> I can guarantee you have never used an LLM to code anything seriously large, complex, or novel
I'm not claiming to have done so, I'm saying almost all of the business ideas in the current bubble are so small and trivial they're also easily duplicated by an LLM.
I agree with the implication that where LLMs are at their best is precisely when they're not doing something large, complex or (particularly) novel.
But most of the ideas being tried out are also not large, complex or novel — just as the dot-com bubble had "making a loss on every sale but making it up with volume" and then the mobile app era had "now you can order your pizza on your phone" (like you couldn't already even without an app), the AI era has many very poorly thought out ideas that are easy to replicate.
I can guarantee you have never used an LLM to code anything seriously large, complex, or novel