> ...it's far from proven that using complex systems necessarily causes the complexity of your system to balloon...

I don't know what "proven" would actually mean in this context, but there's a notion called "leaky abstraction". It bites you hard when you use a "do-everything" framework and something goes wrong.

All of sudden, the magic stops and you have to deal with a ream of obtuse stack traces, problems that completely smash your metal model of what's going on, and you end up in deep rabbit holes of stuff you don't want to get into at the worst possible time. That is a fundamental obstacle.