A look at Waymo's driverless taxi service, in Phoenix suburb Chandler, which seems to work well but still only offers ~100 rides/week and limits cars to 45 mph
Timothy B. Lee / Ars Technica :
Context & Ripple Effects
Two years after Waymo's Waymo One commercial launch in Phoenix — which itself followed a year of free fully driverless rides with employees in the back seat — Ars Technica's Timothy B. Lee revisits the flagship Chandler service and finds it functioning but small: roughly 100 rides a week, cars capped at 45 mph, and about 200 vehicles running without human monitors.
The piece also pulls back the curtain on the operational overhead behind the 'driverless' label, echoing the hidden complexities documented in 2019: Waymo trains some 25,000 humans to assist roughly 3,000 robotaxis, pays workers $20–$24 per door-close and $60–$80 per tow via the Honk app, and its supervisors were overwhelmed by confirmation prompts during a San Francisco power outage.
First-order effects
- Waymo's Chandler service works well enough to carry paying riders driverless, but at ~100 rides/week and a 45 mph cap it remains a bounded pilot, not a scalable product — the constraint is operational scope, not rider demand alone.
- The human-support layer is now a visible cost line: door-closing at $20–$24 and tows at $60–$80 through Honk mean each 'driverless' ride still carries labor the taxi economics have to absorb.
Second-order effects
- Competitors are being measured against this bar: experts cited in the corpus judge Tesla's Austin robotaxi effort to significantly trail Waymo, with far fewer vehicles and safety drivers still in the loop — Waymo's Chandler operation is the reference point rivals get compared to.
- The gap between ~200 monitored-free vehicles and ~3,000 total robotaxis shows the bottleneck for scaling is the remote-assistance and roadside apparatus (Honk-style towing networks), not just the driving software.
Third-order effects
- If the pattern holds, robotaxi economics will be decided by 'AI cost per useful task': the unit cost of a completed ride including human support staff, not the headline absence of a driver — favoring operators who minimize interventions per mile.
- The 45 mph cap and geofenced suburbs point toward capability tiering as the industry norm: services expand street-by-street and speed-band-by-speed-band as intervention rates fall, with regulators (already flagged for lack of transparency back in 2018) likely to formalize such operational design domains.
The trend: Robotaxis are proving out the technology in narrow, speed-capped zones while their economics hinge on shrinking the human support apparatus per ride — the gap between a working pilot and a scalable network is operational, not perceptual.