I feel the need to correct the record. You do the things that don't scale in order to better realize/appreciate where to put the effort to fix them. Tesla has the equivalent of an mvp/poc production line. You'd be kidding yourself if you think they're not improving it every minute. And you know the improvements they're making are the ones with the biggest impact, in that exact order of priority.
If you try to make the perfect assembly line before you deliver 1 car, you'll be making huge over-engineering expenditures and costly assumptions which result in even greater net production losses.