Click here to get this post in PDF
Myra Ahmad has already made her position on artificial intelligence in medicine clear: it should support providers, not replace them. The harder question, she argues, is not whether AI belongs in chronic care; it is whether the data feeding these tools is good enough to trust in the first place. On that question, she thinks the industry is getting ahead of itself.
The gap between AI’s promise and its inputs
Chronic conditions like obesity, diabetes, and metabolic disease are managed over years, not single visits, which means the value of any AI tool applied to them depends heavily on continuity: does the system have a complete, connected picture of a patient’s labs, medications, and history, or is it working from whatever fragment happens to be in front of it at that moment? Ahmad’s critique is that most healthcare data infrastructure fails that test before AI ever enters the picture. Labs live in one system, medication history in another, and a patient’s actual day-to-day experience of their condition in neither.
“Most providers are getting labs from separate systems that never talk to each other,” she says. It’s a fragmentation problem that becomes more acute, not less, once AI tools are layered on top of incomplete records. An algorithm trained to spot patterns across a patient’s history cannot spot a pattern it was never shown, and a fragmented record is, by definition, a record with gaps the system may not recognize.
Why continuity is the real prerequisite
“Chronic disease is about trends, so before I’d trust an AI tool near a treatment decision, it needs months of connected data on the same patient, with their labs and medications in one record instead of scattered across systems,” Ahmad says. “Even then, a provider should make the call.”
This argument extends a thesis that appears to run through Mochi’s own design: that continuity of care, not any single technology, is the precondition for good outcomes in chronic disease. Applied to AI specifically, that means a model operating over fragmented data will reproduce the same blind spots that already limit human-coordinated care; it will simply do so faster and with a veneer of technical authority that can make the blind spots harder to notice. A clinician working from an incomplete chart at least knows the chart is incomplete. A confident-sounding AI output built on the same gaps can read as more authoritative than the data underneath it deserves, which is arguably a worse failure mode than the one it is meant to fix.
What good data infrastructure would need to look like
The fix, in her framing, is not a better algorithm but a better foundation: patient records that follow the person rather than the system, so that any tool built on top of them, whether AI-assisted or not, has an accurate, current picture to work from. That is a less exciting story than a breakthrough diagnostic model, and it is unlikely to generate the same kind of headlines, but Ahmad’s point is that it is the more urgent one. An AI tool layered onto disconnected data can create false confidence in a recommendation that is only as good as the incomplete record behind it. In a chronic-care setting, that false confidence is not a minor inconvenience. It can shape a decision a patient and provider act on for months before anyone realizes the underlying data was never complete.
Where this leaves the industry’s current enthusiasm
This is, in effect, a caution flag aimed at an industry currently racing to bolt AI tools onto existing systems without first asking whether those systems are feeding the tools anything reliable. She is not arguing that the industry should slow down on AI broadly; there is likely real value in AI-assisted clinical support. Her point is that the sequencing matters. Fixing continuity first, and layering AI on top of a genuinely connected record second, is a less flashy roadmap than leading with the AI feature, but it is the one she argues actually produces better outcomes rather than just better demos.
What this means for Mochi’s own roadmap
The critique also lines up with the order Mochi’s own product decisions appear to have followed. The company has spent its effort on the less visible work of connecting labs, medications, and provider notes into a single record a provider can actually see in one place. Any AI capability the platform adds later, by this logic, is meant to sit on top of that foundation rather than substitute for it, a sequencing choice that is easy to state as a principle and considerably harder to hold to once competitive pressure to ship an AI feature quickly starts to build.
A second act to an argument she has already started
This is deliberately a follow-up to Ahmad’s earlier, more general stance that AI should support providers rather than replace them. Where that argument was about AI’s role relative to a doctor’s judgment, this one is about whether AI has the raw material to be useful in the first place. It is a more technical, more specific claim, and one that puts the responsibility back on healthcare’s data infrastructure rather than on the AI tools themselves.
A claim other founders in the space will be able to test
Because this argument is about data infrastructure rather than about Mochi’s own product specifically, it is also a claim other healthcare AI companies can be measured against. Any startup pitching an AI-powered chronic care tool should, on Ahmad’s logic, be able to answer a basic question honestly: what does the underlying patient record actually look like before the AI touches it, and how complete is it? A company that cannot answer that question with any specificity is, in her framing, building the more exciting part of the system on top of a foundation no one has actually checked. That is a useful diagnostic for patients, journalists, and investors evaluating the current wave of healthcare AI products, not just a point about Mochi’s own approach.
Also read: What Doctors Get Wrong About Weight Loss Drugs, According to a Mochi Health’s Myra Ahmad
Image source: Pexels.com
