But programmers are LLMs augmented with the ability to run code. It seems odd to add a restriction when testing if an LLM is "as good as" a programmer, because if the LLM knows what it would need to do with the external code, that's just as good.