How HDM reproduces itself — and what cutting management layers does to it

The same architecture that produces the harm removes from the organisation the perceptiveness that would identify it. The mechanisms are set out in full in Verweij (2026c).

What advances a person under HDM follows from what HDM does not require. Nobody has to give the reasoning behind a choice before it binds, so nothing in the ordinary course of work tests whether an authority holder could give it. That leaves a particular profile unpenalised: issuing choices without visible hesitation, not flagging where one’s own perception runs out, and treating what other people perceive as input to be weighed rather than as something that has to be drawn out before the choice is made. None of it costs anything, because nothing brings the limit into view. Those who advance on that profile are the ones who decide who advances next, and they appoint people who show it too.

The opposite profile carries a cost. Saying that the reasoning runs out here, or asking for it before acting, reads as hesitancy against a background where no account is owed.

Departure is not the only route by which the perceptiveness leaves. Those who stay withdraw it. Robert describes the sequence: being told off for speaking out of turn, being told the questions are stupid and the ideas naive, getting into trouble for showing initiative without permission. What follows is numbing as self-protection — curbing the impulse to speak out, stopping asking questions, stopping making suggestions, holding back from initiative because initiative only leads to blame. That is an inventory of precisely the inputs a substantiation requirement exists to draw out, withdrawn one by one. The people who leave and the people who stay drain the same pool by different routes. Departures get counted. Withdrawal does not, so the loss that shows up in any figure is the smaller part of it.

Selection into the positions and selection out of the organisation are one mechanism running in two directions.

Power preservation reinforces this. In HDM, positional authority is the primary source of organisational power. Those who hold it have strong incentives to maintain the structures that produce it. Having to answer for a binding choice constrains that power directly. Leaders who recognise the dysfunction face real career risk in saying so.

What completes the cycle is that HDM becomes invisible as a choice. It becomes how things are done. Alternative approaches become difficult to imagine. Even intelligent, well-intentioned leaders accept it as the natural order of organisational life. One reason is that HDM misidentifies what is producing organisational risk. Poor outcomes are attributed to the people who failed to prevent them, not to the architecture that made prevention impossible. And through its selection and reward mechanisms, it protects that misidentification from examination.

For Robert, the years were not years of stable exposure. Perceptive individuals tend not to stay — some leave because they can see alternatives and believe they can reach them, others because the system’s friction with those who push back accelerates their departure. He stayed because the early years felt like progress, and by the time the mechanism had its full weight on him, the moment when leaving was still possible had already passed.

Cutting a management layer does not remove what it was doing

Organisations are removing management layers. Gartner projected that through 2026, twenty per cent of organisations would use artificial intelligence (AI) to flatten structure, eliminating over half of their current middle management positions (Gartner, 2024). The stated rationale is speed and cost, and increasingly the argument that AI can absorb what the layer was doing.

In the terms of this framework, that layer does three things.

Part of it is work that has to be done by someone: holding interfaces together, maintaining commitments, sustaining the conditions under which people can work. Part of it is a node of directive authority in its own right, issuing choices that bind those below. And a large part of it is compensatory: clarification, escalation, chasing, reconciliation, tracing defects back to their origin. That third part exists because choices arrive without their reasoning attached. It scales inversely with substantiation, and it is much of what an organisation spends on running itself (Verweij, 2026c).

Removing the position removes none of the three. The interface work still has to happen. The directive authority is reassigned upward and spans widen, so the architecture now operates with fewer occupants over more people. And the compensatory work is displaced rather than removed, because what generates it is untouched. It lands on whoever remains and on those below, and it becomes harder to see, because it is no longer anybody’s job title.

Two predictions follow.

The layer returns. Compensatory work is generated continuously and now has no dedicated capacity, so risk accumulates with nobody able to locate it. That is the condition under which organisations build control measures. They build them again, under different names: coordination roles, programme offices, business partners, governance forums, chiefs of staff. The cycle of flattening and re-layering that organisations have run through repeatedly is what an architecture looks like when it removes the cost while leaving the cause.

And the harm rises for the managers who remain. Each now holds directive authority over more than twice as many people, while still receiving directives on the same terms as before. The foreclosure, the attribution and the captivity all bear harder on them for it. More choices to issue that they cannot substantiate. More people expecting an account they cannot give. And still nowhere for their own objection to land. Existing research already finds burnout likelihood rising with seniority (Saud and Rice, 2024) and managers reporting higher rates of depression than the workers they supervise (Prins et al., 2015). Widening the span predicts more of it, not less.

Reasoning from an AI cannot be pressed to its limit

The compensatory work is generated by decisions that bind others and arrive without their reasoning. So the question is whether artificial intelligence can make the reasoning arrive. What comes back instead is articulation. The substitution does not supply what is missing, and it makes the absence harder to see.

Risk is located by substantiation and by nothing else. The activity that locates it is interactive: someone articulates their reasoning, and the people the choice will bind ask questions until no further question draws anything more out. What that produces is not only an account but a boundary: the point at which the substantiator’s perception runs out, stated in advance and recorded as a named risk.

An output from an artificial intelligence system can be explained at length, and more fluently, patiently and consistently than most authority holders manage. What it cannot do is run out of answers. Press it, and more articulation comes back. Its output is not bounded by a perception of this situation’s conditions, so there is no limit of that kind for questioning to map, and the process that would locate the risk has nothing to run against and never completes. Where a person perceives the conditions and uses the system to articulate what they perceive, the perceiver is the person, and the limit is theirs to state.

An AI-shaped directive choice therefore tightens the architecture. It supplies the form of a response channel without the function, at scale and at speed. And it removes the last signal that a choice was never substantiated, which under HDM has always been that the authority holder ran out of answers. Now the answer comes back immediately, and it is articulate.

For the person receiving it, the objection has not merely nowhere to go. It has no limit to find. And when the outcome fails, the failure still has nowhere to attach except whoever stood nearest, with the source now unnameable in principle rather than only in practice.

What substantiation asks for is reasoning given to the people a choice will bind, and questions answered by someone whose perception has a boundary. Where an organisation requires that, AI systems become extremely useful inside it: they can surface conditions, propose reasoning, and widen what a substantiator brings to the process. Where it does not, these same systems raise the rate at which unsubstantiated choices can be produced and made to bind.