I think the issue there is that in exchange for solving the "one fat finger = outage" problem, you lose the ability to update the server fleet quickly or consistently.