Which is precisely the issue with the idea of LLMs completely replacing human engineers. It doesn't understand this context unless a human tells it to understand that context.