Your belief is sometimes called the "Raymond Chen" school of software engineering, although mostly because he's documented it rather than because he's in some way responsible for the Windows team taking this approach.
You can get yourself tied in some terrible knots this way. Chen's blog The Old New Thing is currently documenting how starting from [let's not annoy CP/M programmers] got them to [making an OS component optional causes security vulnerabilities in third party programs], over the course of a decade or so. Every step along the way is completely rational but the result is a confusing, insecure mess that's hard to reform.
But the alternative school, where everything not tied down and documented is up for grabs, and the tied down stuff might be cut loose and "deprecated" with relatively little notice, causes its fair share of problem as we've seen with the thread's topic.
Let me say this: It is very far from clear which of the alternatives here is better for anyone, from users to developers to OS vendors, let alone which would be best for all.