It’s 10:47 at night.
You’ve closed the last chart of the day, except you haven’t, because one case will not sit still. You can hear a teaching in your head. You know it was covered. You cannot remember which course, which case discussion, which hour. So you start clicking.
I have never had that night as a clinician. I have spent nine years on the other side of it: watching people hunt through teaching material they already paid to learn, knowing it is in there somewhere.
I’m not a clinician. I’ve been at SAFM since 2017, most of it as the Integrator: the person who makes the school actually run. In 2025 I moved into Technology Director so we could meet AI on purpose, not by accident.
The whole point, if we get it right, is that AI gives you back the visit.
The job came with one condition I set for myself: whatever we built would sound less certain than the tools everyone else was shipping. Not because certainty doesn’t sell. Because this school spends its whole curriculum refusing to oversimplify, and a confident tool would undo that in a week.
What we are building
When people hear “AI at a school,” they picture a chatbot that plays doctor. That is exactly what we are not building.
The first job is findability. Get the teaching back in your hands while the case is still warm. Point you to where SAFM covers a topic, in SAFM’s own language, with a source you can open and judge for yourself.
We have been building that as SearchSAFM: a search tool that only looks inside our course materials. It cites what it used. When the material does not cover a question, it says so instead of reaching. It will tell you, in plain language, that it is not a practitioner and it does not give clinical advice.
That last part is not a disclaimer we slapped on the footer. It is the product.
Most AI tools are designed to always have an answer. Ours is designed to sometimes not have one. “I don’t know” is a feature. A confident paragraph with no teaching behind it is how people get hurt.
A chatbot will write you a sure-sounding paragraph with a citation that looks like a real paper. I have watched those citations evaporate when you try to open them. Confidence is not a source.
We hold ourselves to the same rule on our side of the wall. AI helps us turn hours of live teaching into readable transcripts, and a note rides along so nobody mistakes a first pass for finished teaching material. Speed is allowed. Unreviewed certainty is not.
If you take nothing else from how we build it: we would rather a tool look a little less impressive than have it invent a source, a recommendation, or a plan.
Search is the first job, not the last one. You cannot make a better decision from a teaching you cannot find. What comes next stays in that spirit: tools that help you hold a framework and see the considerations around a choice.
The shift the practitioners gave us
In 2025 we sat down with SAFM-trained clinicians and asked how they were already using AI.
They were not waiting for us. They were already using it daily. They were catching made-up citations. They were still doing the thinking. They were also being pulled toward tools that will generate a one-size-fits-all plan on request.
I walked into that room to find out how badly they might misuse it. I walked out having been corrected. The risk was never that they would use AI carelessly. It was that if we said nothing, they would learn their habits from tools that were never built for this work. I had the fear pointed at the wrong thing.
So the strategy shifted. Not “prevent AI by hiding it.” Teach people to use it with the same skepticism they already bring to a lab printout that does not match the person in front of them.
Three commitments came out of that room, and they shape everything we build.
- We only retrieve from SAFM’s own teaching. We do not let AI roam the open web and then dress the result up as ours.
- Every answer that claims to be from the material has to show where.
- If it is not there, the tool says it is not there.
There is a fourth underneath those, and it isn’t technical. We will not build a machine that hands you a formula and calls it care. This school’s whole education is frameworks, considerations, and the judgment to choose between them. A tool that erodes that judgment isn’t a gift.
The thing that stayed with me was how unbothered they were by the limitation. You still have to think. You still have to ask the better question, then ask it again. Nobody in that room wanted that part taken away.
That is not a limitation of the technology. That is the point of the education.
What it’s for
AI should give you back the visit.
Not by seeing the person for you. By taking back the hunt for a lecture you already paid to learn. By turning hours of teaching into something you can search. By helping you see the considerations around a decision.
The part that requires you is not shrinking.
People already arrive with a printout from a chatbot. Some of it is useful. Some of it is a dream with a bibliography. Your job, the job no AI has, is to sit in the uncertainty with another human, hold a systems view, and decide what is true for this person, in this body, in this week of their life.
That is applied functional medicine. AI can fetch the teaching. It can help you hold a framework. It cannot do the applying.
The habits that already make this work are the same ones that keep AI honest.
Ask where that answer came from.
Notice when the tone is more confident than the evidence.
Refuse the one-size-fits-all plan, even when you are tired.
Keep the chair. Nothing has earned it but you.
The future we are aiming at is not a school that practices for you. It is a school that gives you time, sources, and a clean “I don’t know,” so you can stay the one in the room.
