Source code is already a serialization of an AST, we just forgot that and started treating it as text. The practical problem is adoption: every tool in the ecosystem reads bytes.