Thinking out loud.

A new report estimates that spending on generative AI reached roughly 110 billion US dollars this year, growing about three times faster than the internet or mobile did at a comparable stage. But it counts only what customers pay, leaving out time saved and productivity gained. In healthcare, that excluded part is where most of the value sits: the documentation time given back to clinicians, the error that never happened. When a benefit is invisible to markets, someone has to measure it in real clinical work, or it will be assumed instead of demonstrated.

Across very different health systems, the core problems look much the same: growing service demand, a strained workforce and a clear need to use data better. When the problems are shared, it makes more sense to collaborate than to reinvent the wheel. Last week I gave talks at Keio University and its hospital in Tokyo, meeting clinicians, managers and researchers, including collaborators now involved in OPTIMISE and the Agile Leadership Interventions project. There is a need for more of this cross-border, cross-disciplinary work.

Interoperability in digital health is not something a country achieves once the infrastructure is built; it is a capability that requires continuous governance. This was a central insight in the doctoral thesis of Janek Metsallik, where I had the privilege of serving as opponent at Tallinn University of Technology. Even where strong technical infrastructure is in place, the reuse of health data is often held back less by technology than by how stewardship, responsibility and accountability are organised.

On Friday, Desale Tewelde Kahsay defended his doctoral thesis in anaesthesiology and intensive care at the University of Turku, and it has been a privilege to co-supervise this work. His research followed the development of ResuGlove, a soft, textile-based wearable that gives rescuers real-time feedback on chest-compression quality during CPR, from systematic review and user interviews through to a simulation-based randomised crossover trial. I was especially glad to see a development process that looked beyond technical performance, involving users early and throughout. Congratulations, Desale.

A few weeks after MIE 2026 in Genoa, two observations remain on my mind. Much of the work presented addressed problems that are easy to solve rather than the problems that matter most in care delivery, and few clinicians were engaged in the discussions. I think these are connected: when development is driven by what is technically feasible rather than by needs in care, the results have limited relevance for practice. There is a need to redirect investment towards applications that address real needs, defined together by clinicians, patients, managers, researchers, industry and policymakers before development starts.

Same AI tool, same clinic, very different routines. Watching clinicians use ambient documentation systems, I saw them develop personal approaches the system was never designed for: verbal cues to steer the AI, toggling it on and off by encounter type. Implementation scientists call this reinvention, and it tells us flexibility is a prerequisite in system design. The bigger surprise was downstream: in several clinics, AI-supported workflows compressed the time from consultation to GP letter from weeks into hours, meaning faster communication and better continuity of care.

We often talk about 'AI scribes' as if they were one technology. They are not. In practice, clinicians work with very different approaches to clinical documentation, each with its own workflow, level of AI involvement and benefits. From my clinic observations these fall broadly into three categories, and treating them as a single thing hides what actually helps and what does not.

As one of the co-founders of IMIA SEP, it makes me very happy to see this mentorship initiative moving forward, led by JaMor Hairston and Mollie Hobensack. Mentorship is such an important opportunity for students and emerging professionals in health informatics: it is how the field grows its future leaders.

Last week, for the first time in over five years, I left my desk and went on-site to observe clinicians using ambient voice technology in a real clinical environment. It changed how I see our work. The clearest lesson: the technology is needed, but implementation is everything. The clinicians I observed work under extraordinary pressure, with constant interruptions, system crashes and multiple patients at once, and that reality has to shape the research we do.

Finland's hospital-at-home sector is at an inflection point. At the Kotisairaala 2026 event, the programme covered the full breadth of advanced care delivered at home: IV therapy, heart and lung disease, ALS, liver cirrhosis, wound care and end-of-life care. These are high-acuity patients, managed outside hospital walls, and each group generates clinical signals we are only beginning to use well.