Innovation and caution: Humanitas is focusing on AI as support during visits, while in the United States the leap towards autonomy is being tested.
THE’artificial intelligence in medicine it is no longer a promise, but an increasingly integrated component in advanced healthcare systems. From diagnostic imaging to clinical data management, AI is redefining times, processes and priorities, with a clear goal: restore centrality to the patient while lightening the administrative burden of professionals. The experience ofIRCCS Humanitas Clinical Institute of Rozzano, which has started a collaboration with Tandem Health by introducing an advanced system of AI-based automatic transcription in daily clinical practice. This is one of the first structured implementations in Europe, initially intended for nine Operating Units. The operation is as simple as it is innovative: during the visit, the system automatically transcribes the conversation between doctor and patientgenerating a first draft of a clinical report. The result is a paradigm shift: the doctor is no longer forced to divide his attention between the keyboard and the patient, but can concentrate on listening, observation and clinical evaluation. As he points out Professor Andrea Laniahead of the Endocrinology department at Humanitas, and among the first to use the device. «we can concentrate more on the patient, look at him and listen to him without interruptions, while artificial intelligence supports us in compiling the report». A crucial passage, because it clarifies the central point of the debate: AI does not replace the doctor, but enhances his work.
Humanitas and the challenge of the Smart Hospital: less bureaucracy, more care
The introduction of these tools is part of a broader strategy that has led Humanitas to be recognized as first Italian Smart Hospital for four consecutive years in the Newsweek rankings. The stated goal is to build an ecosystem in which technology and clinical skills integrate seamlessly. The system developed with Tandem Health is designed to integrate into existing clinical flowswithout changing doctors’ operating methods and without direct impact on patients. An aspect that is anything but secondary: technological adoption in healthcare often fails when it imposes too radical changes in daily practices. Furthermore, the theme of responsibility remains central: every report generated is reviewed, verified and signed by the doctorwhich maintains full control of the clinical process. Also on the front of data protection and privacythe implementation takes place in compliance with current regulations. The innovation, however, does not stop at automatic transcription. Inside the Humanitas AI Center, one of the first centers integrated into an IRCCS, we work on advanced frontiers such as Digital Twinvirtual patient models built from clinical, genomic and imaging data. These tools promise to simulate the evolution of diseases and predict the response to treatmentspaving the way for increasingly personalized medicine.
At the same time, AI systems are already used in clinical practice, for example in mammographs to improve the early diagnosis of breast and other cancers endoscopywhere they increase the detection of lesions during colonoscopy. In all these cases, the role of AI remains that of decision supportnot replacement.
The Utah case: when AI tries to become autonomous
If the European model focuses on AI like ally of the doctorsomething more radical is being experimented in the United States. In January 2026, the State of Utah announced a pilot program that allows a autonomous agent based on artificial intelligence to renew medical prescriptions. The project, developed by the Doctronic company, represents one of the first attempts at large-scale implementation of systems agents and autonomous in medicine. The program includes 192 drugs for chronic pathologies and initially includes medical supervision, which however is destined to be reduced over time. The choice has raised strong doubts on the part of medical and pharmacist associations, according to which AI should not make care decisions. The issue is clear: while tools such as those adopted by Humanitas are limited to supporting clinical work, here we enter the territory ofdecision-making autonomy. Yet, the reasons behind the experiment are concrete. In many areas, especially rural ones, patients encounter logistical and economic difficulties in renewing prescriptionsaggravated by the shortage of doctors and cuts to health coverage. Furthermore, many chronic therapies remain stable over time, making renewal an often repetitive act. In this context, AI could represent a solution for reduce bureaucracy and waiting timesbut the risk is that of sliding towards one excessive delegation of clinical responsibility. The Utah program, rather than a definitive answer, therefore appears to be a test bed: to understand how far it is possible to push automation without compromising the safety and quality of care. The comparison between models is now open. On the one hand, an AI that supports and frees up time for the doctor-patient relationship; on the other, systems that aspire to growing autonomy. The future direction of digital healthcare will depend precisely on this balance.
