At my heaviest I was 135 kilos, and I was paying two professionals to fix it. A nutritionist built my meal plans. A fitness coach built my training. Both were competent. Both were expensive. And in all the months I worked with them, they never once spoke to each other.
I was the only connection between them. Every week I carried messages from one to the other, badly, like a child relaying an argument between adults. The nutritionist cut my calories the same week the coach raised my training volume. Neither knew. The person deciding how those two plans combined, the single most consequential health decision being made, was the least qualified person in the room: me.
This is the first post in a series about what AI can actually fix in health. Each post stakes one falsifiable claim and shows the mechanism. Here is claim one.
The hypothesis
The breakthrough in AI health coaching is not smarter advice. It is specialists that talk to each other.
Everyone building in this space is racing to make the advice smarter. Better models, better prompts, more medical literature in the training data. I think that race misses where the value is. Advice quality was never the binding problem. Coordination was.
The nutritionist's advice was good. The coach's advice was good. The combination was incoherent, and the combination is the only thing your body experiences.
Why humans do not fix this
You could hire a team that coordinates. It exists: it is called an integrated performance staff, and professional athletes have one. A nutritionist, a strength coach, a sleep specialist and a physician who share notes weekly costs more per month than most people spend on health in a year. Coordination is a luxury good. That is the actual gap AI can close, because software agents do not bill per meeting.
What is an AI health coach, then?
The term is newer than the search traffic suggests, so here is a working definition: an AI health coach is software that does not just record what you did, but decides with you what to do next, across the areas of health that affect each other.
That last clause is the load-bearing one. Food, training, sleep and stress are one system wearing four names. Any coach that sees only one of them will keep handing you advice the other three quietly contradict.
The mechanism
In Trackr, the coaching is a council of five: nutrition, fitness, sleep, mindfulness, and a head coach that reads the whole picture. The interesting part is not any one of them. It is what happens when they disagree.
A real example from our own testing. Three short nights of sleep in a row. The fitness coach has a heavy leg session scheduled. In a single-assistant app, nothing happens: the workout app does not know about the sleep, and the sleep app has no opinion about squats.
In the council, the sleep coach objects. The fitness coach yields. Leg day moves to Thursday, and the user is told why in one sentence. Nobody carried a message between professionals. The professionals talked.
That is the whole product thesis in one interaction. Not smarter advice. Fewer contradictions.
What would prove this wrong
A falsifiable claim needs a failure condition, so here is mine. If users of coordinated multi-coach systems keep their plans running no longer than users of single-assistant apps, this hypothesis is wrong. Coordination should show up as retention: plans that survive contact with real weeks, because the plan bends before it breaks. If that never shows up in our data, the council is theatre, and I will write that post too.
What is next
Next week: why logging fails. It is not a motivation problem. We asked humans to be data clerks, and humans are terrible clerks.