The people who will build it
Getting expert medicine to everyone will take decades of work by researchers, clinicians and health leaders in many countries, and the Wai Foundation exists so that what one of them learns, the rest can use.
One study, many countries
The case that follows is illustrative.
On a Monday morning, a registrar in Leeds, a research nurse in Kisumu and a public health official in Lima join the same call. They are about to start one evaluation of a system that helps clinicians spot sepsis earlier in adults admitted to hospital. The protocol was agreed a month ago, down to what will be measured and what will count as success. Each hospital runs the study on its own patients, under its own ethics approval, with consent and deidentification built in. The data stays in each country. The analysis travels to it.
Four months later the results are published, including the parts that did not work. In one setting the system helped. In another it made no difference, and the team can see why. That week, a ministry of health in a fourth country reads the paper and decides how to proceed with the evidence already in hand.
That study has not happened yet, though nothing in it is out of reach. The Wai Foundation is being built to run studies like it as a matter of routine, and in public.

The skills it takes
Bringing these systems into medicine takes several kinds of skill at once. It needs people who understand models well enough to know where they fail, and people who know clinical practice well enough to know what a registrar needs at four in the afternoon. It needs people who design for regulation from the first day and can make the numbers work for a hospital, a health system or a single practice. Most of all it needs people who keep the patient in mind, including the many patients they will never meet. At Wai, engineers learn what a clinical coder does all day, and clinicians learn how a model is evaluated.
I trained as a physician at Cardiff, Oxford and Harvard, and I hold a master's in public health from Harvard. Clinical training taught me to look closely at one person. Public health made me count the people I would never see. Founding the London clinic where Wai was first built and used is where I learned that both are the same job, done by a team over many ordinary days.
In How intelligence reaches a person I wrote about the goal, a longer, healthier and happier life for everyone, and Sofian Youssef's Medicine on one record describes the system underneath it. Neither gets built without people.
Who it is for
The United Nations projects that people aged 65 and over will outnumber children under 18 by the late 2070s 1. For the rest of this century, more and more of medicine will be care for long lives, often with several conditions at once, followed over decades, while care for children and mothers is still being carried to every community. The World Health Organization projects a shortfall of 11 million health workers by 2030, mostly in low and lower middle income countries 2. We need to train more people. We also need each clinician already at work to reach more patients with the full force of their skill, and the knowledge of experienced specialists to reach places that have never had one.
Clinicians want a say in how this happens. In the American Medical Association's 2026 survey, 85 percent of physicians said they wanted to be consulted or directly involved in decisions about adopting these tools, and 88 percent said validation of safety and efficacy was critical to wider use 3. A profession that insists on evidence before it adopts a tool will move quickly once the evidence is there. Our job is to produce it.

Why a network
Research can reach practice in months when the right people work to one protocol. In 2020 the RECOVERY trial in the United Kingdom randomly assigned 6,425 hospitalized patients with Covid to dexamethasone or usual care. Among patients on invasive mechanical ventilation, 28 day mortality was 29.3 percent with dexamethasone and 41.4 percent with usual care, and the first results were published online in July 2020 4. That speed came from many hospitals enrolling patients into one shared protocol, with one team analyzing the results. AlphaFold spread for related reasons. It predicted the structures of virtually all of the proteins researchers have identified, and more than two million people in 190 countries have used it 5, because deep research was paired with open access and a global community ready to build on the results.
Medicine needs that kind of network for the tools now arriving, across ministries, health systems, hospitals, payers and clinics. One hospital can test a tool on its own patients. A network across dozens of countries can run the same evaluation in a teaching hospital in London and a district hospital in Kenya, compare what happens, publish both, and carry the finding to every health system that can use it.
As models get cheaper and more capable, the technical parts of an evaluation, drafting the protocol, cleaning the data and running the analysis, get faster every year. The human parts do not. Hospitals have to agree on what counts as success. Ethics committees have to be satisfied. Patients have to trust that their information is protected, and leaders have to be willing to publish a result that does not flatter them. Those agreements are made between people who know and respect each other's work, and they take years. A network that makes them once, and keeps them in good repair, lets every later study start where the last one finished.
Over decades, a network like that could test every important change in care openly and in many countries at once. A clinician anywhere would be able to see how a system performed on patients like hers before she used it. The years between a discovery and its arrival in ordinary care could shrink toward the months that RECOVERY showed are possible.
Working across countries means different laws on data, different ethics processes and different health systems. Some of it will be slow, and some questions will not be answerable everywhere at once.
Wai is preparing research in this spirit, including evaluation protocols, work on expert disagreement as a signal for learning, a time study of documentation, deidentification tested against reidentification attacks, and benchmarks against the standard of care. The Foundation will open that work to anyone who wants to take part.

The standards we hold
We tell the truth about evidence. When we cite a study, we cite it accurately and state its limits. A randomized trial at UCLA Health found that one ambient scribe reduced time in notes by 9.5 percent against usual care while another showed no significant change, and physicians reported occasional clinically significant inaccuracies 6. Both results belong in the account, and we expect everyone who joins us to report their own the same way.
Clinicians stay accountable for clinical decisions. Wherever Wai is used, the doctor signs every note and nothing reaches a patient without a named person's approval. That follows the World Health Organization's 2021 guidance on artificial intelligence in health, which asks that people remain in control of medical decisions 7.
Safety is engineering work. In England, manufacturers of health software manage clinical risk under DCB0129, with a hazard log and a safety case 8, and Wai carries out its clinical safety work under that standard and revisits it each time the system changes. Each person who opens the record sees what their role requires, and every access is logged.
What we ask
We ask for excellence in a craft. Whether you are a machine learning researcher, a clinician, an engineer, a designer, a regulatory specialist, a health system leader or an operator, we hope you are among the best at what you do and still getting better. We look for curiosity past the edges of that craft, because the people who do this work well end up learning the rest of the system, from how a visit is coded to how a safety case is written. We ask for honesty early, about what you know and what you do not, since medicine has little patience for confident errors. We ask for care, because every decision about how a record is built is a decision about how a person is treated. And we hope for generosity with what you know, in the old tradition of clinicians who teach as well as practice.
We have not yet settled how the Foundation will be governed or which research questions it will take on first. We will decide both in the open, with the first people who join.
Each year the Foundation will report in public on whom it has connected and what came of it, and Wai will keep building for hospitals, health systems, governments, payers, clinics and individuals.
The registrar, the research nurse and the public health official in that first study are out there now. Some of them are reading this. Come and build it with us.
References
- United Nations Department of Economic and Social Affairs. UN projects world population to peak within this century, World Population Prospects 2024, July 11, 2024. https://www.un.org/sustainabledevelopment/blog/2024/07/press-release-wpp2024/
- World Health Organization. Health workforce. https://www.who.int/health-topics/health-workforce
- American Medical Association. More than 80% of physicians use AI professionally, AMA survey, March 2026. https://www.ama-assn.org/practice-management/digital-health/more-80-physicians-use-ai-professionally-ama-survey
- RECOVERY Collaborative Group. Dexamethasone in Hospitalized Patients with Covid-19. New England Journal of Medicine 384(8), 2021, 693 to 704. https://pubmed.ncbi.nlm.nih.gov/32678530/
- The Nobel Prize. Press release, The Nobel Prize in Chemistry 2024. https://www.nobelprize.org/prizes/chemistry/2024/press-release/
- Lukac, P. J., et al. Ambient AI Scribes in Clinical Practice, A Randomized Trial. NEJM AI 2(12), 2025. https://pubmed.ncbi.nlm.nih.gov/41497288/
- World Health Organization. WHO issues first global report on Artificial Intelligence in health and six guiding principles for its design and use, June 28, 2021. https://www.who.int/news/item/28-06-2021-who-issues-first-global-report-on-ai-in-health-and-six-guiding-principles-for-its-design-and-use
- NHS England. Clinical Risk Management, its Application in the Manufacture of Health IT Systems (DCB0129). https://standards.nhs.uk/published-standards/clinical-risk-management-its-application-in-the-manufacture-of-health-it-systems