I don't agree with that "not needed" explanation. But they are quite useless if the compiler can not deduce that both the values returned from the functions you are calling are monads, and what kind of binding should apply to them.
I find them to be at least somewhat useful in a language like javascript. I imagine they are more useful in a typed language with proper inference, but I wouldn't say they're entirely useless in a dynamic language.
What I think you said is that, in an untyped language, a function will just return "something" (as far as the compiler is concerned). But in a typed language, the compiler knows that a monad was returned. So the point of the monad is to make the compiler do magic, not to make the runtime do magic.
Well.. It's like 25% accurate, and you hit the part you were most worried about. But you have been seriously misled about monads in the past.
There is no magic in monads. At the level of runtime operations, they're just a couple of functions.
The magic taking place in the compiler is type inference guiding the selection of the right pair of functions during compilation. Without that, you have to explicitly use the correct functions. It's much easier to offload that bookkeeping to a computer than it is to do it yourself.