You're right, doesn't sound great. In the example they shared, sounds like the issue wasn't that the car couldn't drive around the fire truck, but that it couldn't break a programming rule about crossing into a lane that would normally be opposing traffic. Once given the "ok" to follow such a route, the car handled it on its own, the human doesn't actually drive it.

I could imagine a scenario where you need one human operator for every two vehicles. That's still reducing labor by 50%.

Obviously they want it to be better than that, they want it to be one operator per ten vehicles or no operator at all.

And the fundamental problem with these systems is they will be owned by big corporations, and any gained efficiency will be consumed by the corporation, not enjoyed by the worker or passed on to the customer.

But I think there's true value to be found there. Imagine a transportation cooperative - we're a thousand households, we don't all need our own car, but we need a car sometimes. We pool our resources and have a small fleet that minimizes our cost and environmental impact, and potentially drives more safely than human drivers.