Google did a huge study of what makes a good manager (after trying to do without them) and found that a big deliverable of management is “psychological safety” for individuals when they are working as a team. Without that, people hoard information and spend all their effort to undercut each other, which is detrimental to the team and company overall.

At Google, where they foster a culture of intellectual safety. Their systems are so complex that no one person knows everything, and having people bullshit their way around things is dangerous, so people have up feel safe to ask "dumb" questions in order to make good decisions. At other companies with different cultural values, say Microsoft of the 90's with stack ranking, a good manager may need to foster the opposite in order for their direct reports to succeed in that culture.

Some people feel safer asking the LLM dumb questions, at least I do because of social pressure, but answers could be bullshit. I am retired but I could imagine that I only would proceed if I either had good enough references backing the LLM's answer or if I decided that my question is not so dumb after all and took the plunge to ask a human.

This seems to indicate that AI could relieve managers from answering dumb questions at least partially.

Are we talking about middle managers or line managers?

For line managers, the manager has to talk with human employees, and bad managers asides, that is where the human element comes in: fundamentally, your employees are still human.

On the other hand, if the line managers are all human, then the middle managers have the same human elements, too.