“The status quo is unacceptable.”
That’s how chair of the UC San Francisco Department of Medicine Robert M. Wachter, MD, describes the health care system.
Costs continue to rise. Clinicians spend hours on documentation and administrative work. Medical records contain more information than any person can reasonably absorb. Health systems are asking their staff to do more than ever before.
In that environment, Wachter sees artificial intelligence as an essential part of the solution.
“AI is probably the most exciting possibility we’ve had in the last 20 or 30 years to make health care better, safer, more convenient, better for caregivers, and hopefully less expensive,” said Wachter, MD.
At the Children’s Hospital Association’s 2026 Annual Leadership Conference, Wachter will discuss AI’s implications for children’s hospitals during his keynote address.
Ahead of the Oct. 26-28 event in Phoenix, he spoke with CHA about where organizations should focus, how they should evaluate opportunities, and why caution cannot become paralysis.
Why are you optimistic about AI in health care?
The system is buckling under the weight of everything we need to do for patients, the costs and inconvenience of care, and the demands involving safety and quality.
The reason I’m optimistic is not that the technology is perfect. It’s because the current system isn’t.
We have terrific hospitals and talented people working hard and trying to do the right thing. Yet nobody looks at health care and says, “Let’s leave it exactly as it is.”
Where should hospitals focus first?
The earliest opportunities are in administrative work and documentation.
Clinicians spend enormous amounts of time dealing with paperwork, prior authorizations, insurance requirements, and records that have grown far beyond what anyone can realistically review in the time available.
AI scribes are a good example of a practical starting point. They can turn a clinician-patient conversation into a properly formatted note and significantly reduce the documentation burden. And, in doing so, they can help restore the humanity of a clinician-patient visit.
But applications like AI scribes are singles. You can’t win a baseball game without singles, but the home run will be sophisticated decision support that helps clinicians do their jobs better.
What makes decision support so promising?
When I’m caring for patients in the hospital, I may ask AI five or 10 questions in a morning. In the past, I might have hoped to run into a specialist in the hallway and stop them for a “curbside consult.” Now I have something in my pocket with specialty-level knowledge in every field.
While this is a big plus, the bigger opportunity comes when those capabilities are paired with patient-specific information.
A generic answer has limited value, particularly for a child with a complex medical history. A tool that understands the patient’s conditions, medications, and circumstances (and the family’s preferences) can provide guidance tailored to the situation.
That’s where AI becomes truly powerful.
What other opportunities do you see?
AI can identify patients at risk for sepsis, surface potential safety concerns in medical records, and help clinicians recognize patterns they might otherwise miss.
It is also showing promise in imaging, diagnostics, and procedures. In some cases, AI can extract information from an electrocardiogram, such as ejection fraction, that previously required an echocardiogram. In procedural settings, it can help clinicians identify abnormalities or provide guidance during surgery.
The challenge will be getting the signal-to-noise ratio right. If systems generate too many alerts, clinicians stop paying attention to them.
With so many possibilities, how should leaders decide what to pursue?
Every health system needs a governance process for AI.
The biggest mistake is trying to do everything at once. Too many pilots overwhelm clinicians and IT departments. They create the impression that the organization is chasing every new opportunity without a coherent strategy.
Start by identifying problems that would meaningfully improve care or improve the experience of clinicians if they were solved. Then focus on applications that offer substantial benefit at relatively low risk.
What should hospitals look for when evaluating tools?
First, determine whether the tool addresses a problem worth solving.
Then decide how success will be measured. The outcomes may be clinical. They may be financial. They may involve the experience of clinicians, patients, or families.
Hospitals also need to think beyond implementation. A tool that performs well on day one must still perform well on day 200. There needs to be a system for ongoing monitoring.
Leaders will have choices to make about whether they work through their electronic health record vendor, partner with larger technology companies, or pursue more specialized products. Those decisions require discipline and governance, not enthusiasm alone.
What role should clinicians continue to play?
We’re going to build a new kind of team in which people and technology work side by side.
Some tasks may eventually be handled safely by AI alone. Other tasks will involve AI but will require clinician oversight. Some responsibilities will remain fundamentally human.
One of the interesting challenges is that human oversight can sound safer than it sometimes is. If people come to trust the technology too much, they may stop questioning it. Clinicians can also lose skills if they become overly dependent on the system.
Those risks need to be managed carefully as organizations expand their use of AI.
How should leaders think about the risk of moving too slowly?
Organizations don’t need to be at the front of the pack. But they don’t want to be last, either.
This is a marathon, not a sprint. Hospitals should move thoughtfully and avoid taking on more than they can manage.
But moving too slowly carries risks of its own.
These tools have the potential to improve care, operations, and the experience of clinicians and patients. Standing still may feel safe, but it doesn’t solve the problems health systems are facing today.
We need to start with the recognition that the status quo isn’t working. We have to do better.
