Expertise · Agentic AI in Medicine
Clinical decision support, autonomous prescribing workflows, AI-drafted records and direct-to-consumer health agents — reviewed by a physician who builds these systems as well as practises alongside them.
When an algorithm sits between the clinician and the patient, the usual medicolegal questions do not disappear — they move. Someone still had to decide whether the tool was appropriate for this patient, whether its output was plausible, and whether a human reviewed it before it reached the chart or the prescription. Answering that requires an expert who can read both the medicine and the system that produced the recommendation.
I sit on both sides of that line: a board-certified endocrinologist in active practice, founder of an AI health company, and Director of the AI Academy at the Geneva College of Longevity Science. I can explain to a court how these systems are built, validated and deployed — and, just as importantly, what they do not know — without either overstating their capability or dismissing them as a black box.
Case Patterns
Intake bots and agentic workflows that gather a history, order a panel and route a prescription with a physician signature applied at the end. The question a court needs answered is where independent medical judgment actually entered the encounter — and whether the clinician who signed ever saw what the agent decided.
A clinician who follows an AI recommendation into harm, or overrides a correct one, is still measured against the standard of care. I address what a reasonable endocrinologist should have done with that output: the validation the tool did or did not have, the population it was trained on, and whether the recommendation was plausible on the record in front of them.
Ambient scribes and LLM-drafted notes that record examinations never performed, carry forward resolved diagnoses, or hallucinate results outright. When the chart itself is partly machine-written, establishing what actually happened at the visit becomes an evidentiary problem before it becomes a clinical one.
Chatbots and longevity apps advising on peptides, testosterone, supplements and dosing without a clinician in the loop — and the gap between the marketing claim, the disclaimer in the terms of service, and what a reasonable user understood they were being told.
Whether a tool was clinically validated, in whom, and against what endpoint; whether it was deployed as regulated software or launched as a "wellness" product to avoid that path; and whether performance degraded in the population it was actually used on.
Algorithmic titration of insulin, GLP-1 agonists and hormone therapy, and risk scores applied to metabolic disease. These are the systems I both practise in and build with — I can speak to the model behaviour and to the endocrinology it is acting on, rather than one at the expense of the other.
What the Record Shows
For expert review, opinion or testimony in a clinical AI, agentic prescribing or digital health matter.
Enquire about a matter