Fun thought experiment. What if they would try to do this for the Tesla Bot Gigafactory? Teach one bot how to assemble one bot. Currently they have humans doing it, so it should be possible. Have KUKA robots etc make the parts needed, then let optimus bots assemble them. Might not be the most efficient way to assemble them, but at least it would scale very fast. If it takes one day for one bot to assemble one bot, the doubling time of their assembly would be 1day or 1000x in 10days.
We will eventually end up there, the relevant question is the most optimal path.
For bots and cars, I think we will see assembly tasks designed for humans, initially done by humans, and with bots slowly being phased in.
Say a line with humans can assemble 100 bots per day, as bots are phased in the aim will be to keep the same run rate.
Then Tesla can up the run rate by speeding up the bots, or duplicating the line.
Once Tesla is mostly using bots, a large factory floor space with duplicate lines is the most optimal solution.
What this probably means for most human factory workers is more variety of work, more time spent solving problems, training bots, identifying problems, monitoring bot performance, quality control, inspections, scheduling etc.
In the end there should be very little need for humans to regularly do simple repetitive tasks. The manual jobs humans end up doing will require judgement, skill and dexterity that robots can't currently achieve, and that bar will probably get progressively higher.