A leadership team says: "we want to become flatter." Fewer management layers, shorter lines, decisions closer to the work. In the same meeting one person means: fewer managers per employee. Another means: fewer handoff moments between departments. A third means: teams that decide for themselves without having to go back to a manager. Three different organizations, one sentence.
The ambition only becomes concrete once you add a layer: flattening *because of what*. Because AI takes over part of the work that currently justifies a management layer — coordination, control, passing on information between levels. That is a different ambition than flattening through reorganization without that shift. The capabilities that go with it differ too.
Management layers exist partly to do work that falls into three categories. Some of it can be taken over by AI: summarizing reports, flagging deviations, preparing a decision with the relevant data attached. Part happens with human oversight: AI proposes an assessment, a person approves or rejects it and gives a reason for doing so. And part remains human work: the conversation with an employee who is not performing, the trade-off between two equally weighted but incomparable risks, the responsibility that must sit with a name.
At organizations where this already works, it has been worked out exactly which tasks fall into which category, per layer and per role. At organizations where it does not work, the assumption is that AI takes over "the work of a manager," without anyone having recorded which part of that work exactly that is. The difference does not lie in the technology but in the precision of that breakdown.
Flattening without these capabilities produces an org chart that changes, while the work stays where it was — just distributed among fewer people.
Decision rights that move along with the layer. If a team decides for itself on something a manager used to decide on, it must be recorded who holds which mandate, and up to what amount or impact someone may decide without escalation. Without that record, the decision does not move, only the delay does.
Oversight that actually reviews AI proposals. If AI makes a recommendation and a team member approves it, that person must be able to substantiate a rejection. That requires knowledge of the underlying process that previously sat with the manager and must now sit with the team.
Escalation paths that do not silently revert to the old layer. Flat structures that, under pressure, fall back on "just checking with the manager" are not flatter — they have only added an informal layer that is described nowhere.
Capacity that becomes visible in hours, not in headcount. If coordination work falls away, freed-up capacity emerges. What happens with it — more work per team, different tasks, something else — is a choice that must be substantiated with facts about those hours, not with an assumption.
The role that changes most is not the employee but the manager who did coordination and handoff work. That work does not disappear entirely, it shifts: part to AI, part to oversight within the team, part remains with a smaller number of people with a different task content. What happens with this for the people currently filling this role falls outside what can be answered here — separate legal requirements apply to that, and it is up to the employer, not up to an assessment of ambition and readiness.
What this page does answer: which capability needs to be in place before the layer can be removed, and who takes over which decision right. This overlaps with questions that arise around growing without hiring extra people and around becoming less dependent on a small number of key people — in all three cases the question is not whether AI takes over work, but whether the organization is ready to actually let the decision right that is freed up land somewhere.
It has succeeded if a decision that used to have to go three levels up now stays at the level where it originates — and you can verify that by the turnaround time of decisions, not by the number of job titles that have disappeared. It has succeeded if teams can demonstrate why they rejected an AI proposal, instead of blindly following or ignoring it. And it has succeeded if the freed-up hours end up somewhere you can point to, instead of evaporating into extra meetings that informally replace the old layer.
Whether the ambition "flatter" is sharp enough to be tested this way is a different question from whether it is achievable — the difference between a goal and an ambition lies precisely there: a goal is testable, an ambition is often still just a direction. And how quickly a layer can actually be removed without the work being left undone depends on how the organization has changed before — see how fast an organization can really change.
The underlying question — which work in your organization can, right now, genuinely be taken over by AI, and which part will keep requiring oversight — is answered per task with the work scan from FTE TO AI. For the ambition itself, there is the free readiness check: eight short questions, one per dimension, giving you a picture of where you are furthest along and where you are not. The full ambition assessment, with the four layers and the five confidence gates, is under construction.
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.