I think this lack of an arbitrary feature boundary makes it a useful research language. However, for the same reason, I imagine it's not so good in production — it's like Perl with "more than one way to do it," but now you need a PhD [1] to understand several of the myriad ways of doing things.
[1] You don't need a PhD, but I'm guessing only a very small proportion of working programmers know what, e.g. a higher-kinded type is.