Digital health funding reached $14.2B in 2025, up 35 percent year over year, with AI capturing somewhere between 54 and 60 percent of it. The money found the category. The category has not found the people.
You feel this the moment you try to hire. There is no shortage of engineers who can ship an agent. There is a real shortage of engineers who can ship one into a clinical workflow, survive a payer integration, and explain the audit trail when someone asks how a decision was made. Those are not the same person, and the market keeps pricing them as if they were.
We call the missing skill dual fluency. It is the ability to speak two languages at once: modern AI systems on one side, and the regulated healthcare domain they deploy into on the other. Most candidates are native in one and tourists in the other. The work lives at the seam.
Why one skill is not the same as the intersection
A strong machine learning engineer can describe retrieval, evaluation, and model behavior with precision. Ask the same person how a false negative propagates through a care pathway, or what a HIPAA-bound data flow does to their training set, and the fluency thins. The reverse is just as common. A clinician-turned-operator understands the workflow in their bones and goes quiet when the conversation turns to latency, drift, or what the model actually does under load.
Neither of these people is the hire. The hire is the one who holds both worlds in their head without dropping either. They can sit with a physician and a staff engineer in the same meeting and translate in both directions. That person is the market. The AI skill alone is not, and the healthcare context alone is not. The intersection is.
The intersection is small because the path to it is unusual. You do not get there by accident. You get there by building something real inside a regulated system and being changed by the constraints. The constraints are the curriculum.
Why generalist recruiters cannot screen for it
A generalist recruiter screens for keywords and titles. That works when you can read the role from the outside. Dual fluency you cannot read from the outside. It does not show up cleanly on a résumé, because the people who have it tend to describe themselves in the language of whichever side they came from. The clinician calls themselves a clinician. The engineer calls themselves an engineer. The thing you are hiring for is invisible in both self-descriptions.
So the generalist forwards you two stacks of plausible profiles and asks you to sort them. You spend your scarce engineering hours doing the screening the recruiter could not. The funnel looks full. The signal is thin. In our experience this is the single most expensive failure mode in the category, because it costs you the one thing you have least of: the time of the people who could actually tell the difference.
Dual fluency does not show up on a résumé, because the people who have it describe themselves in the language of whichever side they came from.
What the interview signals actually are
You cannot screen for dual fluency by asking whether someone has it. Everyone says yes. You screen for it by watching how a person moves between the two domains under a little pressure. A few signals hold up:
- Translation in both directions. Ask the engineer to explain a model decision to a skeptical clinician, then ask them to explain a clinical constraint back to their own team. Fluency is bidirectional. People who only translate one way are fluent in one language and rehearsed in the other.
- Comfort with the audit trail. The fluent candidate reaches for provenance, evaluation, and explainability without being prompted, because they have answered to a regulator or a safety review and the reflex stuck.
- Respect for the workflow as it is. They do not propose to replace the clinical process; they propose to enter it. People without healthcare fluency tend to treat the workflow as a bug.
- Specific failure stories. Ask what broke when their system met a real patient population or a real payer. A genuine answer is concrete, slightly uncomfortable, and full of detail no one invents.
None of these can be faked at depth, which is the point. They reward the candidate who has actually done the work and quietly expose the one who has read about it. A screen built around them stops measuring vocabulary and starts measuring judgment.
The bottleneck is the moat
It is worth naming the upside. The scarcity that makes this hire hard is the same scarcity that makes it valuable. If dual fluency were common, your competitors would have it too. They do not, mostly, and neither do most of the candidates in your inbox. The team that solves the dual-fluency problem early is building on ground the rest of the market cannot stand on yet.
That is the quiet reason to take the screening seriously rather than fast. The bottleneck is not an inconvenience on the way to the hire. It is the hire. Find the people at the intersection, and the rest of the build gets easier than you expected.