▶Both sources report Karaz's prediction that "Dr. AI" will replace "Dr. Google" for initial patient self-research and guidance, with patients learning to discern when to seek human medical judgment. [1, 8]Apr 2026
▶Karaz consistently claims that service providers with real-world integration, like ZocDoc, Uber, and Airbnb, have significant negotiating leverage over the competing major AI agent platforms, a dynamic that didn't exist with Google's search monopoly. [14, 19]Apr 2026
▶Across both interviews, Karaz asserts that ZocDoc's competitive moat is its deep, multi-year experience in solving the complex, real-world operational challenges of healthcare scheduling, which he refers to as the "anthropology problem" or the "Coast of England problem." [15, 20]Apr 2026
▶Karaz's claim that ZocDoc's AI scheduling agent achieves a ~52% conversion rate, outperforming competitors (below 40%) and average humans (high 40s), is mentioned in both podcast appearances. [5, 24]Apr 2026
▶Karaz presents a nuanced view on AI vs. human performance, highlighting that while ZocDoc's AI (52% conversion) outperforms average humans (high 40s), it is still significantly less effective than the best-performing human agents (65%). [4, 5, 23, 24]Apr 2026
▶There is a tension in Karaz's predictions for AI in medicine: he forecasts "Dr. AI" will become the primary tool for patient self-guidance [1, 8], yet he also firmly believes medicine will be one of the last professions to be made obsolete by AGI due to the need for physical examination. [21]Apr 2026
▶Karaz advocates for a specific AI architecture, emphasizing ZocDoc's use of a "deterministic orchestration layer" that selectively employs LLMs. This contrasts with a pure, open-ended LLM approach, suggesting a debate on the right balance between control and capability for high-stakes applications. [3, 11]Apr 2026
▶While Karaz is bullish on AI for scheduling and patient research, he is skeptical about near-term AGI, citing OpenAI's Sora as evidence that the industry is "many, many years away" from achieving it. [16]Apr 2026
Not enough data for timeline
Sign up free to see the full intelligence report
Get started free