At least 71 percent of U.S. hospitals now integrate predictive AI into their electronic health records. The debate over whether AI belongs in health care is fading, but a harder question has taken its place: how do health systems change the way people actually work so the technology delivers on its promises?
Health system leaders say AI is here to stay, but they are still figuring out how to make longstanding workflows, processes and job descriptions fit with the new tools. AI in all forms-agentic, generative, predictive-still requires human oversight and sign-off. The challenge is that AI is changing not only the tools clinicians use but the work they are expected to do.
A 2026 survey from the Scottsdale Institute and Deloitte Center for Health Solutions found a substantial gap between health systems' enthusiasm for transformation and their ability to make it operational. Every surveyed health system said care-delivery transformation was either a top enterprise priority (63 percent) or very important (37 percent), but organizations scored themselves just 2.7 out of five on their ability to turn transformation efforts into scaled, AI-enabled operations.
The practical approach: changing how information gets used
Tampa General Hospital offers one example of how that work is taking shape. The health system now runs more than 110 AI applications around the clock, but those tools have not led to layoffs. John Couris, president and CEO of Tampa General, said the organization has taken what he calls a "very practical approach" to AI and predictive analytics.
"We're not deploying AI to reduce FTEs," Couris said. "AI is allowing us to do more work with the same amount of people, and that's the journey we're on."
Sepsis care shows what that looks like in practice. Tampa General uses AI to monitor changes in patients' physiology. When the technology detects a potential problem, it directs caregivers to the medical record. Clinicians confirm whether the concern is legitimate and, if it is, bring the appropriate teams to the bedside. The health system's 48-hour mortality rate for sepsis is three percent, well below the national average of 20 percent. Couris attributes much of that success to the workforce acting effectively on the information AI provides, not the technology alone.
Couris added that he would rather see AI additions done right than done quickly, ensuring clinicians are comfortable with their new copilots. "We have to deploy technology that actually makes a difference in the lives of the people who are responsible for caring for patients, and in the lives of the patients that we care for," he said.
The workforce constraint problem
Changing how people work becomes more complicated when there are not enough people to begin with. Dr. David Carmouche, chief medical and commercial officer at Lumeris, sees that problem acutely in primary care.
"There's a little bit of a problem in that we just don't have enough human beings to deliver primary care to all Americans, and yet there's this notion that we don't want to let go of the fact that humans have been part of health care," Carmouche said. "But how do we then take a constrained workforce and also meet all of the needs of the population?"
Contextual clinical decision support could help primary care physicians manage some conditions that might otherwise require specialists. Similar tools could give nurse practitioners, physician assistants, social workers and care coordinators more support in their existing roles. But reorganizing work inside the health system does not guarantee that patients outside of it will benefit. Carmouche pointed to the unresolved problem of reaching people who fall through the cracks of health care delivery. "We haven't solved for that yet," he said.
For healthcare professionals learning to integrate these tools into daily practice, AI for Healthcare training resources address the practical side of that transition.
Rethinking clinical research workflows
The same tension between what AI can do and how work actually gets done appears in clinical research. At MD Anderson Cancer Center, Chief Clinical Research Officer Dr. Jennifer Litton is looking beyond whether AI can help write a clinical trial protocol. She wants to know whether it can properly connect the disjointed processes that come next.
A trial needs a coverage determination, a budget, a contract, and order sets built into the electronic medical record. Each represents another step between designing a study and enrolling a patient. "Could it get us the coverage determination, the budget, the contract, and put together the electronic medical record order sets, so that something that can take six months could get down to 60 days?" Litton asked.
Wearables and telemedicine could move parts of clinical research closer to patients' homes. MD Anderson is also working with national network partners on a clinical trial program that could open studies across multiple sites, including for patients who might never travel to the cancer center. Paradigm Health CEO Kent Thoelke pointed to related work his company is doing with the FDA on what he called "real-time clinical trials," combining trial matching with new approaches to data collection.
When the technology goes down
As AI changes how clinical work is organized, it also increases health systems' dependence on the technology underneath that work. The number of patient records affected by reported health care data breaches climbed from about 6 million in 2010 to 170 million in 2024, according to a recent study published in JAMA Network Open. Hacking or IT incidents accounted for 91 percent of affected records in 2024.
At Jefferson Health, Chief Information Officer Luis Taveras said that threat environment has already forced the organization to reconsider something as routine as installing a security patch. Traditionally, Jefferson would wait for third-party vendors to send patches, bundle them together and install them during scheduled monthly downtime. Taveras said that approach no longer moves quickly enough. Jefferson now uses same-day or immediate patching depending on the vulnerability.
The shift illustrates the same tension playing out across health care AI. New technology can move quickly, but organizations built around older processes do not automatically move with it. For teams managing the operational side of these changes, AI Agents & Automation coursework addresses how to redesign workflows around what the technology actually requires.
Why this matters for healthcare professionals
The examples from Tampa General, MD Anderson, and Jefferson Health point to a familiar management problem beneath the surface of health care's AI transition. The technology may be new, but many of the questions it raises are not: Who does what? In what order? Who makes the final decision? What needs to change when the old way of working no longer fits?
As Carmouche told Newsweek, "It's not really a technology story here." For clinicians, administrators, and IT leaders, the practical task is not adopting AI itself but redesigning the human processes around it-and that work is just beginning.
Your membership also unlocks:
