In such case the density would probably be large enough to efficiently match e.g. 4 people to a car.

Commutes are predictable (people do the same exact trip at roughly the same time) so easy to optimize.

You can therefore pack more people in the same fleet during peak times than during non-peak times.

The second obvious step is to create additional fleet made from half-buses like the 12 seater for peak traffic, a better Chariot.

Surge pricing can be flattened out by making these autonomous vehicles convertible and capacious. During peak commute load, they function as high-density passenger vehicles. During low commute load, they function as light freight vehicles -- courier services, delivery services, even mail.

I could even see someone like Amazon running their own fleet of delivery vehicles and offsetting the maintenance and management costs by letting the vehicles run in "taxi mode" (with flip-up seats) during commute times. Think AWS for vans/light trucks.

In such a scenario, they wouldn't even need to necessarily charge market rates to commuters. They could offer the service below cost and still make money, as their primary profit driver would be their delivery services.

The sheer number of parties that stand to gain from this makes it almost inevitable, to my mind.

It's already close to that cheap, where public transit is relatively expensive and rideshare is relatively widely used and thus cheap. As soon as you take human drivers out of the scenario it gets even cheaper, of course.