"Modern" devops (and kubernetes). It seems to me that people want to avoid human error and dependence by replacing human operators with complex networks of dozen of services, each of which can fail in isolation or in conjunction with others. How do we measure the costs and benefits of putting in all that extra complexity in our stacks, how do we ensure that the friction they cause will not be a greater detriment than the previous situation they aim to improve? Are we in a situation where the solutions to existing problems have so many problems themselves that they need solutions of their own, thus creating an infinite tower of ever more complex software?