At a recent Becker’s Healthcare webinar, Dr. Matt Sakumoto (Chief Clinical Product Officer, Nabla) moderated a discussion with Dr. Karandeep Singh (Chief Health AI Officer, UC San Diego Health) and Dr. Marlene Millen (Chief Medical Information Officer, UC San Diego Health) on what it takes to move AI beyond pilots and into the core of care delivery.
Their message was clear: AI success is about building the systems, processes, and culture to support it, not merely about deploying more tools.
Here are five key themes from the conversation.
1. AI governance must be deeply embedded.
UC San Diego Health’s approach starts with a simple principle: AI should not be treated as an exception.
In addition to governing AI like other technologies, implementation and oversight of AI is built into multiple existing structures by bringing in clinical, operational, ethical, and compliance perspectives.
“There’s not much you can do if you can’t be confident behind the AI you’re using,” said Dr. Karandeep Singh.
By creating a closed-loop system for intake, monitoring, and evaluation, AI becomes part of how the health system makes decisions, not a parallel process.
2. AI should solve real constraints.
A core question guiding their strategy: Do we actually need AI to solve this problem?
“AI is a means to an end. It’s not the end in and of itself,” Dr. Singh said.
Matt Sakumoto echoed this, “Sometimes people use automation to take a bad process and just do it more times.”
In general, AI addresses fundamental health system constraints, particularly resource scarcity and access to care. This leads to two primary applications:
- Predictive AI to prioritize limited resources to the people who need them most
- Generative and agentic AI to extend care delivery beyond clinician time
The takeaway: AI must be evaluated alongside every other operational priority and not above them.
3. Integration linked to evaluation is what drives adoption.
Rather than launching standalone pilots, UC San Diego Health embedded AI into daily workflows from the start, with evaluations baked into the workflows.
“We made it part of our work, not a separate thing,” said Dr. Marlene Millen.
This shift from isolated tools to integrated capabilities enabled more natural adoption. AI became part of how clinicians and staff deliver care, not an additional layer on top.
4. Ambient AI has become foundational infrastructure.
Ambient documentation emerged as a clear inflection point.
What started as a high-cost, emerging technology quickly became:
- A key driver of clinician wellness
- A standard expectation across health systems
- A foundation for future AI capabilities
Dr. Singh described this shift as moving from a “car” to the “road” or infrastructure that enables everything from documentation to downstream automation.
“It went from being a use case to being the way we do business,” added Dr. Millen.
Importantly, adoption was organic. Clinicians didn’t need to be convinced. They actively sought it out.
That demand also surfaced a new reality: clinicians will use AI tools whether health systems formally adopt them or not.
As Dr. Singh noted, the default risk is not zero. Clinicians will find ways to use AI independently. In response, UC San Diego Health has taken a harm reduction approach, ensuring clinicians have access to approved, secure tools rather than turning to unvetted alternatives.
The takeaway: as ambient AI becomes infrastructure, health systems must not only deploy it, but guide its safe and effective use.
5. Culture and ownership are the real differentiators.
While governance and technology matter, both speakers emphasized that culture is what makes AI scalable.
Adoption ultimately comes down to usability and impact.
“Does it help them? Is it easy to use?” Dr. Millen said. “If it doesn’t make their day better, it’s not going to stick.”
UC San Diego Health has focused on:
- Strong cross-functional relationships
- A data-driven, learning health system mindset
- Shared ownership across teams
When teams feel ownership over AI tools, adoption follows naturally.
The bottom line
Moving from AI pilots to an AI operating model requires more than technology.
It requires:
- Integrated governance
- Clear prioritization
- Workflow alignment
- And a culture of continuous learning
As Dr. Singh put it, “Faced with urgency, every health system has the capability to become a learning health system.”
UC San Diego Health’s approach shows that when those elements come together, AI can move from experimentation to a core part of how care is delivered. To learn more, watch the webinar here.
