Managers cannot sponsor AI from a distance and call it leadership.
They do not need to become the most technical person in the room. They do not need to know every tool. They do need enough fluency to understand how the work is changing and how to judge the outputs their teams now produce.
The manager's job moves toward standards.
What work should use AI by default? What work should not? What data is allowed? What requires review? What counts as good enough? What should be documented so someone else can reuse it? What stops if the new workflow works?
Without those answers, AI use becomes hidden and uneven. The eager people run ahead. The cautious people wait. The careful people over-review everything. The reckless people ship too much. The team gets private workflows instead of shared learning.
Managers also need to inspect quality differently.
A polished AI output can look finished before it has been thought through. The manager has to ask whether the work answered the right question, used permitted inputs, checked the right risks, and made the next action clearer.
This is harder than checking whether the deck looks good.
Managers also have to create learning loops. If one person finds a better way to do recurring work, that pattern should not stay private. It should become a short note, a template, a prompt, a checklist, a demo, or a review standard that the team can adapt.
The manager should not centralize every experiment. That will slow the team down. But the manager does need to centralize learning enough that the team gets smarter together.
The hardest management problem may be people whose work is changing faster than their identity. Someone who was valued for producing a certain kind of output may now be valued for judging, editing, reviewing, or designing the workflow that produces it.
That shift needs clarity. It also needs honesty.
The manager's role is not to cheerlead AI. It is to make the new standard visible and help the team meet it.