Nothing. If you accept only workable specific subset of Unicode and validate input, you can get the point back. Using utf-8 encoding for text is almost always the right ting to do, treating it as generic Unicode string is usually wrong.

Unicode is as complex as UTF-8 encoding is simple an neat.

Joel Spolsky studied Unicode and was so sure that he understood it that he wrote a short intro for programmers without knowing that code points in Unicode don't correspond to "platonic ideal letters" aka user perceived characters. They correspond only in every case he used them (user-perceived character may actually be a sequence of code points).

Yeah. There's probably all kind of obscure use cases for ASCII, but generally, any text meant to be human readable should be Unicode encoded, whether you call it "string" or "blob".

In the case of Postgres, you need string types if you intend to do any string operations on those strings in the database. As mentioned in the article, you also have no choice if you want to use jsonb.