Without abstraction enforcement mechanisms, an abstraction you have correctly implemented can still misbehave, because someone else has tampered with its internal representation.
Regardless, Rich Hickey's last talk (last month at Clojure Conj) offered a counter argument, why ADTs are usually not the best way to organize software (as part of his larger discussion against static typing).
It depends on context
> why ADTs are usually not the best way to organize software
He clearly hasn't proven anything about his code.
Is that a bad thing? It seems like many people are able to write software that works well and gets the job done without "proving anything" about their code in the way you're describing. Personally, I'm fine with not proving anything if I can deliver quickly and everything works, but I'd like to be convinced otherwise if there's real value there.
It’s a mirage to which many have succumb but like in the desert most died of thirst trying to catch it.
Let's try not to make inflammatory statements.
The condescending tone of the post is completely unwarranted, especially since it glosses over the real limitations of procedural values as abstraction building blocks, which don't magically go away when you rename them from “objects” to “functions”. In no particular order:
(0) There is no modular way to distinguish procedures that meet a nontrivial specification from all other procedures. Guaranteeing that a procedural argument never takes a silly value requires whole-program analysis, which doesn't scale.
(1) There is no modular way to communicate the fact that two or more procedures manipulate similar or related data structures. Try implementing abstract data types with binary operations (e.g. mergeable heaps) as objects.
(2) There is no modular way to control which procedures exchange information with one another. Say Foo is designed to communicate with Bar. I can replace Bar with a type-compatible but ill-behaved impostor Qux. As far as Foo can tell: https://az616578.vo.msecnd.net/files/2016/09/26/636105299923...
The cure for this illness is using abstract data types, not objects.