UC San Diego Health’s AI strategy shows how one health system is trying to move artificial intelligence from pilot project to clinical infrastructure. For physicians, that shift raises a practical question: who governs AI before it changes the way doctors document, communicate, and deliver care?

A Becker’s Hospital Review sponsored article, published in collaboration with Nabla, reported that UC San Diego Health leaders described AI as moving from a collection of point solutions into a broader operating model. The article summarized a Becker’s webinar sponsored by Nabla. It featured Karandeep Singh, MD, chief health AI officer at UC San Diego Health, and Marlene Millen, MD, chief medical information officer.

The story matters because AI now reaches into the daily mechanics of care. It can shape how physicians create notes, how teams redesign workflows, how health systems manage patient communication, and how clinicians decide which tools they can trust.

UC San Diego Health leaders described AI governance as a health-system issue, not simply a technical issue. Becker’s reported that UCSD Health uses an AI Think Shop, a structured intake process that evaluates proposed AI initiatives as health-system decisions. The article also reported that UCSD embeds AI review into existing governance meetings, procurement workflows, and medical record oversight.

For physicians, that governance model raises practical questions. Who reviews an AI tool before it reaches the exam room? Who decides whether it fits clinical practice? Who monitors accuracy, privacy, and safety? Where do medical staff leaders fit before AI changes documentation or care processes?

Becker’s identified ambient scribes as UCSD Health’s highest-profile AI rollout. The article reported that UCSD leaders described ambient scribes as moving from a single use case toward foundational infrastructure. 

For physicians, the key issue is accountability. While an AI-generated note may reduce documentation burden, physicians must still review all information added to the medical record. Ambient documentation raises practical questions about note accuracy, error correction, billing capture, patient disclosure, liability, and downstream use of the record.

UCSD leaders also emphasized workflow redesign. Becker’s reported that Marlene Millen, MD, urged leaders to observe how frontline staff actually work before automating a process, including on understaffed days, during surge scenarios, and in edge cases.

That point may offer the most practical lesson for physicians. AI can pose risks when a health system automates an idealized workflow that does not reflect real clinical practice. Before deploying AI, leaders need to understand how care teams manage interruptions, handoffs, staffing shortages, patient volume, and exceptions.

For physicians, the test is not whether an AI tool works in a demonstration. The test is whether it works in the clinic, on a busy day, when the patient story is incomplete, and the workflow is already under strain.