My current thinking on the matter is, would be neat, it might be possible, but IDK whether it would be useful. If types are inferred completely, i.e. there's no actual type definitions, perhaps things not even having types themselves, but rather just rules extracted and tested such that the compiler evaluates and says "yup you can pass X into F" (which may be all you can say in cases like the omega combinator--there's no "type" you can associate with it afaict), then there's nothing to apply type classes to. "Can't have type classes, without having types". Also it takes away one of the big advantages of defined types (records / classes), in that you can do `.` and get precise autocomplete. If everything is just inferred, then the editor can't help you out as much.
