Interoperability Quietly Became Real: TEFCA, FHIR, and Agentic Healthcare

Interoperability has been the punchline of health tech for as long as I have worked in it. The standing joke was that the fax machine would outlive us all. So I want to flag a genuinely big milestone that slipped by with almost no fanfare: TEFCA, the national health-information exchange framework, just reported that nearly 500 million health records have been exchanged across the network, up from about 10 million a year earlier. That is a fifty-fold jump in twelve months. While everyone was arguing about AI, the plumbing quietly got built.

The boring milestone that changes everything

I do not get excited about exchange frameworks for their own sake, and neither should you. I get excited about what they make possible. For two decades the reason healthcare AI stayed stuck in demos was not the model, it was that the data lived in a hundred locked basements and you could not get at it. Look at the curve that just bent.

A network that goes from 10 million to 500 million exchanges in a year is not a pilot anymore, it is infrastructure. But here is the honest caveat, and it matters: most of that volume is document exchange, the digital equivalent of faxing a PDF very efficiently. Useful, but not yet the thing an AI agent can reason over. For that you need the next layer.

FHIR is the part that matters for AI

The roadmap that should be on every health-AI leader's radar is the plan to bring FHIR, the standard for structured, queryable health data, into TEFCA exchange. It rolls out in three phases, and where you are on this ladder decides what your agents can actually do.

The difference between phase one and a pile of PDFs is the difference between an agent that can answer a question and an agent that can only hand you a document to read yourself. Structured data is what turns exchange into intelligence.

What agentic healthcare actually needs from the pipes

When I think about an AI agent doing real work across health systems, booking a referral, reconciling a med list, answering a coverage question, it needs five specific things from the data layer. FHIR over TEFCA is, not by coincidence, built to provide exactly these.

The mechanic underneath is worth picturing, because it is more elegant than the old fetch-everything approach. In facilitated FHIR, your agent does not go hunting blindly. It asks the network where the patient's data lives, the network points it to the right endpoints, and the agent queries those endpoints directly for exactly the FHIR resources it needs.

That pattern is quietly profound. It means an agent can pull a live, structured, sourced answer from wherever the patient has been, with the permissions handled by the framework rather than a hand-rolled integration for every hospital. Ten years ago that was a fantasy. It is now a phased rollout with a date on it.

So here is my ask for anyone building healthcare AI: stop treating interoperability as someone else's problem, the boring compliance thing two floors down. It just became the substrate your product runs on. The teams that learn the FHIR-over-TEFCA stack now, while it is still rolling out, will be the ones whose agents can actually reach a record and act on it when the rest of the market is still writing point-to-point integrations one hospital at a time.