Chances are there's some cyclical dependencies. These can creep up unnoticed without regular testing, which is not really possible at AWS scale unless they want to have regular planned outages to guard against that.