Just because a few companies failed at factory automation in 1980 doesn't mean it couldn't succeed in 2018. It's patently absurd to knock Tesla for trying it again in the era of ultra-high resolution cameras and advanced machine learning algorithms.
But it goes beyond just the fact that technology has advanced: Tesla is willing to pursue alternative approaches where others have long stopped (and this is a point the author of this article makes later on). The fact that Tesla doesn't have the conservative culture of other car makers means that they'll make mistakes like this, but also that they'll find new solutions (or recognize existing solutions in similar fields that haven't become industry standard) whereas others would not.
You cannot automate EXISTING car designs much. And Model 3 does make improvements, but in many ways is built similar. It's like trying to automate soldering components by hand. Automation of electronics required a change in the fundamental way electronics were built. Through-hole is hard to automate but relatively easy to build by hand. Surface mount is simple to automate and ultimately better in several ways, but is super annoying to build by hand. Tesla needs to find the surface mount of car manufacturing. And I think they're trying.
Something particularly hard to automate for automobiles is the wire harness. There are many degrees of freedom; it's kind of like tying your shoe: easy for humans, hard for machines. The Model 3 uses less wire than the Model S/X, but only by a factor of 2 or so. Model Y, on the other hand, is supposed to use like an order of magnitude less. This is like optimizing a circuit board for automation by only using a few through-hole components, relying mostly on surface mount.
So Tesla is going to have to continually redesign their vehicles to be more and more amenable to automation. They can't do this as a step function, and Musk has realized that now. (This is another good point the author makes.)
And don't give me that crap about R&D blah blah blah. Other car makers include R&D and other related capex in their per-car profit accounting.
Note: Most other car companies make hundreds of thousands of cars at each of their factories, and sell hundreds of thousands to millions of cars each year, which substantially reduces the per-car capex. Tesla sells comparatively very few cars, so its per-car capex dwarfs the rest of the industry.
But they're still ramping up. The factory will be outputting something like 500,000 to 700,000 cars per year after they're done ramping. They'll need Model Y and/or Semi and at least another factory before they're going to be comparable to other mid-range manufacturers. They'll need at least half a dozen factories to be a major manufacturer. Or they'll need to change how cars are made entirely.
If they can get millions of cars from a single factory, then the high automation that they're shooting for (and which Model 3 had to back off of) might make sense (and is required for getting that kind of production from a single line). But even that will likely need multiple factories to justify the R&D into the factory line itself.
That's only true if you're a time traveler from the future. But it's only April 2018 here, and Tesla is producing about 2600 cars a week. They produced 12000 cars during the first 3 months of the year.
https://www.bloomberg.com/graphics/2018-tesla-tracker/
Also it would be very difficult to get millions of cars out of the Fremont facility.. they already ran out of space there over a year ago.
although q1 production was 34k x 4 = 136k.. so they still have some growth before it's hundredS.
Both are important, but if you're looking to invest in a business that's growing, you care more about the first number because you're assuming that if the business grows enough then the second number will take care of itself. If Tesla has been in business for 100 years and is unlikely to grow its base, like more established automakers, then you care alot more about that second number. And you also care about a lot of other numbers like the dividends paid out, how likely that something will change to move the stock price, etc.
Tesla R&D isn't just about creating yet another model, but about creating the very first car model in that category that they have ever done.
I think it would be very odd if they kept having equally high R&D when upgrading their models.
"ultra-high resolution" doesn't mean much if you can't see the problem.
"advanced machine learning" sounds dangerous once the machine damaged itself or killed a bunch of people trying to figure out things.
Given a 99% chance of working for every single process for hundreds of processes you end up with... well... not much output.
When I was younger I had a less severe but equally enlightening experience. A transmission case line with 24 CNC stations had our UI software running on PCs to make them easier to use. They could run fine without the fancy UI but it was how the operators had come to run things. I went in one afternoon to load a software update from floppy disk. The plant manager came out. "What are you doing?" "I'm updating this UI software." "You work for $company?" "Yes." "Is there gonna be a $company guy on site for 3rd shift tonight if something goes wrong with that software?" "No." "You aren't gonna update it then." I went home.
Go fast and break things doesn't work in that environment, and yes the critical processes work far more than 99% of the time because as you say there wouldn't be much output. It sounds like that's the exact lesson Tesla is learning now.
[edit] the reason all the processes on the line have to work is because it's basically impossible to store a bunch of cars (or doors or engines) anywhere until things get fixed. Ad-hoc fixes are just not feasible at that scale.
ps: not a Musk fanboy, just a middleground spectator.
Is it really time to say this?
Sure we'll know after more samples but .. well I said it.