framed the broader shift as "more about adapting systems, processes, technology to people rather than the people adapting to those technologies." A flatten that asks people to adapt to a thinner structure, with the same incentives and the same approval paths, does the reverse. The same logic runs through our post on why enterprise AI fails the way agile did: the tool arrives and the change work gets skipped.

Four checks before you remove a management layer

Trace one decision across two teams. Pick a recent decision that needed both teams and follow it from request to resolution. Mark every point where someone in the layer did something. Those points are the coordination the cut will drop.

Write down what the layer decides. List the decisions that route through it and assign each one a decider and a rule for resolving conflict. If you can't write one down, the layer is holding it. Our earlier post on decision rights covers why assigning a decider isn't enough on its own.

Find the constraint before you pick the target. Ask Leybourn's question: where is the point of greatest constraint right now? If the answer is funding or approvals, removing managers won't move it.

Change the defaults before the headcount. Change how people are rewarded and how decisions route so the coordination you want happens without the layer. Then cut the layer whose work AI can absorb and keep the one whose work you just wrote down.

Start with the first check this week. It takes one decision and one afternoon, and the answer tells you which of the two hypothetical companies you're about to be.

Related Podcast Episodes

Episode 139: The 5th Industrial Revolution: Why 40% of Jobs Will Disappear, with Vijay Luthra. Luthra puts technology last and the organization first, and he doesn't pretend jobs survive untouched. Listen for the client who asked how to start a very big transformation, because his answer is the spine of this post.

 Episode 58: AI Hype! Breaking down the Buzz, with Chris Daily. Daily starts from the problem and treats AI as one tool among several. His light-switch example makes the cost of tool-first thinking easy to remember.

 Episode 167: Business Agility in Crisis: 2025 Trends, with Evan Leybourn. Leybourn's production-line example is the quickest way to see why improving the wrong function changes nothing. His data on where organizations struggle most points at governance before it points at headcount. 

If you're weighing a layer removal and want a second set of eyes on how much of the coordination is written down, book a 30-minute conversation. It's a listening session, and you'll leave with the first decision to trace.