Gabe Wilson demands AI workflow rebuild

Healthcare AI showed its clearest gains in Portugal when it changed the route into care, not just the software on top. (nejm.ai) The evidence comes from ESSENCE, a prospective quality-improvement study inside CUF, Portugal’s largest private healthcare network, using Ada Health’s symptom-assessment system in the myCUF app. (clinicaltrials.gov) (nejm.ai) Patients entered symptoms before care, answered follow-up questions, and got guidance on whether to use self-care, primary care, specialist care, or the emergency department. Researchers then checked what people actually did through electronic health records and surveys. (clinicaltrials.gov) (nejm.ai) That setup matters because most healthcare AI products still arrive as sidecars: a chatbot, a note tool, or an assistant added after the visit has already started. ESSENCE tested AI at the front door, before the clinician encounter. (clinicaltrials.gov) (mckinsey.com) The study enrolled 1,470 adults, and 1,338 had both preassessment and postassessment care intentions recorded. After the assessment, 33.0% revised their planned care level and uncertainty fell from 12.6% to 5.0%. (nejm.ai) Among 721 participants with observed behavior, 59.1% changed their care pathway. Of those, 28.9% de-escalated, 17.2% escalated, and 13.0% resolved earlier uncertainty. (nejm.ai) Primary-care consultations rose from 16.3% to 42.1%, while specialist visits dropped from 49.7% to 29.8%. In 382 nonemergency cases with enough documentation, appropriate care rose from 29.8% before assessment to 64.4% after patients acted. (nejm.ai) Emergency-department demand also shifted. Of 96 participants who initially planned an emergency visit, 38.5% chose lower-acuity care after the assessment, and physicians later judged 27 of 29 followed-up cases to have appropriately avoided an unnecessary emergency visit. (nejm.ai) The clinical trial record shows Ada and CUF were not only studying advice quality. They also set consultation efficiency and health-seeking behavior as explicit objectives, which is another sign the intervention was built around workflow. (clinicaltrials.gov) That is the backdrop for Gabe Wilson’s argument that healthcare AI should rebuild workflows instead of being bolted onto electronic health record screens. The ESSENCE results support the narrower factual claim behind that argument: the measurable gains came when AI was embedded before the visit and connected to the next care step. (nejm.ai) (clinicaltrials.gov) For product teams, the lesson is less about adding another assistant and more about deciding where the patient starts, what data gets collected before the visit, and how that information changes the handoff. ESSENCE moved those steps upstream, and the care pattern moved with them. (clinicaltrials.gov) (nejm.ai)