What My AI Scribe Taught Me About Being a Doctor

I started using an AI scribe to generate clinic notes about two years ago. I take it for granted now, but I remember being genuinely startled the first time I used it.

Why? It certainly transcribed the conversation, but we’ve had transcription technology for years.

What surprised me was that it somehow made sense of the conversation without requiring me to impose structure on it.

That day, the patient and I spent forty-five minutes discussing knee pain, insomnia, gardening, a recent family wedding, medication side effects, grief after losing a friend, blood pressure readings, and a new puppy. Yet somehow the AI sorted all of that into a coherent history, organized it into a SOAP note, and separated what belonged in the medical record from what did not.

My first thought was, “Well, that’s eerie. Maybe AI is closer to replacing me than I thought.”

My second thought was, “Look at everything we talked about. No wonder I’m tired at the end of a clinic session.”

For the first time in my career, I was seeing a visible record of work that had always happened invisibly inside my head—captured in far greater detail than I could ever recreate from memory.

As physicians, we tend to think of ourselves as listeners. But listening is only part of what we’re doing.

While a patient talks, we question and refine, sort information, track timelines, recognize patterns, weigh significance, filter distractions, form hypotheses, and decide whether to chase the hints that body language betrays but an AI scribe would never see. We’re building a map while simultaneously walking the terrain beside our patients.

Most of us do this so automatically that we barely notice it. That day, the AI note held up a mirror. Suddenly I could see the residue of decades of cognitive work that no quality metric will ever measure.

The surprising thing was not that the AI could produce a decent note. It was realizing how much work had been happening all along.

The AI also taught me something else.

It was remarkably good at identifying information relevant to the medical problem. It left out several minutes scattered throughout the visit where we talked about our dogs, recent vacations, grief, children leaving for college, and the absurdities of everyday life.

From a documentation standpoint, it was completely correct.

From a human standpoint, however, it was completely wrong.

Those conversations were not irrelevant. They were doing some of the most important work of the visit.

The AI was documenting the medical encounter. My patient and I were building a relationship.

And relationships do serious work.

Familiarity does work. Being known does work. And trust works its metaphoric butt off.

The note didn’t need those stories, but the relationship certainly did.

That experience ultimately made me somewhat less worried about AI and much more appreciative of the craft of medicine.

The technology was impressive, but it also revealed something important.

Medicine is not just information processing. It is interpretation and judgment and relationship. It is knowing which details belong in the note and which live in the space between two people who have cared about one another long enough to ask about the dog.

The AI showed me something I had forgotten after thirty years in practice.

Being a doctor is exhausting not because we spend all day talking and listening. It’s exhausting because we spend all day making sense of human stories.

It took a machine to remind me how much work that really is.