Org flattening is being sold as an AI workforce strategy, but it solves the wrong problem. Middle managers are not coordination overhead; they are the human infrastructure that makes collaboration across departments and competing priorities possible. The real issue in most organizations is underskilled managers running poorly designed processes, and the solution to that is investment in leadership capability, not elimination of a layer. The organizations that will succeed with AI integration are the ones whose managers have the skills to lead effectively at every level, not the ones with the fewest layers.

Org flattening has become one of the most discussed AI workforce strategies of the moment. The pitch is straightforward: AI can handle coordination, communication routing, and status aggregation, so the management layers that existed to perform those functions are now redundant. Cut them, compress the hierarchy, and have employees manage agents directly. It sounds efficient. It may even be partially right for some organizations.

But for most organizations, and particularly for the complex, multi-functional, globally distributed ones that make up the majority of the workforce, this logic skips over something critical. It treats middle managers as a coordination tax when they are actually something else entirely: the human infrastructure that makes collaboration possible.

Removing that infrastructure before building something to replace it does not produce a leaner organization. It produces an organization that cannot work across itself.

What Middle Managers Actually Do

The case for flattening rests on a narrow view of what middle managers contribute. Yes, some management work is coordination: aggregating updates, routing decisions, tracking progress. And yes, AI tools are increasingly capable of performing those functions. But coordination is not the same as collaboration.

Collaboration is what happens when people with different expertise, different departmental priorities, and different definitions of success have to produce something together. It requires negotiation, trust-building, and the kind of contextual judgment that only comes from understanding both the work and the people doing it. Middle managers are often the ones who hold that understanding across organizational boundaries.

Consider something as straightforward as getting IT approval to deploy an AI tool for a business function. Someone has to negotiate that. Someone has to understand the business case well enough to speak the IT team’s language, and understand the IT constraints well enough to represent them back to the business. That someone is almost always a manager in the middle, not an executive, and not an agent. AI bots cannot negotiate with people. Not yet, and not in the ways that actually matter.

The instinct to make organizations more efficient is right. The mechanism of eliminating an entire class of managers to do it is not. It is the equivalent of cutting off fingers because you want a glove to fit better. The problem is not the fingers; the problem is that you need a better-fitting glove.

The Problem Has Been Misidentified

When organizations look at their middle management layer and see inefficiency, they are usually seeing something real. But they are misreading what is causing it.

In our experience, the pattern looks like this: organizations have processes that are poorly designed or never formally designed at all. They then place managers into those processes who do not have the skills to make them work well. Underskilled managers in broken processes do not produce nothing. They produce compensating behavior: unnecessary meetings, redundant approvals, information hoarding, work-around systems. That is what inefficiency in the middle actually looks like. And it is not a headcount problem.

Hiring better-skilled people into inefficient processes does not fix the inefficiency. Redesigning processes without developing the people who run them produces the same result. Both things need to happen together, and neither of them happens by cutting a layer of the org chart.

The organizations that will actually succeed with AI integration are not the ones with the fewest layers. They are the ones whose managers at every level have the skills to lead effectively in conditions that are changing faster than their training anticipated.

What to Do Instead

The alternative to flattening is not defending the status quo. Most organizations do have too much management overhead, and the question of how to increase efficiency and reduce the cost of coordination is a legitimate one. The difference is in what you optimize for.

Start with processes, not people. Before you can know whether you have the right number of managers, you have to understand what the work actually requires. Which processes need to exist? Which ones exist out of habit or historical accident? What does effective coordination look like for the organization you are trying to become, not the one you inherited? Those questions have to come before any structural decisions.

Then invest in the skills that make management effective at every level. Collaboration is a learned capability, not a personality trait. So is conflict management, performance coaching, cross-functional influence, and the kind of data reasoning that lets a manager make a defensible decision rather than just a fast one. These are the competencies that determine whether your management layer creates value or drag. And they are exactly the competencies that AI cannot substitute for, because they are fundamentally human and fundamentally relational.

Organizations that promote managers without giving them the training to succeed in their new scope are not saving money. They are deferring costs, and adding interest. The underskilled manager in an unclear process does not just perform poorly. They train their teams to work around them, erode the trust that makes collaboration possible, and create the very conditions that make the organization look like it has a middle management problem.

The Leadership Competency Question Underneath the AI Strategy

There is a version of the AI workforce conversation that is entirely about structural design: how many layers, what spans of control, which roles are redundant. And there is a version that asks a harder question: do the managers we have, at every level, have the skills to lead in an environment that is changing this fast?

Those are different questions, and the second one is more important.

The OGC Leadership Competency Wheel organizes leadership capability across six domains, from self-leadership and personal accountability at the foundation, to the people leadership, collaboration, and strategic reasoning skills that become critical as scope expands. The domains most at risk in an AI-accelerated environment are not the analytical ones. They are the relational ones: the ability to build trust across functions, to manage conflict productively, to develop others and create conditions for collective performance. These are the capabilities that determine whether an organization can actually absorb change, rather than just announce it.

The organizations that will navigate this period well are not the ones that moved fastest to flatten. They are the ones that asked what their managers needed to be capable of, and invested accordingly.

If your organization is working through questions about management structure, leadership effectiveness, or how to build the capabilities that AI integration actually requires, we are happy to have that conversation. Explore the OGC Leadership Competency Wheel or reach out to talk through what you are seeing.

Have a specific question about how this applies to your organization? Our FAQ page covers common questions on talent systems, leadership development, and engagement data.