A look at the hidden complexities of an autonomous taxi service, after taking a fully driverless ride, now being tested as part of Waymo One service in Phoenix
After more than a decade, Waymo's driverless ride-hailing service is open to customers — Ed Niedermeyer is an author, columnist and co-host of The Autonocast.
Context & Ripple Effects
The road here was long and staged: Waymo offered free fully driverless rides in Phoenix back in 2017, though most cars carried an employee in the back seat, then opened Waymo One commercially to Early Rider members in late 2018 — a debut captured in early test-ride impressions. Six months into the paid rollout, reporting still found safety drivers at the wheel for most rides, making this account of a genuinely empty driver seat the first proof the service has crossed over.
What Ed Niedermeyer's ride exposes is that 'driverless' does not mean 'human-free': the relationships around this story describe workers paid $20–$24 to close doors manually and $60–$80 tows dispatched through the Honk app, plus roughly 25,000 trained humans supporting around 3,000 robotaxis — the operational substrate a customer never sees.
First-order effects
- Phoenix riders can now hail a car with nobody up front, but every such trip leans on an unseen support layer — remote assistants, door-closing crews, and Honk-dispatched tow operators — that Waymo funds per incident rather than per vehicle.
- With only about 200 of its vehicles running without human monitors out of a much larger fleet, Waymo is effectively operating two tiers of service simultaneously under one Waymo One brand.
Second-order effects
- Scaling beyond Phoenix forces Waymo to industrialize its human-support apparatus — the 25,000-people-trained-to-3,000-robotaxis ratio suggests headcount grows nearly linearly with fleet size, pressuring the unit economics of driverless rides.
- Rivals are exposed by comparison: experts cited in related coverage judge Tesla's Austin robotaxi operation as significantly trailing Waymo, with far fewer vehicles and reliance on safety drivers, so the bar for 'driverless' claims is now set by what Waymo actually ships.
Third-order effects
- If fleets scale on this model, robotaxi reliability events become human-workload events — Waymo's own post-incident report found supervisors overwhelmed by confirmation prompts during a San Francisco power outage, hinting that oversight capacity, not driving software, may cap how large these networks can grow.
- The industry's competitive framing shifts from 'does it drive itself' to 'how few humans per hundred cars' — a metric regulators and insurers are likely to demand once the hidden-labor structure becomes widely understood.
The trend: Commercial robotaxis are crossing into true driverlessness while quietly building a parallel human support workforce, making operational labor rather than autonomous software the binding constraint on scale.