Delayering means cutting a level of supervisors or coordinators out of an org chart, so that fewer people sit between a frontline decision and the person ultimately accountable for it. It predates AI by decades; the “flattened pyramid” restructurings of the 1990s used the same term for the same move, usually justified as removing bureaucratic drag and speeding up decisions.
What’s different about the current wave is the stated cause. Companies including Amazon and Cloudflare have cut management layers in 2025 and 2026 while explicitly naming AI tooling as the reason the layer is no longer needed. The logic is that a manager’s coordination role, keeping information moving between people who can’t otherwise see the whole picture, checking work before it goes further, stops requiring a person once a system can do that coordination directly. That’s a different claim than “the manager’s tasks got automated,” and it’s the harder one to plan around, because it targets the role’s reason for existing rather than its task list.
The useful test for telling a genuine efficiency gain from a stealth reduction in accountability is what happens to the checkpoint the removed layer used to own. If a named remaining role now explicitly holds that sign-off, the coordination cost really was absorbed by better tooling. If the checkpoint simply stops happening, the organisation hasn’t automated a task, it’s made a class of decision unaccountable.