For decades, the shape of an organization was decided by the cost of coordination. Layers existed to move information, translate across functions, and keep everyone pointed the same way. Management was never only oversight. It helped people in different parts of a large organization understand the same priorities and work toward them together.
AI changes that math. Information moves instantly. Context is shared without a meeting. Analysis no longer waits for three handoffs and a deck. So the conclusion writes itself: fewer layers, more direct execution. Flatten the chart, run faster.
In companies I watch, removing layers simplifies the chart but creates new demands on the people who remain.
The Flattening Is Already Here — And It Is Not Working
Gartner projects that through 2026, 20% of organizations will use AI to flatten their structure, eliminating more than half of their middle management positions. The job market already shows it: US employers advertised 42% fewer middle management roles at the end of 2024 than in spring 2022.
Korn Ferry surveyed 15,000 employees across industries and roles. Forty-one percent said their organization had already slashed management layers. Of those left behind, 37% reported feeling directionless, and 43% said their leaders were not aligned. Different cuts of the same survey, but the pattern is hard to miss: companies removed the layers they were told were overhead, and many employees reported losing direction.
Flattening became a verb executives reach for the way they once reached for “synergy” — aspirational, vague, and expensive when it goes sideways. The promise was less bureaucracy. The result, where alignment was never rebuilt, was a faster machine that could not agree where it was going.
If the layers were just friction, removing them should have helped. So what were they actually doing?
The judgment inside a status update
The conventional story says middle managers aggregated information, tracked progress, and translated strategy downward. This coordination work is increasingly being automated.
That story is incomplete. Every coordination routine carried alignment decisions inside it. When a manager passed a status update upward, they were framing it — deciding what to escalate, what to suppress as noise, how to position competing priorities for whoever read it next. The information moved, but judgment rode along with it. That is why removing the role hurts more than the org chart predicts. Middle managers provide qualitative contextualization AI cannot replicate — frontline insight, historical nuance, the customer subtlety that explains why the dashboard is technically right and practically wrong.
There is also evidence that the role was poorly designed before AI. Thirty-six percent of managers say they are insufficiently prepared to be people managers, and 40% report a decline in mental health after taking the job. Those results suggest serious problems with the role’s current design. Those one-on-ones have a recognizable shape — the manager drowning in dashboards they did not ask for, expected to coach a team they barely have time to talk to. The role was failing before AI showed up to make it look optional.
But broken role design is not the same as a dispensable function. The work of reconciling priorities does not vanish because the person doing it was untrained and underwater. Six in ten managers spend more than half their time on admin AI can automate, part of why an estimated 1.5 million US management jobs sit at risk by 2030. Automating that administration can give managers time back. The harder question is how to preserve the judgment and support their teams still need.
Removing layers does not remove the underlying need to reconcile priorities. It removes the structure that absorbed it. The role becomes less about managing flow and more about managing tension.
If middle managers were running an alignment function while everyone assumed they were forwarding emails, automating the coordination should have left alignment alone. It didn’t.
Coordination Is Cheap Now. Alignment Is Not.
Coordination is moving information through the system. Alignment is getting people to agree on what it means and what to do about it. AI is extraordinary at the first and does little for the second on its own.
A Harvard Business School study provides an example: it followed 50,032 software developers from 2022 to 2024. As AI adoption rose, coding grew 5% as a share of work while project management fell 10%. AI did not come for the expertise at the bottom of the chart. It came for the coordination in the middle.
So coordination gets cheaper, and the part that catches everyone off guard is the other side of the ledger: alignment gets more expensive. Speed up the information and you multiply the moments where teams act before they agree. The system gets richer in possibilities and more ambiguous at once — more options, more parallel work, more that can quietly diverge.
As information becomes easier to move, agreement becomes harder to achieve.
Sixty-eight percent of organizations report AI-driven efficiency gains, yet managers struggle to reallocate the freed-up hours to strategic work. The time got cheaper. The judgment did not.
There is a tempting objection that the split is too clean — that all coordination is alignment work, since framing a message is already an alignment act. That objection is right, and it makes things worse, not better. If alignment was embedded inside coordination routines, automating them strips out the substrate it rode on. You do not get to keep the judgment after deleting the vehicle that carried it.
Fewer layers can make communication faster while leaving decisions harder to reconcile. This is the same compression I described in Phase Collapse: removing a stage can also remove the place where people settled an important question.
An organization ran this experiment years before AI existed.
The Spotify Warning
Spotify, with more than 10,000 employees, built its famous squad model in the early 2010s. Autonomous cross-functional teams in tribes, with chapters and guilds for cross-cutting expertise, and minimal hierarchy. The bet was that coordination would emerge organically instead of routing through managers — the pre-AI version of the same wager.
It became the most copied org-design framework in tech, even though Spotify itself has acknowledged that the model never fully worked.
Dual-reporting created persistent ambiguity about priorities, accountability, and career paths. Autonomy outran competency. Squads pulled in different directions: architectural decisions diverged, one squad’s update forced changes across several systems, technical debt piled up at the seams. Spotify eventually reintroduced structured leadership roles.
The lesson is the one AI is now teaching at speed. Autonomy without alignment does not scale. The same compression flattening project structures is now flattening org structures.
The obvious rebuttal: Spotify failed because it lacked AI. Give those squads real-time dashboards and alignment would have emerged. Would it?
But What About the Dashboards?
The case for dashboards is that they could replace some of this coordination. If AI gives every team real-time visibility into priorities, dependencies, and conflicts, the dashboard becomes the alignment mechanism — no intermediary required. GitLab and Automattic have run radically flat for years on transparent tooling. Where misalignment is just missing information, AI fixes it.
I concede the informational half of that argument. AI makes priorities visible and surfaces conflicts earlier. In loosely coupled organizations, where teams are mostly independent, this objection has real force and the thesis weakens.
But visibility is not agreement. Spotify’s transparency predated real-time AI, granted — yet it could see its own work and still drifted. If a shared screen produced alignment, every company with a Jira board would already be perfectly coordinated, which is not a sentence anyone says with a straight face.
The failures that wreck interdependent organizations are not informational. They are competing values, resource conflicts, and strategic disagreements no dashboard resolves. AI improves the visibility of these issues — it does not resolve them. It can even accelerate the damage by letting more work run in parallel before anyone reconciles it. That 43% of employees say their leaders are not aligned is not a data-access problem. They have the data. They lack agreement — the expensive thing.
Teams still need a way to resolve disagreements once they become visible.
From Flattening to Redesign
The fix is not to keep the old hierarchy. It is to stop treating flattening as subtraction and start treating it as redesign. I use a five-step sequence, starting with the judgment embedded in existing routines.
1. Coordination Audit. Before automating a routine, sort it: pure coordination (safe to automate), embedded alignment (the framing must be preserved elsewhere), or redundant legacy. Skip this and you automate the workflow while destroying the judgment inside it. Most teams skip it, because coordination looks trivial right up until it is gone.
2. Decision Rights Redistribution. When a layer disappears, its decisions land somewhere or fall into gaps. Decision rights need to be explicit, and interfaces between functions defined: which decisions move to AI, which to the people closest to the work, which need a new alignment role. The trap is redistributing authority without accountability — decisions get faster while no one owns the consequences.
3. Alignment Mechanism Design. Replace implicit alignment with explicit structures: forums that arbitrate competing priorities rather than report them, trade-off protocols, coherence checks. These need real decision authority — a status meeting with a new name is theater.
4. Role Mutation. Redefine the roles that remain around managing tension, trade-offs, and decision quality — not flow. Give them judgment criteria instead of throughput KPIs and the job of reintroducing deliberation into timelines AI has compressed into hours. Change the title without the incentives and behavior stays put.
5. Feedback Instrumentation. Flattening severs the upward flow managers carried. Without intermediaries, employees self-filter bad news, so decisions get made on cheerful, incomplete information. Rebuild it with structured sensing and alignment-health metrics like decision-reversal rates. It matters doubly because removing the layer also removes where mentoring and context lived — the dynamic that breaks junior-talent development when the middle vanishes, the same erosion I traced in The Expertise Cliff.
The sequence is not universal. Skip it in teams under 30, where everyone can just talk, and preserve compliance layers in regulated settings. In low-interdependence organizations, the Spotify failure mode never fully appears. Flattening without redesign creates instability — but only where interdependence makes alignment load-bearing.
The Layer Did Not Disappear
So before you draw the new chart, audit your coordination routines for the alignment buried inside them. Decide how those decisions will be made before removing the roles that currently handle them.
Before removing a role, make sure the people affected know where to take a disagreement, who can decide, and who will support them when priorities conflict. The chart may be simpler; that work still needs an owner.
References
- Gartner (2024). “Top Predictions for IT Organizations and Users in 2025 and Beyond.”
- Korn Ferry (2025). “Workforce 2025 Survey: Power Shifts.”
- Harvard Business Review (2025). “How AI Is Redefining Managerial Roles.”
- Fast Company (2025). “What happens to middle management when AI flattens your organization?”
- Deloitte (2025). “What’s the future of management?” Global Human Capital Trends.
- Andy Cleff (2025). “Spotify’s Organizational Evolution.” andycleff.com.
- Fortune (2025). “Surviving the Great Flattening.”
- Gartner / Korn Ferry (2024-2025). “Mentoring pathways break when layers are removed.”