Alexander Sheppert says medicine has already crossed its first AI threshold: note-writing. The internal medicine resident physician and AI researcher built Matic after years of writing code before medical school, and he called AI scribing the first obvious clinical use for AI. He said, “That bet was right.”
Nearly a third of physician practices now use AI scribing, according to KFF Health News, while the AMA reports that more than 80% of doctors use AI of some kind. Sheppert argues the next targets are the first and third parts of the visit: chart review before a doctor enters the room, and coding and billing after the visit ends.
Alexander Sheppert and Matic
Sheppert’s path to Matic started before medical school, when he spent years writing code. When AI got better at reading and writing language, he built Matic, one of the first AI scribes. The tool reflects the piece of medicine that has become easiest to automate first: turning a conversation into a note.
The broader market has moved with it. About $60 billion went into medical AI startups in the last decade, according to research from Flare Capital Partners cited by Healthcare Dive, and the FDA has authorized a growing list of AI-enabled medical devices. AI scribing sits in that mix as the first mainstream workflow tool to gain broad use.
ICD-10-CM and billing
The harder step comes after the visit. Sheppert’s argument is that chart review and coding and billing are the next parts of medicine ready for AI because they involve information gathering, comparison, and classification. In that framework, AI would not replace the visit itself; it would move through the paperwork that surrounds it.
The billing side is especially dense. The U.S. diagnosis code set ICD-10-CM encompasses about 70,000 entries, and some large organizations use dedicated coding teams to audit charts and file bills. A McKinsey report cited by the American Hospital Association estimated that billing and collections cost $40 billion a year.
AMA and the next step
That leaves doctors with a practical shift, not a finished system. AI already writes notes for many practices, but the next gains depend on whether tools can read older records cleanly enough to surface contradictions and suggest the right diagnosis codes faster than a clinician can search them manually.
For physicians, that means the next phase is likely to be judged less by polished note summaries and more by whether AI can handle chart review and billing without adding work back to the workflow. For now, Sheppert’s claim is simple: medicine has moved past the first use case, and the next one has to prove it can do more than write the middle of the visit.
