I went with too limiting, but I would say it's quite a complex issue. Different classes give over different amounts of power to the subclasses in a way that feels clumsy and inconsistent, rather than highly intentional. Base classes vary a bit in how solidly-designed they are, too, without the subclasses.
Cleric is the outright worst design re: subclasses, not because it's too weak overall, but because it's remarkably both bland and inconsistent (always hard to combine those two but they managed it), and the entire concept of Clerics is messed-up by making the subclass "be" one of the aspects of a god, chosen from a ridiculously tiny list. And there's not much to work with, either.
I agree with people saying this was probably to stop the Cleric being a monster, but it's not a good design, and definitely too limiting in a bad way.
A lot of other classes it feels like because subclasses exist, they didn't actually build the base class in an interesting way, which isn't great.
There's also a big issue where you're kind of supposed to pick the subclass that fits the story/tone of the character you want to roleplay, but the very limited number of subclasses, combined with the wild variance in how powerful they are creates a double-sided problem. Some people pick a very weak or very boring subclass, mechanics-wise, because it's thematically right, and have less fun and feel less effective. Others pick a subclass that might not be a great match thematically, but is mechanically not rubbish, which can be uncomfortable (but is probably less bad than for the game than the inverse at least).
They also bring back the PrC problem, in that some of them are really specific, in a generic game. You can get around this, but it's weird and unnecessary. At least they lowered the specificity early on in 5E.
I think if I was wanting to maintain subclasses in a 6E, I'd want to move more power to them, but also make sure they were more generic, and much more tightly balanced, mechanically. I'd also take the Arcana Unearthed (i.e. Monte Cook d20 book) approach and design all the subclasses for all classes, based on what players wanted, re: archetypes/playstyles. That is very clearly NOT the approach taken with 5E, which has instead been a weird mix of that (in a few cases), tradition (taken to mindless lengths), whimsy (in the bad sense), simulation (which is terrible in this context), just outright arbitrary-ness.
For ultra-specific stuff I'd want another layer of mechanics of some kind, one which was more a matter of flavouring.
A number of players I have played with have bemoaned the lack of decision points for their character when leveling, apart from spells for spell-casters.
I've seen similar, even with the less mechanically-inclined players, which surprised me. I think it was the final straw for the player who usually plays Rogues - he loved 4E and picking his precise abilities and so on (despite never having really done that before) - and with 5E he's now clearly having more fun with classes that have actual choices (like Warlock, particularly).
