Daily Briefing

What Changed

Healthcare AI's value debate is moving from experimentation to ownership and operational control: physicians want a share of productivity gains, while Epic is giving health systems tools to build workflow-specific agents [1][4].

Clinical Practice | Physicians want a share of AI productivity gains. In Doximity's survey, 44% of physicians - and 50% of primary care physicians - said physicians should be the primary financial beneficiaries when AI lets them complete more clinical work in the same amount of time [1].

Operations & Workflow | Qventus attributes fewer excess stays and more surgical capacity to its perioperative AI. In a company-authored analysis, Qventus reports eliminating more than 36,000 excess patient stays, producing over USD 110 million in annualized contribution margin, and enabling 14,000 additional surgeries across customer health systems [2].

Industry & Products | VZVC is using AI to support a concentrated biotech investing model. TechCrunch reports that the firm relies heavily on AI for day-to-day operations while making roughly five concentrated biotech and healthcare investments per year [3].

Industry & Products | Epic is opening an Agent Factory for health-system-built assistants. Epic says health systems will be able to create custom agents for workflows such as patient questions, revenue cycle management, and clinician support, building on assistants including Emmie, Penny, and Art [4].

Clinical Practice | A sepsis early-warning model progressed from research to prospective multi-site implementation. On NEJM AI Grand Rounds, Suchi Saria described work that began with a 2015 Science Translational Medicine model and later moved into a prospective multi-site Nature Medicine study using EHR data to identify patients diverging from their baseline [5].

Medical Imaging | A small dental study found similar patient and expert ratings for AI-generated treatment previews. Among 14 edentulous patients, preferences and expert evaluations did not differ significantly between actual and AI-generated facial images; the sample size makes this an early feasibility result, not deployment evidence [6].