>"but that you can manage with software"

This is a take that seems accurate on it's surface but belies a lot of additional problems. Leaning on software for improving reliability can be a dangerous path. It's tempting because it's cheap and easy to iterate on. Software doesn't wear out like mechanical components. But software makes reliability pretty tough when you constantly iterate because it tends to fail at interfaces. Not only do the number of these interfaces tend to grow exponentially, they are notoriously difficult to test fully. Nancy Leveson has some good articles that demonstrate this distinction.

A automotive example from my own experience is a comparison with mechanical vs. electrical robots used in the stamping process of body panels. Plant managers like electrical retrofits because they held the promise of being much cheaper to build and modify. But they needed near-constant tweaking of the control system. The number of interfaces causes a lot more failure modes and impacts the reliability. By contrast, a mechanical system is, by it's nature, constrained by mechanical linkages. Once the mechanical system is tuned and working, it works for a long time. Gradually the parts wear, but they tend to be much easier to predict and troubleshoot.

Since this thread is about aerospace, you can see where software failure caused catastrophic failures by lulling engineers into complacency about how software can enhance reliability. From Ariane V, to Falcon, to CST-100 all have examples of how software did not fulfill its promises to increase reliability and often it does the opposite. Don't confuse the ability to move a design change faster with an inherent ability to make things more reliable.

But that brings ups up a specific example of where Tesla has struggled. They used to keep HUGE backstocks of excess inventory (I believe they have since tamed this) which is a classic manufacturing mistake. So if you have any production mistakes or improvements in the part, you have to scrap or just cycle through the entire inventory until the improvements show up in the car.

This is why Just-In-Time is such an important concept in manufacturing. You can iterate constantly. Meanwhile, Tesla would crank out huge batches of identical fenders and doors and mirrors to maximise efficiency (economies of scale) but then struggle with part fits and panel gaps.

Tesla has since dealt with a lot of this when they went through their production hell period. But a lot of what they had to do was re-learn things like Lean Manufacturing and JIT.