My LinkedIn feed has been flooded with discussion about a recent Nature Medicine study showing that general-purpose large language models outperform specialized clinical AI tools on medical benchmarks. Clinical informatics leaders are calling it a wake-up call for health care, pointing to a timeless truth: Technology rarely fails at the technical layer. It fails because we don’t map how the work actually happens on the ground.

As a clinician working in a resource-limited setting, on an unsalaried internship, I can tell you exactly how the work happens on the ground. And I can tell you exactly where the technology is failing us.

It isn’t failing at the technical layer. It’s failing at the equity layer.

Last October, I was rotating through internal medicine. On call, our team would see around 200 patients. Most arrived in their worst state. By the time someone reaches a hospital in South Sudan, the disease has usually had the last word for hours, sometimes days.

We have very few consultants. We are trained alongside clinical officers, brilliant, hardworking people whose training runs only three years. They staff both rural and urban hospitals, and routinely manage cases that would challenge a specialist. When something goes seriously wrong during a call, they consult us. Intern doctors. Because South Sudan has no internal medicine residency program, there is no senior trainee above us in the room. I am not a specialist. I am simply the most trained person available, and that was enough to make me “senior” in someone else’s eyes.

It is hard to carry that weight. I trained at the Latin American School of Medicine in Havana, where I saw what structured residency training looks like. Here, that structure doesn’t exist yet, not because the talent isn’t there, but because the system isn’t built yet.

In one of those call nights, surrounded by patients who simply wanted someone to save them, I thought: Why not ask ChatGPT for humanitarian access, just for a year, to help carry this load? I wrote to OpenAI, explained where I worked and why, and attached my Ministry of Health deployment letter.

The response was polite. It was also entirely automated. It praised my “dedication to medical care,” pointed me back to the free tier, and closed with a footnote: “This response was generated with AI support which can make mistakes.”

They used the technology to tell a frontline doctor why he couldn’t have the technology.

I am the first doctor from my village. Not a specialist, a doctor. My village has roughly 40,000 people, and in my entire life, I have never personally met a specialist from any field who came from there. I am currently working toward an internal medicine residency, with the long-term goal of specializing in oncology, so that one day my community has its first.

The road there is expensive, and I am unsalaried. If I were paid for this internship, I would be saving toward that residency right now. Instead, I am asking my own brother, who did very well in school and wants to become a doctor himself, to wait. There is no one to pay for him either. I told him maybe after I finish my own training, I can help. That is not how a health system should have to work.

But this story isn’t only about rejection. A few weeks before that OpenAI email, I had been searching YouTube late one night for any way to access UpToDate, a clinical reference tool far beyond an intern’s budget. I found a mention of the Better Evidence program and wrote to them. Within two weeks, I had a full year of free access.

I now share that subscription with colleagues who don’t own smartphones, especially during dangerous cases where we need a fast, trustworthy reference at the bedside. We don’t use it to replace our clinical judgment. We use it cautiously, the way you’d use a second opinion from someone who happens to know everything ever published.

That experience taught me something I didn’t expect: The world makes a lot more sense when you let someone else shine. Better Evidence didn’t have to say yes to an intern doctor in South Sudan. They did anyway.

When data shows that frontier AI models now demonstrate sharper clinical reasoning than some multi-thousand-dollar institutional software, access stops being a subscription question. It becomes a global health equity question. The phone in my pocket has more computational power than entire hospitals had a generation ago. Yet the builders of these tools are operating on the same blueprint as traditional medical publishers: no institutional budget, no access to the best tools.

If AI developers don’t map the reality of resource-strapped clinics into their pricing and accessibility models, they aren’t democratizing medical knowledge. They are digitizing disparity.

My education has been one long struggle, and it has taught me something strange: Struggle is bearable precisely because you can’t predict the outcome. I never imagined I would one day be sharing this story with an audience like this one. That unpredictability is the only reason it was worth enduring.

If the future of medicine is AI-driven, we have to ask a harder question: Who gets left in the dark?

Buga Charles George Kenyi is a physician in South Sudan.