A robotaxi removes the driver from the seat. It does not remove the work around the ride. Someone still has to get the car ready, keep it available, respond when it stops, and bring it back when it cannot continue. The interesting question is how much of that work each additional vehicle creates.
The car asks for help without handing over the wheel
Waymo described its remote assistance system in February 2026. Its staff respond when the vehicle requests information. They do not continuously watch each car or drive it remotely. The vehicle can accept or reject their advice. A separate event response team handles more complex incidents, including coordination with emergency responders.
Waymo said approximately 70 remote assistance agents were on duty worldwide at a given time, alongside a fleet of 3,000 vehicles. It also reported more than 400,000 rides a week. Those figures describe Waymo’s operation on that date. They do not tell us the full labor required per ride: cleaning, charging, maintenance, roadside work, and customer support sit elsewhere in the system.
That gap is the metric I would want from every operator. Rides per remote agent sound impressive until a stalled vehicle needs a truck, or an open door needs a person at the curb. I wrote about Waymo paying DoorDash workers to close car doors. The task is small. Its existence shows how a physical exception can turn into a dispatch job.
Expansion multiplies the operating problem
Waymo said it began welcoming public riders in Denver, San Diego, and Tampa on September 1, bringing its count to 14 cities with fully autonomous trips. Access in those new cities was limited and expanding gradually. A city on a launch list is therefore not the same thing as an open service area for every resident.
Each city adds parking, charging, cleaning, towing, emergency response, and regulatory relationships. Local operating conditions differ. Waymo’s Munich plan starts with manual driving and mapping before a proposed public launch near the end of 2027. The sequence matters more than the launch headline: validate the streets, then build the service.
California permits offer another reality check. The DMV’s permit list distinguished testing with a driver, driverless testing, and deployment when I checked it on September 23, 2026. Tesla Robotaxi LLC appeared on the testing-with-a-driver list, while Waymo appeared in the driverless testing and deployment sections. A ride marketed as a robotaxi does not tell you which permission the operator actually holds.
Marketplaces sell the missing pieces
Uber’s second-quarter 2026 prepared remarks described its autonomous-vehicle business as a commercialization platform. It named fleet operations, financing, charging, insurance, policy work, and its rider marketplace as parts of that offer. This explains the Waymo–Uber split in Austin and Atlanta: the software that drives and the system that runs the rides can belong to different companies.
The economic test is whether those services grow more slowly than ride volume. Public trip counts show demand and deployment. They do not reveal the cost of a recovered vehicle, a cleaning visit, or an idle hour. California’s quarterly reporting page also notes that some Waymo deployment fields remain redacted while confidentiality claims are reviewed. Anyone presenting a precise industry-wide cost per driverless mile from public filings should explain what they had to estimate.
I would track three numbers as operators expand: paid rides per vehicle, interventions per thousand rides, and hours a vehicle is available for bookings. The first shows demand. The second measures exceptions. The third captures the unglamorous work between rides. Until those move together, removing the driver is only the start of the business case.