I see many debates on this and other threads about whether a system is truly "full self-driving" if it requires human intervention in extremely rare cases. I think this misses the point. A system that requires human assistance only one in a hundred times is still automating away roughly 99% of the human work.
When customer support is 99% automated with the remaining 1% handled by remote humans it is often economically and societally the same as 100% automation.
The same applies to self-driving cars. If there's 99% automation and humans can handle the remaining 1% remotely, and if that combined system (including the handoff process) is safer in aggregate than a standard human driver, then you basically don't need drivers anymore.
You don't actually need 100% AI-only automation to make dramatic economic and societal difference. You just need 100% coverage through the human-AI combination.