Part III: Building What Works
Chapter 14
AI in Medicine
Shiv Rao was a cardiologist at UPMC in Pittsburgh who, like most American physicians of the 2010s, spent a large part of each clinic day typing. In March 2018 he founded Abridge with Sandeep Konam, a machine-learning engineer, and Florian Metze, a Carnegie Mellon speech researcher, out of a data alliance between UPMC, the University of Pittsburgh and Carnegie Mellon. The first product, launched in October 2020, was a free app that let patients record a visit and receive a summary. Patients would not pay, and some doctors disliked being recorded. Between 2021 and 2022 Abridge turned to the clinician who had to write the note: with consent, the conversation would be captured and a draft note placed in the EHR for the doctor to review and sign. Large language models, arriving at the end of 2022, made the draft good enough.
Then growth came faster than almost anything in health IT. The University of Kansas Health System rolled Abridge out to about 1,500 clinicians in March 2023. In August 2023 Abridge became the first ambient-documentation "Pal" in Epic's partner programme (reported), and in August 2024 Kaiser Permanente agreed a rollout across about 40 hospitals and more than 600 medical offices. Abridge raised $250 million at about $2.75 billion in February 2025 and $300 million at about $5.3 billion in June. In August 2025 Epic announced its own ambient charting, AI Charting, which became broadly available on 4 February 2026 (reported). The company that had ridden Epic's workflow into hundreds of health systems now competed with it. Chapter 4 told how IBM's Watson failed to show that patients did better. Abridge's question is different and more commercial: when AI works, who captures the value?
The preview ends here. The full chapter is in the book.
In this chapter
- 14.1Four jobs, four standards
- 14.2What the evidence shows
- 14.3Base rates decide what a hospital sees
- 14.4Who pays for AI
- 14.5How an expert reads an AI business