Elon was surprised too. The model 3 was supposed to have a fully automated line. He couldn't make it happen. Industrial robots are bad at wiggling stuff into place, and hopeless at handling error cases. https://qz.com/1261214/how-exactly-tesla-shot-itself-in-the-...
Elon is the biggest day dreamer who imagines everything can be solved with technology around the corner in 6 months, but has a hard time acknowledging when the harsh realities hit him in the face, that real world engineering challenges aren't as simple as those in building successful websites.
He's gotten further with subsequent products. A lot of the learnings with model 3 went into the model y and later iterations of the model 3. And of course cybertruck.
Industrial robot usage and other manufacturing innovations are a big reason they are leading the industry in manufacturing cost and speed. Other manufacturers are starting to emulate a lot of the things they are doing with structural battery manufacturing, casting big parts, etc. The likes of VW and others are (by their own admission) behind and struggling with the number of parts, amount of manual work, etc. It might not be as automated as Elon Musk would like but it is pretty impressive.
It will be interesting to see what they do with their upcoming cheaper model. He seems to be hinting at the notion that further manufacturing improvements are going to be a big part of that.
The one that really surprised me for some reason was PCB manufacturing. The Strange Parts YouTube channel has a few tours of Chinese factories, and I guess I pictured that it's just some sort of giant clean room with a long conveyor belt that goes through all the processes on its own. So I was not expecting it to start with grabbing a blank PCB from a pile stored outside and taking a band saw to it. Such a mix of high-tech automation and human hands picking things up, racking them on the table, and sticking them into the next machine.
The board level electronics assembly is mostly an solved problem that can be automated perfectly fine, but when you get to through-hole almost nobody does it, because (a) the THT capable pick and place machines are uncummon and thus somewhat expensive (b) it often requires more expensive parts that are compatible with such machines (better geometry tolerances, surfaces for the machine to pick them...) and finally (c) the NRE costs of getting that to work.
That said, then there are niches where the automated THT assembly apparently makes sense and all the NRE costs seem to be offset by lower per-unit costs of the PCB material itself. Prime example is mid-range to high-end home AV equipment (Japanese branded if not made in Japan outright), where the large mostly empty single-sided FR-2 boards with truly ridiculous amounts of obviously machine populated wire-links and discrete THT radial components are particularly striking.
I was surprised to learn that the production of stamping tools, used to shape panels from sheet stock, involves lots of manual passes to bring them to the quality and finish required by car manufacturers.
After machine milling them roughly to spec there are employees that angle grind, sandpaper and polish every critical surface of these 10 ton chunks of metal.
People think automation solves problems - it's the opposite. In order to use automation, you have to solve every problem that the automation could encounter.
It's not clearly shown in this, but the majority of the work is manually adding parts.