A sandstorm in Phoenix. A camera that can barely see anything. A lidar sensor that, instead, clearly detects a pedestrian standing at the side of the road. That's the example Dmitri Dolgov, Waymo's co-CEO, shared on stage at Y Combinator's Startup School to explain a technical decision that is actually about much more.
A difference that goes beyond technology
Waymo uses cameras, lidar, and radar together. Tesla has chosen to rely exclusively on cameras through "Tesla Vision," after removing radar and ultrasonic sensors in recent years. Dolgov never mentioned Tesla during his talk. He didn't need to. The point he raised is simple, almost obvious: eyes are enough to drive like a human. They are not enough to drive better than a human, reliably, millions of times over, with no one ready to take control.
For now, the numbers appear to support his argument. Waymo reports more than 500,000 paid rides per week and over 220 million miles driven in fully autonomous mode, with a serious accident rate far lower than that of human drivers. Tesla's robotaxi service in Austin, by contrast, has recorded around 14 incidents over 800,000 miles, with a human safety supervisor always on board. Tesla says its vehicles have driven 380,000 driverless miles since the service launched. Waymo covers that distance in a single day.
The cost of the last nine
There is a concept that often comes up among those building autonomous driving systems: the exponential cost of each additional nine. Reaching 90% reliability is relatively easy. Every extra decimal nine costs significantly more than the one before it. A good driver-assistance system can work with just a few nines. A car with no one behind the wheel, carrying children in the back seat, needs many more. It is the least glamorous part of technological progress—the part you don't see in promotional videos, but that ultimately has to prove itself on real roads, over many years.
Meanwhile, Tesla is facing a growing number of lawsuits, a California ruling over misleading advertising related to Autopilot and Full Self-Driving (FSD), and a federal investigation involving nearly three million vehicles. None of these issues, on their own, prove that cameras alone are insufficient. Taken together, however, they paint the picture of a company that is still testing the limits of its technology on real users while continuing to promise full autonomy.
Who pays the price for trust?
Meanwhile, the rest of the ecosystem is moving quickly—and in different directions. Waymo is importing compact Zeekr vans from China despite tariffs that have tripled their cost, simply to secure a vehicle designed specifically for autonomous driving. Uber has put more than $10 billion on the table to become the infrastructure platform on which other companies will deploy their robotaxis, while reportedly considering ending its partnership with Waymo by 2028. Each player is betting on a different piece of the same future: some on sensors, others on platforms, others on scale.
Beneath the rivalry between these companies lies a question that concerns everyone who will eventually step into one of these vehicles. The issue is not so much whether autonomous driving works, but how willing we are to trust a system we cannot independently verify—one developed by companies that also compete over the narrative of how safe their technology really is. The difference between an impressive demo and a service that genuinely protects people isn't visible from the passenger seat. It only becomes clear afterward, when something goes wrong and we have to determine who—or what—should have recognized the problem first.