If robotaxis take over the entire market for cars, and that's a very big if, it will take at least a decade, quite possibly longer. In the meantime, many companies will have learned how to make a robotaxi, and not just the ones that have already started development today. The vast majority of what companies trying to make a robotaxi are spending their time on today is learning how not to make a robotaxi. This is important work, but it's also work that will very quickly become non-proprietary. Sure, you won't be able to steal Waymo's code (if Waymo comes out with a self-driving car first). But many many people will know what works and what doesn't, and much of that knowledge will be impossible to protect. Furthermore, hardware is increasing rapidly, and increases in hardware make the software much much easier.

Furthermore, the very hypothetical that robotaxis are going to start taking over the market for cars any time soon is, I think, an unlikely one. Robotaxis are only even remotely close to becoming a reality in very limited areas. Even there, they don't really offer a whole lot of benefit, because they're not truly autonomous. There's either a "safety driver" in the car, or someone in a control center monitoring things. For long-haul trucking such remote-controlled vehicles can save significantly on costs. For taxi service, where you're competing against a gig worker making less-than-minimum wage, maybe not.

You say that Google judged that level 3 was not a good idea. But how is Waymo's robotaxi system, which sometimes asks humans in a control center to make driving decisions, any different from level 3?

I can see why Google previously judged that level 3 was not a good idea. I think over the several years since they made that judgment things have become more clear. Is Audi's level 3 Traffic Jam Pilot not a good idea? It seems to be fine, in terms of design (I don't know if it is sufficiently bug-free). As long as it works as design, it's relatively non-risky, and the only real risk is when a driver is being extremely negligent. I think at the time when Google judged that level 3 was not a good idea, the picture we had of level 3 was fairly different from Audi's Traffic Jam Pilot.

I think Level 3 can work as long as things are safe while waiting 10-15 seconds for the driver to take over, and relatively safe even if the driver never takes over at all. And I think most of the really difficult problems of self-driving can fit into that, because they are usually situations where a human would be more aggressive than a car designed to follow the rules.

I think the only way you can say that Google's system is completely different from level 3 is if you say it is also completely different from level 4. I think if you had to give it a level, it would have to be level 3, because it requires humans to make driving decisions at unpredictable times when it is not properly parked.

I think that's reasonable, because I don't think that the "level 3" we are going to see is anything like the "level 3" that some people envisioned just a few years ago. There certainly will be no need to trust the driver to take over reliably within 10 seconds. In fact, the vast majority of "take overs" will consist of a human solving strategic problems when the car is stopped. Maybe that's not "level 3." If the car is stopped in traffic where it isn't supposed to be stopped, I wouldn't call it level 4 either, though. If what you're saying is that the very concept of the levels are poorly defined, I agree.

(Do you really think it's going to be acceptable for Waymo vehicles to stop in the middle of the road for significantly more than 10 seconds? You complain about your Tesla stopping and blocking traffic in a parking lot for just a few seconds.)

When (I mean if) the change comes to robotaxis, it will be much slower than you imagine today. It's not at all clear that the market for cars will go into decline for many many years, and even once it does, that's not going to spell doom for all of today's car companies, just the ones that can't adapt. Someone is going to have to build the robotaxis, and it's not going to be Waymo. (It's not clear that Waymo will play any role, in fact. They're good at R&D, but other companies will be better at production, and other companies will be better at logistics, and other companies will be better at operations. The car companies (and even the ridesharing companies like Uber and Lyft) will be much better at collecting data to feed into neural networks. Waymo is the Xerox PARC of the self-driving car world. Their best hope is probably that they can get bought out before their technologies become ubiquitous.)