Our Alliances

Key takeaways

• Every Cure, the non-profit founded by Dr David Fajgenbaum, uses an AI platform to score tens of millions of drug–disease combinations and find approved drugs that can treat diseases they were never developed for. It is backed by a three-year, $48.3 million ARPA-H contract awarded in 2024 and, since February 2026, up to $76 million more to take AI-identified candidates into trials.
• The lesson for healthcare organisations lies less in the algorithm than in the team around it: knowledge-graph engineering, clinicians embedded inside the loop, and validation infrastructure, three layers hired as one system.
• Most healthcare AI teams over-invest in the first layer and under-invest in the other two. That gap is where medical AI initiatives stall.
• With the FDA’s January 2026 revision of its clinical decision support guidance allowing more clinician-facing AI to reach practice without device review, internal clinical-AI governance has quietly become the safety net, and the scarcest hire in the market.

Dr David Fajgenbaum was a 25-year-old medical student when his own immune system tried to kill him. Castleman disease, which is rare, poorly understood and frequently fatal, put him in intensive care repeatedly. Between relapses he did something few patients are in a position to do: he went looking for his own treatment among drugs that already existed, and found a candidate in sirolimus, a decades-old transplant drug that had never been used for his condition [Newsweek, June 2025].

The organisation he co-founded, Every Cure, is the systematised version of that survival story. Instead of asking which new drug can be invented for a disease, it asks which of the roughly 3,000 already-approved drugs can treat one of the roughly 12,000 human diseases [Every Cure / ARPA-H, February 2024]. Its MATRIX platform integrates biomedical knowledge graphs, machine-learning prediction and real-world evidence to score around 75 million drug–disease combinations by the organisation’s own count [Newsweek, June 2025], surfacing the most promising repurposing opportunities for clinical follow-up. In February 2026 ARPA-H committed up to a further $76 million to move at least twenty prioritised opportunities into preclinical work and ten into clinical trials [Every Cure, 26 February 2026].

For the leaders we work with, people building AI capability inside health systems, biotech and life sciences companies, the most useful thing about Every Cure is the way the organisation is designed, more than its mission or even its results so far.

The platform is a team, not a model

Read Every Cure’s own technical description carefully and a phrase recurs: human-in-the-loop. Algorithmic scoring is only the first pass; ranked candidates flow into structured review by clinicians and researchers who assess biological plausibility, evidence quality and patient impact before anything moves toward trials. The team behind it is more than forty data scientists, engineers and MD/PhD medical experts working with patient groups and research organisations [Newsweek, June 2025].

That design implies three distinct talent layers, hired to work as one system. We set out in February how a healthcare AI leadership team should be sequenced; Every Cure is a worked example.

Layer one: the graph and the models. Someone has to build and maintain a biomedical knowledge graph spanning drugs, targets, pathways and phenotypes, and train the models that score millions of hypotheses. This is specialist work: knowledge-graph engineering, machine learning on heterogeneous biomedical data, and MLOps for pipelines that must remain auditable years later. It is also, notably, the only layer most organisations actually budget for.

Layer two: clinicians inside the loop rather than consulted after it. Every Cure’s reviewers are not a sign-off committee at the end of the process; they are a designed-in stage of the pipeline. A medical advisor who joins a quarterly steering call is a different role, and attracts a different person, from a physician-scientist whose actual job is adjudicating machine-generated hypotheses every week. The second role barely existed five years ago. It is now the defining hire of serious medical AI teams.

Layer three: validation and governance. Between an exciting model output and a patient lies everything regulators, ethicists and hospital counsel care about: evidence standards, bias assessment, monitoring, accountability when the model is wrong. Every Cure’s structure treats this as core infrastructure rather than overhead. In May we described this layer as the thing that separates teams that get AI from pilot to production from teams that stall.

Why this matters more after January

The regulatory context shifted this year. On 6 January 2026 the FDA issued a revised final guidance on clinical decision support software [FDA, January 2026]. Among the changes, the agency now exercises enforcement discretion for tools that offer a clinician a single recommendation where that is clinically appropriate, rather than requiring a list of options as the 2022 version did [Covington, January 2026]. The practical effect is that more AI which suggests a diagnosis or a treatment can reach clinical settings without pre-market review.

The direction of travel, in our reading, is that regulators are shifting responsibility toward the organisations deploying AI, expecting them to monitor, validate and correct their systems in production. That makes the governance and clinical-safety layer the de facto safety net of medical AI. In Fergal Nolan’s experience of the market over 25 years, demand for people who combine clinical credibility, AI literacy and regulatory fluency exceeds supply by a wide margin, and the profile is the same one that AI governance mandates in Europe are competing for.

What hiring leaders should take from this

Having spent twenty-five years placing data and AI leaders, the pattern I see most often is teams built in the wrong order: brilliant machine learning hires first, clinical integration bolted on late, governance added when a regulator or board asks about it.

Every Cure’s example suggests the opposite sequencing. The questions worth asking before your next AI hire:

1. Is clinical expertise a stage in the pipeline or a signature at the end? If your clinicians cannot reject the model’s output as part of the designed workflow, what you have is a demonstration rather than a working system.
2. Who owns the knowledge layer? Not the models but the representation of biomedical knowledge that the models reason over. In most organisations that question has no answer, because the role has never been defined.
3. Could you evidence, today, why your system’s last hundred recommendations were safe? If the answer lives in a slide rather than a monitoring pipeline, the FDA’s new posture should worry you.
4. Would a mission-driven candidate choose you? The people capable of layer-two and layer-three work are overwhelmingly motivated by patient impact. Organisations that can articulate a genuine clinical mission consistently out-hire richer competitors for this cohort.

A problem worthy of the people you are hiring

Banba’s founding principle borrows a phrase from Stanford lecturer Matthew Rabinowitz: solve a problem worthy of you. It is hard to think of a purer embodiment than a physician who nearly died repeatedly, found his own treatment in a forgotten drug, and then built an AI platform so the next patient does not have to be their own last resort.

The healthcare AI talent market is tight and getting tighter. The organisations winning it pay well, but what distinguishes them is that they offer problems worthy of the people they want. If you are building a team to work on one, that story may be your strongest recruiting asset. If you want help finding the people for it, our work in machine learning, AI and data leadership and retained executive search is built for exactly this, and you can talk to us at any stage.

Fergal Nolan is the founder of Banba, a specialist executive search firm for Healthcare AI and Life Sciences AI, with offices in New York, London and Berlin. He has spent more than 25 years in talent acquisition, including as a founding team member at SThree / Real Staffing and as founder of Upstream, where he has built Data Science & AI teams and AI leaders for start-ups, scale-ups and global enterprises.