On a Friday afternoon, my cardiologist sent me a message through the patient portal. It read, in full: “You do have coronary atherosclerosis resulting in moderate narrowing but no major narrowing of the arteries. I unfortunately do not have a clear sense of total plaque burden based on this study. Look forward to discussing in more detail next visit.”
My next appointment was four months away.
I am 55 years old. I had just been told, in two sentences sent at the end of a workweek, that I have heart disease. The message contained no statement of what this diagnosis meant for the care plan I was already on. No mention of what additional tests existed. No interim instructions. No acknowledgment that I might have questions before September. Just a clinical fragment, full stop. I did not know whether I should keep training for my next race or stop. I did not know whether to adjust my medication. I did not know whether this changed my immediate risk.
So I did what I had been doing for weeks. I opened my patient portal, pulled up the radiology reports already sitting there, and asked several AI systems to help me read them.
Real physician voices, twice a week
Free, and one click to unsubscribe.
I want to be careful about what I am describing. AI did not diagnose me. It did not replace my cardiologist. What it did was translate. Within an hour, I understood what my diagnosis category meant, what my imaging actually showed, and what additional tests would sharpen the picture. I learned that more aggressive lipid targets could reasonably be discussed for someone with documented coronary plaque. I learned that newer analyses could quantify my plaque burden from images I had already paid for, and that those tools exist but were not routinely offered. I learned which questions were worth asking at my appointment in September.
I stress-tested every important claim across multiple AI systems, with multiple iterations. One was confidently wrong about a specific guideline; I caught it because two others contradicted it. This is the kind of thing that should make readers nervous about AI in health care, and they should be. It is also exactly why the patients best positioned to benefit from AI today are the ones who can afford to verify what it tells them.
None of the information I gathered was hidden. My imaging reports were sitting in my portal the whole time, written by radiologists who clearly knew what they were doing. The data was excellent. The clinical care, in its narrowest sense, was excellent. What was missing was something else entirely: the synthesis, the translation, the conversation. The work of taking a patient who has just been told they have heart disease and helping them understand what that means and what to do about it. That work, in modern American health care, has been quietly squeezed out of the appointment. The medicine is twenty-first century. The communication is stuck in the 1990s.
My cardiologist is not negligent. He completed the imaging, ordered the right analyses, made the correct diagnosis, prescribed the right medication. By the metrics the system measures, he did his job. But by the metric that actually matters, did the patient walk away understanding their situation and equipped to act on it, he did not. And he didn’t, in part, because the system isn’t structured to pay him to. Appointment slots are short. Documentation requirements are crushing. Insurance reimbursement rewards procedures, not conversations. The result is a health care experience in which the technical work happens correctly and the human work happens poorly or not at all.
I was able to compensate because I have specific resources: literacy, time, the ability to pay for additional tests out of pocket, the technical capacity to use multiple AI systems and cross-validate their outputs. Now imagine, instead, a 78-year-old patient who is not computer-literate, who has the same diagnosis, who receives the same two-sentence message. That patient does not have a portal report they can decode. They do not have AI tools they can consult. They do not have the energy or the knowledge to ask about additional labs or emerging analyses. They sit with a two-sentence diagnosis for four months, drawing whatever conclusions they can, often the wrong ones, often shaping decisions about exercise and diet and end-of-life planning based on a fundamental misunderstanding of what their doctor actually meant. They get worse care, from the same system, than I do. Not because the medicine is worse. Because the communication around the medicine is worse, and they have no way to make up for it.
This is the inequity that artificial intelligence is, right now in 2026, capable of addressing. Not by replacing doctors. By providing the synthesis, translation, and patient education that doctors are no longer given time to do.
The objections to AI in health care are familiar: It can be wrong, it can hallucinate, it lacks the accountability a licensed clinician carries. These concerns are real. They are also addressable. Clinical AI systems trained specifically on medical decision-making, calibrated to medical uncertainty, deployed by institutions willing to underwrite responsibility, these systems exist in early form today and will be substantially better within five years. The technology to improve this exists. The clinical workflows, safeguards, reimbursement models, and institutional will do not. This is not science fiction. It is a workflow change.
What would deployment look like? Alongside every diagnosis, a patient-facing explanation written at the patient’s literacy level, generated automatically and reviewed by the clinician before it is sent. A system that answers patient questions between visits, escalating to the clinician when appropriate. A system that translates radiology reports, prepares patients for upcoming visits, and remembers a patient’s medical history across the fragmented years of care delivery that characterize modern American medicine. None of this requires technology that doesn’t exist.
I am lucky enough to have options. Most patients do not. The cardiologist who sent me a two-sentence diagnosis on a Friday afternoon will continue in his practice, and will continue to send terse messages to other patients. Most will not push back. Most will sit with their fragments and try to make sense of them, alone. A few will use AI to fill the gap. Until health systems provide that support directly, the patients most able to use AI will continue to receive better explanations than the patients who need explanations most.
Andrew Beck is a patient advocate.
