AI doctors are the self-driving problem. Even if self-driving cars are provably safer, adoption takes time. Waymo has now driven more than 220 million fully autonomous miles. Across those miles, it reports: - 94% fewer crashes causing serious or fatal injuries - 82% fewer injury crashes than human drivers in the same operating areas But Waymo didn't need 220M miles before adoption. At roughly 7M fully autonomous miles, it already had statistically significant evidence showing substantially fewer injury crashes than humans. Then adoption accelerated: - 10k rides per week in 2023 - 100k per week in 2024 - 500k per week by 2026 It took about 2-3 years for usage to scale by orders of magnitude. We’ll have the same thing with health AI. We just need enough real-world data to show that people have better outcomes when AI is involved. By the end of the year, we’ll have a health AI that’s provably safer. It’ll take 2 to 3 years for that to diffuse. But I suspect in the next 3 years, AI is going to be prescribing and performing most of the roles of the doctor.