The syntax shown is completely trivial, it should be parseable with about a dozen lines without using any generators, frameworks, libraries, string replacements, or anything else.
Unless I missed something.
You didn't miss anything. The point (or "the only point") is that he was able to reuse the language's own parser rather than writing any parsing code of his own. For anyone who hasn't seen that technique, it's worth seeing. Whether it's a win depends on taste maybe.
I've done this before with Python, making AST literals directly out of tokens. It's potentially dangerous however, similar to eval-ing input. It can be done safely with care.
You can use ast.literal_eval[1] which is supposedly safe (only accepts literal values, not code).