This is a newsletter of Faces of Digital Health - a podcast that explores the diversity of healthcare systems and healthcare innovation worldwide. Interviews with policymakers, entrepreneurs, and clinicians provide the listeners with insights into market specifics, go-to-market strategies, barriers to success, characteristics of different healthcare systems, challenges of healthcare systems, and access to healthcare. Find out more on the website, tune in on Spotify or iTunes.
In 2026, AI use is widespread. In healthcare, a new unicorn breaking records is OpenEvidence. The startup was valued $12 billion in January 2026, and is the most widely used medical AI among verified U.S. physicians. It’s a free tool for physicians, based on peer reviewed articles from prominent medical journals such as NEJM, JAMA, NCCN, and others.
On the patient side, Dave DeBronkart is gathering patient stories on AI use. Patients, such as Steven Brown, are building multidisciplinary clinical AI agents to help them guide their patient journeys (see him explain how, in the discussion from NextMed Health 2025).
I don’t often get overly excited about new technologies or trends. Most of them pass through the familiar hype cycle, and only a few ultimately deliver lasting impact. But in 2025, through my own patient experience and complications related to IBD, I saw firsthand how profoundly AI can impact a patient story and outcomes.
Patients are already using and will continue to use AI not because they prefer it to their doctors or inherently trust it more, but because healthcare systems around the world are increasingly falling short. In that gap, AI can offer meaningful support: helping people think through problems, generate ideas, ask better questions, and sometimes even find the answers they’ve been struggling to get.
So I will say it: I believe AI agents will transform healthcare to the next level. And we need to start thinking how they will be coordinated and governed.
(If you’re familiar with the basics scroll down to operating systems for agents section)
Ali Parsa, CEO and Founder of Quadrivia sees agentic AI as a solution to a very old problem in healthcare: the structural imbalance between infinitely “elastic” demand (you’ll do anything when your child is sick) and constrained supply. For him, AI, specifically autonomous, real-time clinical agents offers a way to create elastic supply of clinical skills.
He is careful to narrow the scope for Quadrivia: not “everything doctors do,” but the 20–30% of clinical tasks that are repetitive, rules-based, and suitable for automation. The vision is not just co-pilots and scribes, but agents that can autonomously follow structured workflows, talk with patients, record and summarize, and escalate to humans when needed, reliably and within 1–1.5 seconds so they still feel human.
Ali describes AI agents as programmable junior colleagues: configurable, controllable, and anchored in a hierarchical knowledge base that prioritizes clinician- and institution-level sources before general models.
But what happens when suddenly different doctors use several different agents? How do we prevent coordination and data chaos, similar to the interoperability mess we’re still struggling to fix, and may take another decade to resolve?
What we need to think about already today, is an operating system for AI agents.
Bart de Witte has worked in healthcare for over 25 years. As a European who views access to healthcare as a fundamental right, he has spent nearly a decade advocating for open-source AI and greater transparency in AI models. He believes this is essential to keeping healthcare accessible to all, rather than allowing it to become dependent on income.
Bart recently visited Ljubljana, and I picked him up at the airport. Somewhere between traffic lights and business districts, we ended up having a conversations about where healthcare AI is heading with agents.
Bart calls AI agents the first healthcare AI technology with the potential to deliver real return on investment quickly; especially in areas like scheduling, triage, and administration.
Unlike AI scribes that are summarising notes, drafting letters, agents, with proper access controls, can complete a task from start to finish, by calling EHRs, scheduling systems, clinical databases. They can chain tasks, reflect on outcomes, and trigger workflows without a human pressing “enter” each time.
However, agency also introduces risk. Once an AI system can act autonomously, questions multiply:
Who controls what the agent can access?
How do you audit its actions?
Where does sensitive patient data live?
Who is liable when something goes wrong?
Healthcare is very good at asking these questions. It’s much less prepared to answer them at scale.
Agentic AI is promising in theory, but healthcare needs to prep it’s IT systems and data for it first. One of Bart’s core concerns is that most healthcare IT systems were never designed for agent-based workflows. Monolithic EHRs embed business logic deep inside proprietary systems. They work well for documentation and billing, but they struggle when workflows need to cross systems, vendors, or even organisations.
Adding AI “features” on top doesn’t solve that. Fo agents to pull information from multiple systems, reasons over it, take action,and learns over time, healthcare systems will need a control layer for AI agents, which is what Bart calls an operating system. Think of it as an environment that governs access, security, audit trails, compliance, and collaboration between agents.
Another counterintuitive insight Bart believes in is that bigger AI models may not be the future of clinical AI. Emerging research is showing that multiple small, specialised agents, each trained for a specific task, can outperform large general-purpose models when orchestrated together. “Think of it as swarm intelligence: a cardiology agent, a nephrology agent, an oncology agent collaborating on a case,” Bart illustrates.
This matters for regulation. General-purpose AI can’t realistically be certified as a medical product. But narrowly defined agents, doing one thing very well, can. Validation, transparency, and accountability become tractable again. It’s a vision that aligns well with healthcare’s reality: specialisation, boundaries, and trust earned task by task.
Bart is building an OS system with his startup Isaree, which he co-founded together with entrepreneur Mandana Ahmadi. To be GDPR compliant from the get go, Isaree aims to run agents directly on clinicians’ phones, tablets, or desktops, enabling local processing, instead of sending sensitive patient data to cloud APIs. This way:
data doesn’t leave the device,
surveillance-based business models break down,
compliance becomes a design property rather than a legal workaround.
At the moment, most IT systems in healthcare are closed, rigid, hard to change and consequently unfitting to individual doctor’s preferences. With agents build on top of open data specifications, healthcare may move toward marketplaces of certified agents, where clinicians choose what fits their workflow, might even build their own agents, and hospitals control access and governance.
Not every clinician will build their own agent, just as not every smartphone user builds apps. But the ability to customise, combine, and adapt tools lowers the barrier between idea and implementation dramatically. And vibe coding is making building apps increasingly easier, even without coding knowledge.
The real question is not whether agents will be used in healthcare, but under what conditions. And how soon, will solutions such as OpenEvidence be available to patients as well. Why wouldn’t patients build agents on top of their data if they’re getting increasing access to their data through national portals in Europe and vendor patient portals in the US? As written by digital health KOL Hale Tecco for Second Opinion, withholding high-quality AI tools from patients only reinforces a more paternalistic model of medicine.
No one is more passionate and motivated to solve a patient case than a patient herself. So why would they not work together with clinicians, in a true partnership? The very model healthcare has been striving toward for years.
The HIMSS26 European Healthcare Conference and Exhibition is the most important European event for digital health leaders, offering unmatched education and bringing together EU policymakers, executives, clinicians and innovators.
AI related topics you will learn about in Copenhagen, include:
How are AI agents changing the workforce
How do we define appropriate autonomy vs. full autonomy?
Operationalizing LLMs in Healthcare: Early Results and Hard Truths
How are vendors preparing to the EU AI Act?
How Are EU Countries Building AI Regulatory Sanboxes for AI testing?
and more!
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.