Everyone is talking about how Artificial Intelligence (AI) will transform healthcare in America. I’m not convinced that’s actually the core issue. Before AI can do anything meaningful in healthcare, there’s a much more basic problem that still hasn’t really been solved.

I was speaking to a clinician not too long ago who described what was, on the surface, a completely normal day. Nothing unusual about it. What he kept coming back to, though, was how often he still had to rebuild a patient’s history from scratch, not because the information didn’t exist, but because it wasn’t available in one place when he needed it. Scans in one system, notes in another, history somewhere else entirely.

Healthcare doesn’t have a full view of the patient.

A system built around fragments

Think about the last time you saw a specialist. You get asked the same questions again. Medications, allergies, previous conditions, and surgeries. Things that have already been captured somewhere in the system, just not available in that moment. That’s still how it works.

Healthcare has spent billions on electronic medical records, but patient data is still spread across hospitals, specialists, GPs, labs, pharmacies, imaging centres, and more. Each provider holds part of the record, but not the full picture. The result is duplicated tests, delayed diagnoses, medication errors, and a lot of time spent trying to reconstruct basic information. Patients end up acting as the carriers of their own medical history.

Why interoperability isn’t the real issue

Most of the industry calls this an interoperability problem.

I think it’s more fundamental than that. It’s a data problem.

We’ve spent years trying to connect systems that were never designed around the patient in the first place. Each organisation has its own records because that’s how the system evolved. But patients don’t sit inside one organisation. They move across them constantly, and the data doesn’t move cleanly with them.

And then we expect AI to make sense of it.

What AI actually sees

The issue is that AI only works with what it’s given. If the data is incomplete or fragmented, it doesn’t matter how good the model is. The output will still be incomplete, just presented with more confidence.

Why AI can’t fix this

I don’t think the next step is better AI. I think it’s fixing the structure of the data first. Instead of every organisation holding its own version of the truth, you need something closer to a single longitudinal view of the patient that follows them through the system.

Only then does AI become useful in the way people are expecting. Only then do clinicians spend less time searching for information. And only then does the system start to feel connected instead of fragmented.

That conversation with the clinician is probably the simplest way to describe the problem. Nothing was missing in theory. It was all there somewhere. It just wasn’t available in the moment it mattered. And I think that’s still where most of the issue sits today.

AI isn’t the problem here. The data access is.