There is a moment, somewhere between the intake note and the exam room door, when the machine has already finished its work and I have not yet begun mine.

The algorithm is fast. It has read the labs before I have poured my coffee, flagged the outlier, drafted the differential, suggested the order set. By most measures, it is the best medical student or resident you’ve ever come across. And yet I have learned to sit with its answer for a beat longer than seems efficient — not because I doubt the arithmetic, but because the patient behind the numbers has not yet been asked how she is sleeping, or whether her husband still remembers her name, or what she is actually afraid of.

This is the part no model can shortcut. Not because the model is unintelligent, but because it was never in the room.

Discernment is a discipline, and like any discipline, it atrophies without use. Every time I accept a suggestion without turning it over in my hands first, I am not saving time. I am practicing a smaller version of myself.

There is an old idea, older than any of our software, that wisdom is not the accumulation of correct answers but the cultivation of a self capable of recognizing which answer this person, in this moment, actually needs. The tool can tell you what is statistically true. It cannot tell you what is true for her. That gap — small, stubborn, and entirely human — is where the actual practice of medicine still lives.

I don’t say this as a technologist’s warning or a Luddite’s lament. I use the tools. I am pro technology. I grew up in Silicon Valley. I trust them more than I trusted my own uncalibrated instincts a decade ago. But I have started to notice that the danger was never that the machine would replace my judgment. The danger is quieter than that: that I would stop exercising it, one small deferral at a time, until the muscle simply isn’t there when it’s needed.

So I have built myself a habit, unglamorous and unenforceable by any dashboard: before I accept what the screen offers, I ask what it cannot see. Not as ritual, but as recovery of attention. The patient’s silence after a question. The detail she mentioned twice. The thing she didn’t say. These are not data points. They are the reason she came to a person instead of a portal.

Perhaps this is what it means to practice with discernment in an age that rewards speed above almost everything else: to keep insisting, quietly and without much fanfare, that the last decision remain yours. Not out of pride in your own judgment, but out of respect for the fact that someone trusted you — not your software — with something they could not fully put into words.

The machines will keep getting better. That much seems certain. What is less certain, and worth protecting, is whether we will still know how to pause before we let them answer for us.

Frank Tran MD is a Primary Care Physician