I believe it'd be more precise if it used an appropriately chosen and applied LLM in combination with web research – in contrast to juggling together some LLM generated code.