The Hidden Cost of Flattening Your Organization

McKinsey calls this "The Great Flattening." IBM is embedding "digital workers" alongside 150,000 human consultants. The pitch is simple: AI agents automate coordination, generate status updates, and streamline decision-making. So why keep all those management layers?

My reaction: directionally right, oversold.

AI agents remove reporting churn, coordination drag, and low-value administrative work. But they don't replace judgment, coaching, accountability, or cross-functional decision-making.

The distinction matters: flatten bureaucracy, not leadership.

The risk is companies cut management layers, call this innovation, then quietly dump the real work of management onto senior individual contributors and already stretched executives.

What Actually Breaks When You Remove the Control Layer

I've seen this play out in professional services. A custom software firm cuts project managers because AI generates status updates, summarizes calls, and coordinates tickets.

The mistake is assuming those roles were administrative.

They were managing scope, translating between client language and technical talk, resolving trade-offs, and spotting delivery risk early. What breaks is delivery discipline.

Projects start faster, scope creep rises. Engineers get pulled into client coordination. Small misunderstandings turn into rework. Escalations come later. Margins erode because nobody is holding the line between what was sold, what was promised, and what gets delivered.

The firm thinks they flattened bureaucracy. What they removed was the control layer protecting client trust, team focus, and project economics.

This isn't theory. When Wistia tried to build a flat structure, CEO Chris Savage discovered they had "done the exact opposite of what we had intended. We had centralized all the decision-making, and we were relying on a secret implicit structure to make progress."

AI Can Interpret Ambiguity. Humans Underwrite Commitment.

The control layer isn't tracking tasks. It's interpreting subtext, managing scope, protecting margin, and preserving client trust simultaneously.

An AI agent summarizes meetings and flags risks. What an AI agent won't do is handle the live moment where someone has to decide:

• "No, that's out of scope."
• "We'll absorb this one."
• "This needs escalation now."

"Not the real issue. Here's what the client is worried about."

There's the difference.

The control layer turns ambiguity into commitment. Without this layer, ambiguity gets dumped onto engineers, clients, and eventually executives.

Handling ambiguity isn't the same as owning the consequence of resolution. AI is getting better at interpreting messy inputs and surfacing options. But the move from ambiguity to commitment is economic, relational, and accountable.

A human is needed because commitment means choosing one trade-off over another, accepting risk, setting a precedent, and being answerable if things go wrong.

An AI says: "Based on past projects, absorbing this request will preserve client sentiment."

A human decides: "Fine, we'll absorb this one, but only this once, and I'm willing to wear the margin hit because this account matters."

Not prediction. Judgment tied to responsibility.

Where the Risk Actually Goes

When you remove the people who used to underwrite commitments, the responsibility doesn't disappear. The responsibility falls downward and upward at the same time.

Downward, the responsibility lands on senior individual contributors, engineers, product managers, and frontline operators who were never meant to carry commercial or political accountability.

Upward, the responsibility lands on executives, who get pulled into issues earlier and more often because nobody in the middle is confidently making the call.

The company thinks they removed a layer. What they did was redistribute unresolved risk.

This shows up as engineers in client meetings, product managers negotiating scope, salespeople making delivery promises, and executives getting dragged into avoidable escalations.

Research confirms this pattern. Managers with 7 or fewer direct reports score 20% higher on team engagement than those managing 15 or more. When managers exceed optimal span of control, they spend 65% more time in meetings and administrative tasks.

Flattening often feels efficient at first, exhausting later. The work of underwriting commitment still exists. The work gets pushed onto people who are either too junior, too specialized, or too senior to carry this well.

The Four Failure Modes Across Industries

The pattern is consistent. You remove a managerial layer, the risk doesn't vanish, the risk migrates. What changes is where the breakage appears.

In retail: execution drift. Poor in-store follow-through, stock issues, labor strain, inconsistent customer experience.

In professional services: scope leakage, margin erosion, delayed escalations, and client frustration.

In software product companies: priority chaos. Nobody cleanly arbitrates trade-offs, so teams build too much, decisions linger, and senior engineers become unofficial managers.

In manufacturing or supply chain: exception handling failure. The system works until something goes off-plan, then no one owns the cross-functional recovery fast enough.

In healthcare: coordination burdens get pushed onto clinicians, increasing friction, delays, and burnout.

In financial services: control and compliance gaps. Decisions get made faster, oversight, documentation, and escalation discipline weaken.

This comes down to one of four failure modes: execution drift, decision latency, burnout, or control failure.

Flattening isn't one thing. If you remove bureaucracy, performance improves. If you remove the people who absorb ambiguity and underwrite commitments, the organization gets faster only until the first real exception hits.

The Shadow Management Layer You're Actually Building

When companies realize they've created these problems, they don't rebuild the layer cleanly. They patch around the problem.

What happens instead is predictable:

• They create more steering meetings
• They give senior individual contributors unofficial management duties
• They pull executives deeper into operational calls
• They add PMOs, chiefs of staff, rev ops, delivery ops, or governance overlays
• They buy tooling to improve visibility and escalation

Rather than openly restoring the removed layer, they often build a shadow management layer.

Many "flattened" organizations don't stay truly flat. They become flat on the org chart and thick in practice.

The cost is higher than the original layer, less visible. With the original layer, you're paying for named people with clear accountability. With the shadow layer, you're paying in messier ways: more executive time, more meetings, slower decisions, duplicated coordination, role confusion, expensive senior talent doing work below their level.

Instead of one delivery lead making calls, you get a steering committee discussing the issue, a chief of staff preparing the brief, senior individual contributors doing the analysis, and an executive making the call late.

The real cost isn't payroll. It's latency, diffusion of accountability, and organizational drag.

The original layer says: "Jane owns this."

The shadow layer says: "Let's align."

The Question You Should Ask Before Flattening

When something goes off-script, who will have both the authority and the expectation to make the call?

This reveals everything.

If the honest answer is "the team will sort this out" or "we'll handle this through better visibility and escalation," you're not removing bureaucracy. You're removing the control layer.

Bureaucracy slows routine work. The control layer absorbs non-routine work.

If nobody clearly owns the messy middle (scope changes, exceptions, trade-offs, client tension, cross-functional conflict), then the risk hasn't disappeared. The risk has been redistributed.

Even McKinsey's own senior partner Bill Schaninger warns: "If you want to eliminate something, eliminate tasks, tasks with no value. But keep the role and curate the role to help develop your next generation of leaders."

Where AI Should Actually Flatten

The direction is right. The execution is wrong most of the time.

You should use AI to flatten workflow, not judgment.

Use AI aggressively on reporting, status collection, documentation, meeting summaries, ticket routing, first-pass analysis, forecasting support, and routine coordination. Layers often become bureaucratic here.

Protect the human layer around scope decisions, priority trade-offs, coaching, escalation handling, client or customer tension, exception management, cross-functional arbitration, and anything involving accountability for risk, margin, or trust.

Management isn't overhead here. It's control.

Flatten the administrative exhaust. Protect the people who underwrite commitments.

What Good Actually Looks Like

A company doing this well looks leaner, not leaderless.

The organization has fewer status meetings, fewer handoffs, fewer people collecting updates, far more automation around reporting, documentation, routing, routine analysis.

The organization also has clear human ownership in the places where this matters: who makes priority calls, who handles exceptions, who resolves cross-functional conflict, who owns client or customer tension, and who is accountable when trade-offs affect margin, risk, or trust.

Most companies today are trying to use AI to remove managers. The better companies use AI to remove managerial waste.

The healthy version is fast information flow, clean decision rights, wider spans for strong leaders, and a small number of high-judgment people who underwrite commitments.

What most companies are building instead is more tooling, less clarity, flatter org charts, shadow management in practice, and unresolved risk pushed outward into the system.

When this works, the company doesn't feel chaotic or "radically flat." The company feels calmer, clearer, and more accountable.

The one thing you should protect: the people who own exceptions.

Companies are most tempted to cut this layer because the layer looks like coordination overhead. These are the people who step in when reality doesn't match the plan (client tension, delivery risk, cross-functional conflict, store-level issues, margin trade-offs, quality problems).

Routine work gets automated. Exceptions are where the business gets managed.

Don't cut the layer turning surprises into decisions. AI helps this layer, AI doesn't replace this layer.