Health care is on the brink of a radical transformation, propelled by the rapid evolution and adoption of artificial intelligence (AI), as well as the accompanying challenges.1
AI tools that directly influence health and health care delivery generally fall into four categories: clinical tools clinicians use to support diagnosis and treatment; direct-to-consumer tools individuals use outside the clinical setting; business operations tools health systems use to optimize processes like scheduling and revenue cycle management; and hybrid tools that span both administrative functions (documentation, billing) and clinical decision support (diagnoses, treatment plans).2
“Health care systems and providers alike have embraced AI the past couple years—we have moved quickly when we see the benefits,” said Eric Poon, MD, MPH, Professor of Medicine, Biostatistics, and Bioinformatics, and Chief Health Information Officer for Duke University School of Medicine.
Dr. Poon cited the use of ambient digital technologies, such as ambient scribes, which he said are being adopted at a pace unmatched by any other technology, both at Duke and nationwide.
Ambient scribes—health care-specific large language models (LLMs) that capture patient-clinician encounters to draft structured notes—are at the forefront of AI use in health care. In fact, 100% of the 43 US health care systems surveyed in a 2024 study reported piloting or fully deploying ambient scribes.3
The enthusiasm for ambient scribes is fueled in part by their potential to not just improve revenue management but also alleviate the administrative load on clinicians and improve the patient experience.
“We gave every provider at Duke access to ambient digital technology a year ago and have been pleasantly surprised by the enthusiastic and broad adoption, even though it is not yet universal. We now have more than 2,100 providers using it on a regular basis,” Dr. Poon said. “More importantly, we saw a 9-point decrease on a 100-point scale in burnout scores. When we assessed provider satisfaction toward documentation, we saw a 2.5-point improvement on a 7-point scale. We also saw about a 25% increased likelihood of providers closing charts on time.”
He said that early data also suggest increased productivity.
Indeed, Derek C. Angus, MD, MPH, Chair of the Department of Critical Care Medicine at the University of Pittsburgh and Senior Editor of the Journal of the American Medical Association, said other common AI-enabled tools being deployed in health care business and administration are also efficiency-related, including staffing tools, accounting software, and algorithms for inventory management. With the explosion and uptake of AI tools, adequate legal, implementation, reimbursement, and equity frameworks are needed to guide their timely and appropriate adoption.
Vishisht Mehta, MD, FCCP, Director of Interventional Pulmonology at the Lung Center of Nevada and a member of the American Thoracic Society’s AI Task Force, said reimbursement for AI-assisted medical tools is complicated.
Return of investment happens through two broad categories, which Dr. Mehta referred to as offensive deployment—the revenue generated when AI tools are reimbursed through CPT codes—and defensive deployment—cost savings, perhaps by mitigating legal risks or improving efficiency. Based on a survey of more than 1,000 pharmaceutical and life sciences clients, AI could unlock an estimated $868 billion in value across the sector by 2030—about $646 billion from defensive deployment and $222 billion from offensive deployment.4
Dr. Mehta said that although CPT codes for AI software do exist, they are rarely used in routine practice, in part because clinicians do not know these codes and AI-enabled reimbursable software is not widely implemented across health care settings. He also said that many professional medical societies are not fully engaging with AI applications, which may play a role in the lack of knowledge of AI capabilities among some clinicians.
Dr. Mehta, along with his colleague Sameer Avasarala, MD, conducted and published a survey on pulmonologists’ perspectives on AI integration in practice.5 While most pulmonologists reported encountering AI in their practice and claimed to have some familiarity with AI-related terms, less than a quarter felt comfortable teaching AI-related concepts to their trainees.5
Along with the need for education about AI’s capabilities, experts say the adoption of AI tools should follow some guiding principles. Before adopting such tools, health systems should have evidence of each tool’s positive impact and its value to clinicians. Additionally, systems should have the ability to refine or disable tools based on clinician preferences.
Boston Consulting Group’s report, “How AI Agents and Tech Will Transform Health Care in 2026,” makes the point that the real value of these tools and systems will not come from algorithms alone. Using its 10-20-70 rule—which allocates 10% to algorithms, 20% to technology and data, and 70% to people and processes—the report argues that organizations that redesign roles, retrain teams, and manage change effectively will come out ahead.6
“It is important to align AI tool selection with the problems we are trying to solve or pain points in workflows,” Dr. Poon said. “Otherwise, we just deploy technology for technology’s sake.”
This may be particularly challenging when the line between administrative and clinical tools is not clear, as is the case with some hybrid tools—AI software marketed as business or administrative tools, which, in practice, may directly impact clinical care.2
“The poster child for a hybrid AI tool is the ambient AI scribe, which is being sold to health care systems as a way to improve efficiency and improve the daily experience of note-taking,” Dr. Angus said, “However, such hybrid tools are currently not regulated by the US Food and Drug Administration (FDA). Despite this, they have made it into the fabric of health care delivery.”
While studies show that AI scribes save time and reduce clinician burnout, there is an implication that these products can improve clinical outcomes—akin to evidence-based therapies with regulatory approval.7–8 Dr. Mehta said another example of blurred lines is the use of LLMs to generate reimbursement appeal letters using clinical notes.
Conventional development and oversight of health technologies and treatments rely on clinical trial data, regulatory review, cost-effectiveness studies, and health policy assessments. Part of the challenge of establishing governance and regulatory frameworks for deploying AI in health care is the unprecedented pace of development, Dr. Angus said.
Assessing the clinical performance of AI scribes reveals a host of other challenges, including heterogeneity among ambient scribes, the wide range of patient profiles and disease states captured, and designing optimal clinical trials to evaluate these tools as clinical devices. These challenges add to existing gaps around discordance between marketing materials for AI products and FDA clearance specifications.9
“In addition to developing new sophisticated methods for rapid, just-in-time flexible evaluations of AI models at scale, there probably need to be incentives for health care systems to keep better track of the impact of integrating AI algorithms in the clinical workflow,” Dr. Angus said, alluding to the example of the Health Information Technology for Economic and Clinical Health (HITECH) Act of 2009, a federal initiative that incentivized integration of electronic health records.
Dr. Angus said the health care landscape could benefit from a HITECH 2.0 to promote data collection on AI use and enable analyses of the impact of such tools. While he said he does not want to slow the AI-mediated disruption in health care, significant groundwork from all stakeholders is needed to ensure that the disruption is positive.
“It is an exciting time,” he said. “It’s a gold rush; like being in the Wild, Wild West.”
References
1. World Economic Forum. The future of AI-enabled health: leading the way. World Economic Forum. January 2025. https://reports.weforum.org/docs/WEF_The_Future_of_AI_Enabled_Health_2025.pdf
2. Angus DC, Khera R, Lieu T, et al. AI, health, and health care today and tomorrow: the JAMA Summit report on artificial intelligence. JAMA. 2025;334(18):1650-1664. doi:10.1001/jama.2025.18490
3. Poon EG, Lemak CH, Rojas JC, Guptill J, Classen D. Adoption of artificial intelligence in healthcare: survey of health system priorities, successes, and challenges. J Am Med Inform Assoc. 2025;32(7):1093-1100. doi:10.1093/jamia/ocaf065
4. Kaspar C, Solbach T, Bruce A. AI’s US$ 868 billion healthcare revolution: strategic ways to play for pharma and life sciences in the future of health. PwC Strategy&. https://www.strategyand.pwc.com/de/en/industries/pharma-life-sciences/ai-healthcare-revolution.html
5. Mehta V, Avasarala SK. Perceptions of artificial intelligence among pulmonologists. ERJ Open Res. 2025;01304-2025. doi:10.1183/23120541.01304-2025
6. Boston Consulting Group. How AI agents and tech will transform health care in 2026. January 5, 2026. https://www.bcg.com/publications/2025/how-ai-agents-and-tech-will-transform-healthcare
7. Duggan MJ, Gervase J, Schoenbaum A, et al. Clinician experiences with ambient scribe technology to assist with documentation burden and efficiency. JAMA Netw Open. 2025;8(2):e2460637. doi:10.1001/jamanetworkopen.2024.60637
8. Olson KD, Meeker D, Troup M, et al. Use of ambient ai scribes to reduce administrative burden and professional burnout. JAMA Netw Open. 2025;8(10):e2534976. doi:10.1001/jamanetworkopen.2025.34976
9. Clark P, Kim J, Aphinyanaphongs Y. Marketing and US Food and Drug Administration clearance of artificial intelligence and machine learning enabled software in and as medical devices: a systematic review. JAMA Netw Open. 2023;6(7):e2321792. doi:10.1001/jamanetworkopen.2023.21792
