The Driverless Taxi's Hidden Workforce

The Empty Seat, The Full Depot

Drive up to Waymo’s Tolland Street facility in San Francisco around 6 p.m. and you’d swear you were walking into a truck terminal. The iron gates swing open every minute or two. White Jaguars and the newer Ojai models roll in, idle, roll out. In 20 minutes, a reporter counted 13 vehicles cycling through.

There are no drivers in those cars. But there are at least 60 to 100 people inside that fence — and they’re why the robotaxi dream is far more physically grounded than the software brochure suggests.

This depot, the largest of its kind in the U.S., now holds 159 charging spaces plus another 50 nearby. It was a truck terminal before Waymo retrofitted it in 2020. Today, workers in safety vests move between vehicles connecting chargers, wiping interiors, checking cameras and sensors, diagnosing software faults. One technician was troubleshooting a camera malfunction right at the wheel — the empty seat was not an absence but a placeholder for a different kind of labor.

The implication is blunt: the technology promised to erase drivers only displaced them into a wider, heavier support network. Remove the human behind the wheel and you need humans behind the charger, the mop, the diagnostic laptop.

The Bottleneck No One Is Talking About

Waymo now operates in 14 cities. It launched full driverless service in Denver, San Diego, and Tampa starting this month. Amazon’s Zoox has gone paid in Las Vegas and is expanding into San Diego and Houston. Tesla is pushing its robotaxi ambitions forward. The software competition is visible. The infrastructure competition is not.

Axios reported that companies including Waymo, Tesla, and Uber have been quietly competing for the right plots of land long before they announced any service. Securing land near dense urban cores for depots, charging hubs, and maintenance facilities is proving harder than building the cars.

Here is the number that should give policymakers pause: building the supporting infrastructure for a new city takes more than three years. That is longer than most election cycles. That is also longer than the hype cycle governing this industry.

When demand concentrates in downtown cores where land is most expensive and community opposition most organized, the math gets worse fast. A robotaxi fleet needs constant recharging, cleaning, and repair. Those activities cannot happen in a server farm. They require square footage, loading bays, power capacity, and proximity to the streets where the cars operate. That combination of attributes is vanishingly rare in a major American city.

The New NIMBY Facility

Charging depots are noisy, bright, and active around the clock. Vehicles roll in at dawn and roll out at midnight. That is not a design flaw — it is the operational rhythm of a service that must be ready whenever a rider taps their phone.

San Francisco residents learned this in 2024 when unoccupied robotaxis idled in a lot and blasted their horns into the early morning hours. Last year, in Santa Monica, a nighttime Waymo charging facility near homes drew a lawsuit. In July, a judge issued a temporary restraining order, finding that the noise, lighting, and vehicle traffic from the facility inflicted significant harm on nearby residents. Nighttime operations were halted pending further review.

The Tolland Street depot was deliberately sited away from housing, surrounded by warehouses and industrial use. It is enclosed by a three-meter fence and black privacy screening designed to keep interior lights from spilling outward. That is not an accident. It is the blueprint for every future facility.

This is the same dynamic that has turned data centers and electric vehicle battery plants into flashpoints for local opposition. The new NIMBY facility is not a tech campus hidden in a suburb — it is a heavy industrial operation disguised as a software solution.

What This Means for Korea and the World

A Chosun Ilbo report tracking this phenomenon carries weight beyond its American setting because South Korea is simultaneously pursuing its own autonomous vehicle push and watching the U.S. trajectory with acute interest. Korean policymakers and automakers are evaluating whether robotaxis are a viable transport solution at home. The American experience offers a cautionary template.

Korea’s taxi market is densely regulated, its urban geography compressed, its charging infrastructure less developed than California’s. If the bottleneck in San Francisco — a city already struggling with EV charging availability — is land and community opposition, imagine the constraints in Seoul or Busan where depot footprints compete with some of the highest land values on earth.

The lesson is not that robotaxis cannot work. It is that they will scale at the speed of civil engineering and zoning law, not the speed of software iteration. The companies that win will not necessarily be the ones with the best perception stack. They will be the ones that secure land, navigate municipal permitting, build community tolerance, and manage a workforce that is larger and more fragmented than the marketing materials admit.

The empty driver’s seat is real. The workforce that sustains it is invisible — until it becomes the thing standing between a city and a robotaxi service.

The Human Layer

The workers at the Tolland Street depot do not see themselves as relics of a transition. They are the operators of a new kind of transport system — one that happens to run without a person in the front seat. But their existence reframes the entire narrative.

Automation does not eliminate labor. It relocates it, often to sites that are less visible, less celebrated, and more constrained by the physical realities of land, power, and community consent. The robotaxi industry’s next battle is not over algorithms. It is over parking lots, neighborhood associations, and electrical grids.

That is a slower, messier fight than the one the software promised. And it is the one that will determine whether the driverless car ever becomes the driverless service anyone actually wants to use.