That's a false dichotomy, because transistors and ICs are manufactured to be deterministic and nearly perfect. LLMs can never be guaranteed to be like that.

Yes, some things are better when manufactured in highly automated ways (like computer chips), but their design has been thoroughly tested and before shipping the chips themselves go through lots of checks to make sure they are correct. LLM code is almost never treated that way today.

Yes, the point is that only if you're willing to accept crappy results then you can use AI to build bigger things.