Maybe stupid isnt the best word but I feel strongly in both cases.
Decoupling config, app state and the app itself has a lot of upside in certain cases.
Even redundancy tended to be hardware-based to a significant degree. How much did the IT industry put into things like failover Unix clusters (and VAX/VMS etc. before that) over the years?
The fact that humans can develop emotional bonds with machines and objects is strange and fascinating.
A system where you can blow away the server and automatically have a new instance recreated is objectively better than a bespoke server that needs someone to recreate everything. The less time I need to spend manually configuring and tending to systems, the better. This is objectively a better model for software infrastructure and deployment. These are literally machines, not cattle or pets or whatever.
Stay at this game long enough and you start to see the yada yada about cycles (fashions) repeating themselves. Is this debate one? I don't know, but I suspect it to be so. Personally I don't think all the extra moving parts of having "cattle" are worth it unless you're trying to be like Youtube, and the reality is lots of folks think they are when they shouldn't. I'm also not OK with a pay-for-compute model, but I also understand that the accountants and bean counters love it.
In my homelab and in every organization I've ever been a part of, there is a mix of both pets and cattle (servers) depending on need. (Although, maybe that applies to people too?) But drift too far in one direction or the other and you're going to be pissing a lot of people off that you don't need to be.
In any case I think this tangent is bikeshedding. Which is another thing we engineers are great at. I should have realized how much one word could nerd-snipe