Waymo, drawing on more than 15 years of development and 200 million driverless miles, argues that recent advances in AI alone cannot provide a shortcut to safely operating autonomous vehicles at scale.
What happened
Srikanth Thirumalai, Waymo’s vice president of onboard software, told Axios in an exclusive interview and expanded on the company’s findings in a blog post published Wednesday. He said their experience shows that deploying safe AVs requires more than progressively larger or more sophisticated AI models.
Why it matters
The answer could shape whether the autonomous vehicle (AV) race remains a long, costly process that favors Waymo’s substantial head start, or whether newer approaches enable rivals to reach full autonomy faster.
Waymo’s position
Thirumalai said Waymo’s objective from the beginning has been “demonstrably safe AI.” In the blog post the company summarized 10 AI lessons learned from its first 200 million autonomous miles.
Although Waymo shifted over time from many specialized models to fewer, larger foundation models — "riding the AI wave," in Thirumalai’s words — the company warns against treating AI as a silver bullet. It noted that some competitors are moving toward so-called AV 2.0 systems: end-to-end neural networks that take raw sensor inputs and output driving commands. While Waymo has tried similar tools, it concluded that pure end-to-end systems lack sufficient safety guardrails to meet its safety bar at the scale it operates.
Thirumalai also warned that even very large models "still hallucinate," and unlike software on servers, physical AI systems can’t be simply rebooted or refreshed without real-world consequences.
Sensors, maps and the debate over camera-only systems
Waymo weighed in on a major industry debate: whether cameras alone can solve full autonomy. The company’s training experience showed markedly better perception when cameras, lidar and radar worked together than when one or more sensors were removed. As Thirumalai put it, "the AI can only make sense of what it sees, and if you just can't see it, the AI can't do much."
That point is a direct critique of Tesla’s camera-only approach. Waymo also questioned claims by some rivals that high-definition mapping is unnecessary.
Bottom line
Waymo’s long track record in robotaxi development gives weight to its claim that there is no single shortcut to safe autonomy. The company frames its findings both as a caution against overreliance on end-to-end AI and as a competitive advantage rooted in years of real-world testing. However, the debate is not closed: rapid advances in AI could still reveal faster paths to full autonomy in the future.
