A lot of companies have been asking whether AI can replace middle managers.

I started by looking at LLM training data. Code, Medicine, and Law appear as a dedicated category. Management does not.

There is no public evidence that frontier models were deliberately trained, balanced, and benchmarked for the full scope of managerial work: judgment, context, conflict, advocacy, accountability, and protection.

Companies have tried to remove management layers before: Google, Zappos, Medium, GitHub, Valve, Buffer, and others all provided the same lesson - coordination problems do not disappear when management is removed.

Then there is a serious risk: AI may automate managerial tasks while removing the human layer that translates, absorbs ambiguity, protects teams, and pushes ground truth upward.

This article explores all these aspects.