Using an LLM as a live runtime would always be a wildly inefficient and less reliable way to perform routine, deterministic tasks compared to traditional code.