Before I Sign the Note

Seven questions I run on every AI-generated note in the last hour of clinic. The end-of-day vigilance no one trained us to carry.

You find the error at the end of the day, not in the demo.

I wrote earlier about the questions every clinic should ask before adopting an AI tool , the procurement layer. This is the layer after, when the tool is already in your room, the note is on your screen, and your hand is on the keyboard about to sign. The same kind of question, asked at a different moment, of a different thing.

It is late afternoon. You are finishing the last notes, and you almost sign one, because it looks right, the way they all look right now. And then a word stops you.

Not long ago, the word was a medication that does not exist. A smudge of letters where a real drug should have been, sitting in the history as if it belonged there. I knew the patient. I knew the drug it was supposed to be. So I deleted the error, typed the correct one, and moved on.

But it stayed with me. Because the note was almost perfect, and almost perfect is the most dangerous kind.

That is the feeling I want to name. Not fear of AI. Not love of it. The specific, end-of-day vigilance the tools quietly transfer onto us, and no one trained us to carry. I previously wrote about this vigilance tax.

Over the last two years I have built a short list of questions I run before I trust what a tool hands me. They are not technical. You do not need to know how a model works to ask them. You only need to remember to.