Don't cut middle managers first
AI makes coordination cheap, so the obvious move is to remove the layer of managers whose job looked like coordination. Several large companies have done it. The risk is that the damage does not show at once. It arrives a year or two later, when the company discovers what those managers were actually doing.
The work nobody wrote down
Middle managers do more than pass information up and down. They translate: they turn a vague instruction from above into concrete work, and they turn a messy reality on the ground into something leadership can decide on. They integrate: they make sure what one team does fits what another team needs. And they carry knowledge that exists nowhere else: why a client is treated differently, which supplier needs watching, what was tried before and failed.
None of this appears in a job description. Remove the layer and it simply stops happening. The first years look fine. Then quality slips, teams pull in different directions, and new initiatives stall because nobody connects them to the rest of the company.
The order of operations
The objection is right, but it is an argument about sequence, not about whether to redesign. Cut first and hope the systems will cope, and you lose the invisible work. Build the systems first, so that translation, integration and knowledge have somewhere to go, and then redeploy the people.
Translation goes into systems that hold the shared context: the data, the history, the reasons behind past decisions. Integration goes into processes that run across teams instead of being handed over between them. Knowledge goes into deliberate transfer: the managers teach, record and coach before their role changes.
Before you remove a layer
List what the managers in that layer actually do in a typical week, not what their title says. For each item, name where it will be done after the change: by a system, by another role, or not at all. Anything in the third column is a risk you are choosing to take. Make that choice knowingly.