Trivial parsers were trivial before LLMs, too.
Ironically, using the LLM to write trivial parsers loses out on the best part of the LLM: its ability to parse natural language into some sort of structure.
Using the LLM itself as the parser frontend seems more interesting all round...