At the end of every AI agent automation, a human is required to ask:

1. Did it do it right?
2. Does it solve the problem?

I’m not convinced we can ever remove that person. Nor can that person be anyone other than a capable expert who is able to evaluate these questions with a high level of reliability.