You need some sort of a tag if you want to pattern match at runtime, which is the usual use case for tagged unions. Now, a tagged union like

    type Option a = Some a | None    # here Some and None are tags, not types themselves

could also be written as an untagged union by using types as tags:

    type Some a                      # a wrapper type for an a
    type None                        # a unit type
    type Option a = Some a | None

but for that to be useful you still need to be able to check at runtime whether you have a Some or a None… which means that the tags are still there, just hidden by the language.

But the more typical use case for untagged unions is different: given a type such as

    type StringOrInt = String | Int

you're not even supposed to be able to check whether you in fact have a String or an Int, rather the language only allows you to use the subset of operations defined for both String and Int. In other words, slightly ironically, the type StringOrInt exposes the intersection of String API and Int API.

There are many benefits of sum types:

- Their values can be discriminated at runtime (it's depending on the implementation of union types whether that's possible. In C a union is just a structure large enough to hold all constituents)

- You can sum the same type (union(Int, Int) simply is Int)

- The tag breaks recursion. You can build recursive sum types because of it, e.g. 'a List = Empty | Cons('a, 'a List)

Maybe<T> is a tagged union between T and None.

Nullable<T> is an untagged union between T and None.