Philadelphia-based Jefferson Health set out to give its clinicians back 10 million hours by 2028, a systemwide commitment to strip administrative work out of medical — and now nursing — practice.
One year in, the system has banked a little over 1 million of those hours, according to Luis Taveras, PhD, executive vice president and chief digital and information officer of the Philadelphia-based system.
“That’s kind of a ramp-up year, and in that first year, we gave back a little over a million hours,” Dr. Taveras said during a Sept. 14 fireside chat at Becker’s IT + Revenue Cycle Conference in Chicago. “We have a tracker that we use to track that. So it’s not just a number that we make up.”
Nearly all of that first-year total, he said, came from one source: the rollout of ambient AI documentation technology to physicians and advanced practice providers, which the health system announced as part of its broader AI strategy in September 2025, when Jefferson President Baligh Yehia, MD, described the goal as an effort to let clinicians “get back to why they went into medicine.”
The technology’s early impact, Dr. Taveras said, has shown up less in spreadsheets than in what clinicians tell him directly. One physician recently told him the tool changed his home life: “I go home. I can spend time with my family. I don’t have to go home and do my notes anymore.”
Physician and APP adoption of the ambient tool currently sits at roughly 15% to 20% of Jefferson’s 5,000 physicians, Dr. Taveras said, and he wants to push that to 70% or 80%. He doesn’t expect that growth to come from a formal campaign.
“It’s really using physician champions. It’s word of mouth,” he said. “They hear that somebody’s doing great, and they hear these stories that I just told you, which they tell each other, and that’s how things are happening. That’s how the demand is growing.”
That peer-to-peer model tracks with what Jefferson has described elsewhere: the system has built an AI champions network that identifies enthusiastic early adopters in each department and gives them deeper training to carry forward to peers, and has hired a senior vice president focused specifically on AI to help bridge clinical and technology teams.
Physicians, though, are the easier half of the adoption problem. Jefferson employs roughly 5,000 physicians and 15,000 nurses, and Dr. Taveras said the ambient rollout now underway in nursing is running into resistance the physician rollout didn’t face.
“In the nursing space, it’s much more challenging, because the way that nurses function — most of the time, nurses don’t really talk,” he said. “They come to the patients, they take their vitals, and there’s a lot of silence.”
Getting nurses to narrate their assessments out loud — Jefferson’s internal shorthand for the effort has landed on both “speak up, nursing” and, as Dr. Taveras put it, teaching nurses to “nurse out loud” — is only part of the challenge. Nursing documentation is built around templated flow sheets rather than free-form notes, which the ambient tool isn’t designed to fill in as easily.
“The way that nurses’ notes are done is with templates. Everything is templated for nurses,” Dr. Taveras said. “Is there a reason why nursing notes have to be templated, or is that just because that’s the way we’ve grown up in nursing? Does it have to be that way?”
Jefferson’s nursing pilot, running on three units at Abington (Pa.) Hospital, logged about 350,000 hours as of April, with roughly 30% of participating nurses still hesitant to use the tool. Jefferson plans to make the nursing rollout opt-in, similar to how it introduced the technology to physicians.
One unplanned benefit of the nursing pilot, Dr. Taveras said, has been for students. Nursing students who shadow nurses have historically had little to observe during the mostly silent process of taking vitals and assessing patients. With nurses now narrating their assessments aloud for the ambient tool, students say they can follow the clinical reasoning in real time.
Not every AI tool Jefferson evaluates makes it into the workflow. Dr. Taveras described a dermatology AI tool the system declined to deploy after its data science team — physicians who specialize in evaluating AI models before they’re approved — found the tool had been trained almost entirely on light skin tones.
“We have a very rigorous process to really get into the solution that we’re trying to get as an AI solution,” he said. “They dig into the details, and in that case, we found out that they have been trained primarily on light skin. We didn’t pursue that, obviously, because there’s a bias inherent in the way the solution and the model were trained.”
Jefferson has identified more than 100 AI solutions in various stages of deployment, according to Dr. Taveras, including tools focused on imaging, revenue cycle and back-office operations. As physician adoption climbs toward his 70% to 80% target and the nursing rollout expands beyond its current pilot, Dr. Taveras said he expects the pace of hours saved to accelerate substantially heading into year two of the 10-million-hour goal.
