One layer out. Less management, more autonomy on the floor. Shorter lines between who does the work and who decides. It sounds like an organizational structure question, and it is, but by now it's also something else: much of the work a middle management layer does today consists of tasks that are partly or largely taken over by AI. Compiling reports, monitoring progress, flagging deviations, presenting initial choices. If those tasks shift, what the layer below needs to be able to do in order to function without that intermediate layer also changes.
That is the reason "becoming flatter" without AI is a different ambition than "becoming flatter" with AI. Without AI, flatter mainly means: more decision-making authority lower in the organization, and therefore more training, more trust, more risk of inconsistency. With AI in the picture, part of the work that intermediate layer did doesn't shift to the floor, but to a system that needs oversight. The question then becomes not just "who decides now", but "who judges whether the system decided well, and on what grounds does it reject it".
The tasks a middle management layer performs today fall roughly into three categories. Part of it can be taken over by AI: status overviews, flagging deviations against a norm, summarizing progress from multiple sources. Part of it can be partly taken over, with a human approving or rejecting with reason: an investment proposal that a system calculates, but whose context and judgment remain with a human. And part of it remains human work: the conversation with an employee who is stuck, the political trade-off between two departments, the decision no one wants to make with the push of a button.
The intermediate layer you now want to eliminate often consisted of a mix of those three. If you remove the layer without knowing which part of that work goes where, not only does the layer disappear, but so does the oversight that belongs to the second type of work. That is where flatter organizations get stuck in practice: not because the title is gone, but because no one explicitly approves or rejects anymore, and no one records that reason.
Companies where becoming flatter does work generally have three things in order that other companies still lack. First: they know per process which part of the work AI can handle, which part requires oversight, and which part remains human work, rather than assuming that at department level. Second: they have explicitly redistributed decision rights, not let them implicitly erode — someone on the floor now has the right to approve or reject something that previously sat with a manager, and that right is documented somewhere. Third: they have reassigned the capacity that becomes available (in hours, not in headcount) to the work that remains, rather than assuming it will resolve itself.
Companies where it doesn't work have often changed the structure — the org chart is flatter — but haven't answered the underlying question: who now judges what the system delivers, and is that person equipped and mandated for it. That difference isn't in ambition or intention, it's in readiness. The same question, incidentally, applies to ambitions that at first glance seem to have little to do with AI; the ambition to become more customer-focused also splits into work that shifts and work that remains with people.
Becoming flatter this way doesn't require fewer roles, but different ones. There emerges a need for a role that judges the quality of AI output before it enters a decision — not a manager in the old sense, but someone with the mandate to reject and record the reason for doing so. There emerges a need for people on the floor who receive greater decision-making latitude, and that latitude must match what they could already do or still need to learn. And there remains a need for someone who has the non-transferable conversations: performance, conflict, direction. If you let that last role disappear because the title above it falls away, something disappears that no system takes over.
Whether that means fewer people are needed is a question with its own legal requirements and falls outside what can be answered here. What can be answered here: which work in your organization can actually be taken over by AI, which part requires oversight and which part doesn't. That is precisely the question the work scan from FTE TO AI answers per task.
Not the number of layers on the org chart. Three things you can check: is there, for every process that used to run through the intermediate layer, an explicit rule for who approves or rejects something, and is the reason recorded somewhere. Do the people who now have more decision-making latitude actually have the knowledge and the mandate to fill that latitude. And has the time that became available been noticeably assigned to work that matters, rather than everyone becoming a little bit busier without anyone being able to say why.
To know where you currently stand on these points, and where the biggest gap is between what the ambition assumes and what the organization can do, the sequence is often: first get clear on how you reduce dependency on a few key people, because that dependency often sits precisely in the layer you want to eliminate. If you also want to know how to bring the rest of the organization along without a months-long process, that touches on how you involve the shop floor in a strategy without it causing delay.
The free readiness check consists of eight short questions, one per dimension, and gives a picture of where you are furthest along and where you are least far along — no judgment, but a starting point. The full ambition assessment, with the four layers from vision to target state and the translation to roles and decision rights, is under construction. If the ambition itself isn't yet sharp enough to test, start with how you make a vague vision concrete.
Vertel wat u wilt bereiken, dan kijken we samen wat daarvoor moet staan.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.