That's a strawman argument. I'm just saying that we need to find ways to wrangle the mountains of code coming out of LLMs into something we can trust, as best we can.