I am looking forward to smart highlighting that does what you say (marking errors and inconsistencies, like perhaps a truthiness check being used with a variable that could be both null and undefined in case of a typed language) and otherwise is based on context (for example, highlighting a variable that was defined in outer scope is something that could be legitimately useful).

Colouring human language by tokens is a very fitting analogy.