The thing is it's increasingly about the process of stringing things together, rather then the stages themselves. Obviously, not all the stages in your video are in the one video: it's from at least 7 years ago from a company that makes the specific machine class, so I can see why you might be confused about that.
The thing that no one has cracked yet is that sitting 30 people down at a belt and telling them to follow the manual is easy and flexible and while it's a skill, it's not very technical. Designing the line to do the same thing is hard, fiddly and very, very technical. Whoever does that needs to know the machines available, the sensors, IO, networking, wiring, power, PLCs, HMIs, access paths, be able to design and spec anything custom, etc etc. You also need to feed back to the designers if one aspect is hard to automate (poking through wires is a good example). Even loading workflows: You can't just shout "duo kai" at the machinery as you trolley a palette of magnets to the head of the belt and expect them to get out of the way.
That, combined with the outlay of the machines themselves, limits things you can make with it. The kind of factory in your video seems to a very typical Chinese SME where there are 50-200 people in a simple warehouse with some basic equipment and they contract assemble. These places do not generally run the margins you need to stop production, acquire several million in automation and design services, and start up again.
Fully or just highly automated systems thus get used for high-volume stuff that doesn't change much in a run. Cars, phones, that kind of thing are the classics. Food production also due to the massive volumes.
The floor on where it makes sense to even think of automation is, however, lowering quickly. It's not just Siemens NX that can do it any more, factory management platforms are sprouting from everywhere from Huawei's ERP to Hangzhou student bedrooms, and there are other SMEs starting to do this work at lower costs for smaller outfits.
As long as manual CMs exist and are cheaper, CheapNShit Speakers doesn't need to design its product for automatic assembly, even though it definitely could if it needed to. That creates a kind of "snap to automation" where something that wasn't ever automatable suddenly gets automated, because when the machinery is close enough, the last few sticking points are designed away and suddenly you can do it.
Like I said, the real acid test of when this becomes really useful (or dangerous to the workers, depending on perspective) is when you can quickly reconfigure a production line to make the Black Friday CheapNShit 2000+ speakers rather than the ones you made before. The CMs currently will do that with a WeChat message, a new manual and a palette of parts and have them on a truck tomorrow.
This is likely a place where "AI" will find a home, whatever happens to the chatbots. It's things like allowing a fairly generic robot to pick up a new kind of speaker case and put it on something else without having to have an engineer write a program or even G-code to do it.
Also like AI, this isn't quite like last time when the US lost its manufacturing to China. The US lost low-medium manufacturing to China based on, basically, labour costs. Then it lost the high end on labour and automation that means as long as you have the machine you don't need the expert US workers and network effects because everything you need for that is available within 2 hours in Shenzhen.
China losing the middle-high end to cheaper places implies that automating most of it will remain so hard that it's always more expensive in TCO terms than the cheapest global labour. If labour is only a small part, this is not necessarily true or it may be true only for some products (clothing is a good example of something traditionally hard to automate fully).
On the other hand, automation getting to that state also means the US, or anywhere, can get the manufacturing back if it wants to, by competing not on labour costs, which is unlikely within the next decades, but on the technology and network effects that make it possible to acquire, stand up and run these kinds of line quickly.