Well, part of the problem is that sometimes there's process where failure is providing information that may be desirable for changing your approach, but you don't want acquiring it to make the situation worse.
Say you're climbing something; a not-uncommon pattern in a trad game is for a success to make progress (but not necessarily resolve the whole thing), a crit makes faster progress, a failure makes no progress and a fumble creates a potential fall (you usually don't want it to automatically do so if there's multiple rolls involved because of the way maths work in those situations). If you get more than one failure, it may be there to tell you you need to try to do something different (perhaps move sideways to a different part of the face) similar to how failure to strike and do damage in combat may tell you that a different tactic is needed.
Its still serving a purpose, but its not, per se, advancing things. But the ability to do so effects pace-of-resolution in a way that does not enforce partial or complete failure the way "any roll has consequences" does if full success is not made (and as I noted, probabilities of each and how they're framed matter; PbtA games tend to frame "success with a cost" as part of success, which is not how at least some people see it, and embody it as a component of its "success" probability range which is very much not how at least some players will see it).
Ok, but even pretty much all “conventional” game state-of-the-art advice embraces a fail-forward mentality. Daggerheart is probably the purest distillation of the “game as collaborate cinematic experience” and bakes it right in. As long as I’ve been playing 5e, “I roll and nothing happens” has been the most boring thing - I just didn’t know there was alternatives until pretty recently. If I was starting over again now, I’d just add stakes to every roll: “on a success, you get your aim - on a failure X.”