The steelman argumentation is probably that spaghetti doesn't matter to LLMs and no humans will read that code anyway.

Do you feel like this is a very meaningful comment? Will someone go "mm yes, actually it's spaghetti, I didn't think of that.."

Are you trying to succinctly say AL'S spaghetti code outweighs the benefits of what it produces quickly?

If you're not saying it outweighs it, what are you saying?

I mean you can just pause and refactor regularly. I tend to use every ~4th session as an opportunity to refactor, adjust interfaces, break overly large modules into smaller ones and suchlike.