I wonder if this is a reference to "I need you to understand that people don't have conversations where they randomly recommend operating systems to one another"

But to the actual point of the article: my understanding is that there are areas where you can use bidirectional typing (e.g. languages that have subclasses) where HM style type inference might become undecidable.

You should also ask “Does my language need subtyping such as subclasses?” And if the answer to both is yes, you should probably forget about Hindley Milner, or at least pick something far away from it on the spectrum.