>Surely all that would really change is when buying a new car, you are purely insuring the vehicle and not the person!
Thats how insurance works in the state that I live in.
One minor problem is I'm not sure exactly what the average driver experience age is, but I know new teen drivers are incredibly expensive to insure and after 10-20 years of primary driver experience cars start getting pretty cheap to insure (until your kids are teens, of course).
Changing a cars primary driver without informing the insurance company will get you cancelled. I can't basically gift my old car to my teen cousin while paying "safe old man rates". They'd be more than happy to insure my old car, at "crazy teen boy rates". So I suspect a software upgrade of the driving system would require insurance company contact.
Observationally, today's github release will be insured at a small integer multiple of "crazy teen boy" insurance rates.
This develops into other problem areas. At some crossover point its going to be cheaper to hire a human driver... My gut level engineering guess is it'll never be cheaper to run an autonomous car on shared roads than to run an autonomous train on dedicated (elevated?) rails, so throwing capital at the problem is a losing strategy, you'd come out way ahead financially building elevated monorails or just going with modern intermodal trains.