Inside a year of rapid, system-wide adoption at the Veterans Health Administration.
The Veterans Health Administration is not built for speed. It is built for scale. It serves roughly 9 million enrolled veterans in a model that follows each one from active duty through end of life — primary and specialty care, rehab, mental health, dental, home health, housing support and a nationwide telehealth network, most of it under one roof. It is the nation’s largest health system, its largest educator of medical professionals, and a major engine of clinical research.
Systems this size turn slowly, which is why the past year has surprised even the people running it. Ambient AI — generative technology that drafts a clinical note in real time from the conversation between patient and clinician — has spread across the VA faster than its own leaders expected. After a pilot at about 10 medical centers late last year, the department is spending 2026 pushing the technology toward every facility.
That a system this large and this careful moved this fast is the part worth studying. For any other health system weighing the same rollout, the useful question isn’t whether ambient AI helps, it’s how the VA got it to stick.
The speed is more striking because the VA was being pulled in two directions at once. The VA’s Office of Inspector General counted 4,434 vacancies across Veterans Health Administration facilities in 2025, a 50% jump in a single year as demand ran the other way. The PACT Act alone brought in roughly 900,000 newly eligible veterans, one of the largest enrollment expansions in the department’s history. Every one of those encounters — layered with the co-occurring mental health conditions, polytrauma and chronic disease common in veteran care — ends in a note someone has to write.
That note is where ambient AI does its work. In a system this size, the burden it lifts is not abstract: clinician burnout, hours of administrative load, and the cost and complexity that follow when a patient’s own words never fully make it into the record.
Becker’s convened six VA leaders, from a single medical center to a national program office, for a candid conversation about what a rollout at this pace looks like where the work actually happens. Events have moved fast. Just one year ago, several of them feared the VA would be the last system standing without these technologies.
“If we jump into the wayback machine and go back nine months to a year, I was somewhat terrified that the VA was not going to rapidly adopt [ambient AI] because everybody else in the healthcare ecosystem was doing it,” said James McLain, executive director of the Milwaukee VA Medical Center. His fear was competitive: clinicians finishing training now expect these platforms. “Had we been in a position where it was going to be two more years before we did this, I think we would’ve started losing providers or losing our ability to recruit.”
The numbers since have been the opposite of slow. Chuck Ritter, partner success director for Abridge at the VA, put the systemwide curve in numbers: from a standing start in January and February, the program was on track to log its 400,000th recorded visit by the day of the panel in late June. “The growth and uptake of providers using Abridge has just been really impressive,” he said.
The trust was already there
At the Memphis VA, primary care trained and licensed every clinician within a month, and 55% were using the tool in the exam room — a fast climb for a workforce that runs from the tech-savvy to the change-resistant, said Rebekah Kaplowitz, MD, the medical center’s chief of primary care. It spread as much by word of mouth as by rollout plan. In Milwaukee, Mr. McLain credited an early adopter “who was probably the biggest reason it spread so fast because he was telling all the good news stories to the providers early on.” In VISN 21, the network covering California, Nevada and Hawaii, usage is now among the highest in the country, said Richard Kynion, MD, its clinical lead.
The bigger surprise came from the other end of the visit. Leaders had braced for clinician resistance and planned around it; what they hadn’t counted on was how readily veterans welcomed ai into the kind of care the country has pledged to get right.
“The acceptance from our veterans tends to be almost universal,” Dr. Kynion said. “It’s sort of surprising because some of these conversations are going to be very personal.”
The VA also had a head start it hadn’t planned for. Long before ambient AI, care teams had encouraged veterans to record their own visits — a practice suited to a population whose families are often deeply involved in their care, and who replay medication instructions and follow-up dates at home. Consent, in other words, was a conversation the VA had been having for years. “This just seems like the next step,” Mr. McLain said.
For Marcus Meadows, director of clinical practice management at the VA Northern California Health Care System, the value runs deeper than convenience, into the record itself. Veterans leaving active duty with service-connected disabilities have long worried that what they tell a clinician won’t make it into the chart, a gap that can bear directly on their benefits. Capturing the conversation closes it, Mr. Meadows said. Veterans can now point to what they were told and find it “right here in the notes,” and he expects that reassurance to grow as families gain My HealtheVet access and can revisit a provider’s instructions after the visit.
What the conversation unlocks
If the record is going to carry that weight, its accuracy has to hold up — the question any clinician asks first of an AI-drafted note. Mr. Ritter said quality is monitored on two tracks. Inside each note, a clinician can flag a rating or written feedback, which Abridge’s scientists and the VA’s contract team use to retune how the models generate notes. Separately, a national VA evaluation team audits a sample for quality. “We’re always consistently ranked number one in terms of overall quality,” he said.
A trustworthy record is only the beginning. If the first phase of ambient AI is about capturing the visit, the VA is already looking at what a captured visit makes possible. “It’s the conversation that is really the important part that we’re picking up on,” Mr. Ritter said. “And with technologies like Abridge, that becomes computable, which means we can layer intelligence into that workflow.”
That intelligence spreads across the whole visit rather than sitting at the moment of documentation: surfacing a veteran’s history beforehand so clinicians lose less time to pre-charting, taking coding off their plate during the encounter, and closing referral and follow-up loops afterward so no one is chasing them down long after the visit is over. “Unburdening that mental load so that you can focus on the person in front of you is just so vital,” Mr. Ritter said.
The same instinct extends past the visit itself, to what the veteran carries home. Since most patients may not always remember much of what was said aloud, the after-visit summary matters as much as the note. Abridge’s version, personalizable by each clinician, is written in plain language for exactly that reason.
Built to grow with the VA
A pilot proves a technology can work. Scaling it proves a partner can. The leaders were clearest on what separates a vendor genuinely ready for the VA from one that only looks ready on paper.
One date looms over everything: Jan. 16, 2027, when the VA goes live on its new Oracle electronic health record. Whether a documentation technology can plug cleanly into that new system was the leaders’ first test of any vendor.
Scale is the other non-negotiable, at an organization with more than 400,000 employees and close to 300,000 providers across 150-plus inpatient hospitals. Mr. Meadows, a nuclear medicine and PET/CT technologist in his 38th year, pressed on the specialties still waiting — how the tool will handle radiology’s existing HL7 interfaces and the inter-facility consults that route a veteran from Boston to Portland to Northern California.
The sharpest consensus was on voice. A note has to sound like the clinician who saw the patient. “Every provider is different,” Dr. Kynion said; the output has to read “like the note that’s created is in my voice.” Mr. Ritter named the same challenge from the other side — holding that individual voice steady across the entire VA workforce. “You can’t just take an existing product off the shelf and push it out to the VA and expect it to work without adjusting to the nuances of the way that the VA delivers care.”
The VA has done this kind of scaling before. Candace Oliva, director of CIDMO strategic engagements in the VA’s Office of Health Informatics at VA Central Office, offered its own template, drawn from its interoperability work. The Veteran Interoperability Pledge, an API that flags a veteran at check-in across Epic or Oracle systems, has recognized 1.5 million veterans in two years and fielded 95 million queries from partner health systems.
The lesson, Ms. Oliva said, was more about process than technology. “When you try to push something out without getting the perspective of the frontline employees or the ones that actually do the work every day, it leaves more room for failure,” she said.
The VA is, by most measures, ambient AI’s hardest proving ground: the broadest population, the widest geography, the deepest records and the highest stakes. That its leaders are no longer weighing whether the technology belongs, only how quickly to extend it, is the real measure of the past year. The pilot made the case in the exam room. What comes next — new specialties, the Oracle go-live in January, a single platform that follows the clinician and the veteran across every setting — is the work of carrying that case to scale.
That is a different kind of test than a pilot, and the VA has shown it intends to pass it. The rest of the industry is watching to see how.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
