That's not a "best case demo": it's a real system, working in real world environments, and even if they're "constrained" (and go ahead and look at the Phoenix service area and tell me how constrained it really is), they are functioning, now, today. There's one driving by my house as I type this. It's not mowing down children. It's not causing fatal accidents. Its incident rate is dramatically lower than human drivers, and its severe injury rate is lower still. We're a decade into the hype cycle because of how hard a problem this is to solve, and we're finally catching up with the right confluence of technologies at the right time (mostly around machine learning, machine vision, sensor fusion, LIDAR, reinforcement training, and computer power) to make it actually work.

I can understand if you live outside a Waymo market that you might still believe this is still fairy tale "won't work" stuff, but when you live in a market where you see dozens of them every single day, doing their thing _unremarkably_, it's... well, it feels quite a bit like the future.

Note that I'm speaking explicitly of Waymo here. Tesla FSD still terrifies me, vision-only seems like a horrible oversight, pun intended, and while it's meant to be non-constrained, it still has a very, very far way to go to close the gap with Waymo.