IMHO, the fun is that less abstracted systems are actually fixable because you can “see” the failures, as opposed to cloudy stuff where you do a lot of work to prove that the problem is on the provider’s side, and then wait months for them to implement a fix… only then to have already worked around the issue.
Now I have a not so fun job at Big Tech doing SRE type work; all of that mucking around is abstracted away from me.