The next phase of health will be written by AI

The health model we inherited is, in its deepest architecture, an auto-shop model. You run the car until it breaks, you take it to the mechanic, he swaps the broken part, you get back on the road. Modern medicine — all of it, from the neighborhood clinic to the cutting-edge hospital in Boston — operates on this logic: you enter the system when there's already a symptom, when the deviation has become a disease, when the damage is big enough to show up in a test someone ordered too late. The doctor is summoned by the collapse. The entire system was designed to respond to events, not to track processes.

This isn't a failure of execution. It's the architecture. And the architecture has a reason to exist: throughout the whole history of medicine, there was no infrastructure to do anything else. You couldn't measure the body continuously. The best you could do was take a sporadic photograph — a blood test once a year, an appointment when the pain was already in place — and try to infer, from that snapshot, the entire film of a life. Episodic medicine is what's left when all you have is photos and you need to understand motion. All of clinical statistics, the whole notion of a "reference value," every protocol is born from this fundamental limitation: the body was an opaque system, read in rare pulses, with a structural lag between what happens and what gets discovered.

The next phase of health is born from the disappearance of that limitation. And the one who writes it isn't the doctor — it's AI, operating on a body that, for the first time, has become legible in real time.

The body has stopped being opaque

The first thing to understand is that the body was never silent. It always transmitted. Low-grade systemic inflammation begins to climb months, sometimes years, before any diagnosis. Heart rate variability plummets before an infection manifests as fever. Sleep architecture degrades before fatigue becomes a complaint. Blood glucose oscillates in ways that predict insulin resistance long before prediabetes shows up in a fasting test. The body screams all the time. What was missing wasn't a signal — it was a receiver.

That receiver arrived, and it arrived from below, almost without anyone grasping its magnitude. It wasn't a hospital that installed it. It was the consumer industry. The optical sensor on the wrist, the ring that measures peripheral temperature and cardiac variability night after night, the continuous glucose sensor that used to be restricted to diabetics and that today anyone can stick on their arm, the mattress that reads micro-respiratory movements. Each of these devices is, on its own, a toy. Put them all together and you have something no hospital in the world possesses: an uninterrupted, longitudinal, ultra-high-resolution stream from a single organism across years. Medicine always had data from many people at few moments. Now there exists the opposite — data from one person at every moment. And it's the opposite that changes everything, because health isn't a population question. It's a question of deviation from your own baseline.

Here's the point almost everyone gets wrong: the value isn't in the sensor. The sensor is a commodity, it gets cheaper and more accurate every quarter, it's the easy part. The value is in reading what the sensor produces. And what the sensor produces is a volume of data no human being can process. A single continuous glucose sensor generates 288 readings a day. Multiply that by heart rate, variability, temperature, sleep, activity, and you have tens of thousands of points per day, per person. No doctor, however brilliant, reads that. It's not a lack of competence — it's a scale mismatch between human cognition and the data stream of a monitored body. It's exactly the kind of problem for which AI isn't a luxury. It's the only possible receiver.

AI isn't the doctor — it's the reading layer

I want to be precise here, because this is where the discourse degenerates into fantasy. AI in health isn't a robot in a lab coat that replaces the cardiologist. That's the lazy image, sold by people who never built any system at all. What AI is — and this is much deeper — is a layer. A continuous reading layer that interposes itself between the body and any decision, and that does the one thing no one else can do: look at your organism every second, without tiring, without forgetting, comparing your today with your own yesterday across years.

Think in layers, because that's how all serious infrastructure organizes itself. The internet isn't a cable. It's a stack: physical, transport, application. Each layer hides the complexity of the one below and offers something legible to the one above. Health is gaining exactly this stack. At the bottom, the hardware: sensors, wearables, tests. In the middle, the layer that was missing — the AI that turns an ocean of raw readings into interpretable signal, into deviation, into trajectory. On top, the decision: what you change, what the doctor investigates, what the system anticipates. The doctor doesn't disappear. He moves up the stack. He stops being the reader of data — a function for which the human brain is structurally inadequate — and becomes the one who decides on what the reading layer has already distilled.

This is the same move AI made everywhere it became real infrastructure. Stripe didn't eliminate companies' financial work; it created a layer that absorbed the brutal complexity of moving money and offered a clean interface upward, and beneath it there are thousands of automated decisions no one sees. Cloudflare didn't replace network engineers; it became a layer that reads the whole world's traffic continuously and anticipates attacks before they arrive. Health is going down the same path. AI will be the layer that reads the body continuously and anticipates the deviation before it reaches the clinic. It isn't the top of the stack. It's the middle — and the middle is where the power lives, because it's the layer that everyone above depends on and no one below controls.

From photograph to film: what continuity changes

There's an ontological difference between measuring a body once a year and measuring a body all the time, and that difference isn't one of degree. It's one of nature. When you only have the annual photograph, the only question you can ask is: "is this value within the population's normal range?" A population's normal range is an average of millions of people different from you. Your "normal" cholesterol could be dangerously high for your specific organism, and your "abnormal" cortisol could simply be your basal functioning. The population reference is the best you can do in the dark — and it's structurally blind to the individual.

When you have the film, the question changes entirely. It's no longer "is it within range?" It's "is it deviating from your own baseline, and in what direction, and at what speed?" That's an infinitely more powerful question, and it only makes sense with continuity. The inflammation that begins to climb slowly over six weeks sets off no alarm in an isolated test — each point is "within range." But the trajectory is unmistakable to anyone watching the film. The cardiac variability that drops three days before you get sick means nothing in a snapshot, and means everything in a time series. AI isn't performing magic. It's doing the one thing continuity allows and the photograph forbids: it's seeing the derivative. The speed of change. And almost every chronic disease is, at its origin, a derivative before it's a value.

That's what showed up, at brute scale, in early 2020. Studies with monitoring rings showed that signs of viral infection — subtle changes in temperature, resting heart rate, and variability — appeared in the data two or three days before the person felt the first symptom. Two or three days. In an infectious disease, that window is the difference between containing and spreading. In a chronic disease, the equivalent window isn't days — it's years. The body spends years deviating before it breaks. Episodic medicine arrives at the end of that period, when the damage is a diagnosis. The AI layer arrives at the beginning, when it's still a derivative you can correct with sleep, food, movement, and not with surgery.

Human engineering in place of episodic medicine

There's a word that describes what this becomes, and it isn't "health" in the old sense. It's engineering. When you have a legible system, continuously monitored, with signals that predict deviations before they become failures, you're no longer practicing medicine — you're engineering a system. It's what you do with an airplane, with a reactor, with a datacenter. You don't wait for the engine to explode to open the hood. You instrument every component, read the sensors continuously, and intervene in predictive maintenance long before the failure. Airplanes crash far less than cars not because they're simpler, but because aviation abandoned the reactive model decades ago. The human body is the last critical, highly complex system that still operates like an auto shop. That's going to end.

And I want to mark the boundary clearly, because this is exactly where the discourse gets contaminated with garbage. Human engineering isn't biohacking. Biohacking is the narcissistic, anecdotal version of this — the guy who takes forty supplements, takes ice baths for Instagram, and generalizes an n=1 experiment with no control group as if it were a discovery. Biohacking is the aesthetics of optimization with no system underneath. It's the photograph again, except taken by the very subject deluded into thinking he's watching the film. Human engineering is the opposite: it's precisely having the system, having the continuity, having the rigorous reading that disproves the hunch. Most "hacks" don't survive contact with real longitudinal data — the fast that was going to fix everything shows up in the time series worsening your sleep, the trendy supplement moves no marker. The AI layer is, above all, a machine for killing personal superstition. It turns belief into measurement.

Nor is it wellness. Wellness is the industry of performative well-being, of scented candles and retreats, of health as a feeling and not as a system. Wellness sells the sensation of caring for the body without ever measuring anything. Human engineering measures everything and sells no sensation at all — it sells a corrected trajectory. The difference between the two is the difference between believing you're fine and knowing, with longitudinal data in hand, where you're headed. One is a middle-class religion. The other is infrastructure.

Health as personal infrastructure

Here's the thesis, down to the bone: health is going to become personal infrastructure. Like power. Like the internet. Like sanitation. A continuous layer, always on, that you don't notice when it works and that becomes intolerable to live without.

Think about what happened with electricity. For millennia, light and heat were events — you lit a fire, you burned a candle, and when the fuel ran out the darkness returned. Episodic. Then electricity became infrastructure: continuous, in the wall, invisible, and the whole city reorganized itself around the premise that it's always there. No one "consults" power. It's present as a background condition. The internet made the same move a generation later — from dialing up and waiting for a connection to a permanent layer that envelops everything. Health is behind by exactly one revolution. It's still a campfire: you light it when you need it, at the clinic, and then you put it out and go back to the darkness of knowing nothing about your own body until the next crisis.

The AI layer is what makes health become electricity. Continuous instead of episodic. Background instead of event. Always reading, always present, intervening at the right level of abstraction — not with panic alarms, but with adjustments that integrate into life. And like all infrastructure, the sign that it has matured will be its disappearance. The best infrastructure is invisible. You don't think about the power grid; you think about what you do with it. Mature health as personal infrastructure won't fill you with charts and quantified anxiety — that's the adolescent stage, the stage of the app that bombards you with numbers you can't read. The mature stage is the layer that absorbed the complexity and delivers, up top, only what matters: fix this, investigate that, everything's fine, carry on. The complexity doesn't vanish. It drops a layer and stays hidden, which is what infrastructure does.

This is the point that separates those who understand from those who are merely excited about a gadget. The revolution isn't the data. It's the rendering-invisible of the data. It's the moment the layer gets good enough to stop showing you your glucose point by point and start simply adjusting the recommendation, silently, the way your body has responded over the last two years. Raw data is the primitive stage. Intelligence is making the data disappear inside the decision.

Personalization: the end of the medicine of the average

All of evidence-based medicine was built on the average. The randomized clinical trial — the gold standard, and rightly so — answers a specific question: does this treatment work better than the placebo, on average, in a large population? It's a powerful question and it was the basis of nearly every medical advance of the last century. But there's a hole at the center, and the hole is you. The average hides the variance. A drug that works "on average" might not work in you, might work too much, might harm you — and the trial, designed for the population, had no way of knowing, because you aren't a population. You're an n=1, and the medicine of the average never knew what to do with the n=1 except treat it as if it were the average.

The AI layer with continuity solves this in a way no previous generation of medicine could dream of. It has your clinical trial running all the time — in you, on you, with the control group being your own past. When you change something, the time series shows how your specific organism responded. Not the population average. You. The same meal that stabilizes one person's glucose spikes another's, and this has already been demonstrated at scale: glycemic responses to identical foods vary enormously across individuals, to the point that the notion of a universal "healthy food" is, in part, fiction. There's a food that's healthy for your body, read in your response. The AI layer is what discovers this, continuously, with no a priori theory, just observing how you respond and adjusting.

This is a philosophical inversion, not just a technical one. The medicine of the average starts from the general and applies it to the particular, hoping the particular behaves like the general. Human engineering starts from the particular and never leaves it. It doesn't ask "what works for people like you." It asks "what works for you," and it has data to answer. The concept of a "reference value" — that pillar of episodic medicine, that gray range where everyone and no one fits — starts to look like what it actually is: a crutch from the era when you couldn't measure the individual. When you can measure the individual continuously, the reference becomes your own baseline, and the population range becomes a historical detail, the way the candle became a decoration after the light bulb.

What could go very wrong

I build infrastructure for a living, and anyone who builds infrastructure learns early that the same layer that grants power concentrates power. I'm not going to paint this as a utopia, because utopia is how people who never operated a system talk. The body's reading layer is, simultaneously, the most powerful and the most dangerous thing that's going to emerge in health, and the two sides are inseparable.

First, the data. A continuous stream of biomarkers is the most intimate map that can exist of a person — more revealing than browsing history, than location, than conversation. It exposes pregnancy before the person tells anyone, depression before the diagnosis, decline before awareness, disease before the symptom. Whoever controls that layer controls the substrate of decisions about insurance, employment, credit, relationships. The question of who holds the health stack isn't technical. It's a question of power, and it's the central question of the next decade in this industry. If the AI layer of health consolidates into three advertising platforms that make money off your attention, it will optimize for engagement and anxiety, not for your trajectory. The incentive architecture of the layer determines whether it serves you or exploits you, and that's a design choice, not a destiny. That's why the local-first thesis matters so much: the reading of your body has to run where you control it, not on the server of whoever profits from you.

Second, the tyranny of the number. The immature version of this technology — the one that already exists — produces quantified anxiety. People who slept well became insomniacs because the app gave their sleep a low score. Healthy people became patients of an invented disease because a wearable scared them with a false positive. A poorly calibrated reading layer doesn't cure — it medicalizes all of life, turns every normal fluctuation into an alarm, and produces exactly the opposite of health: obsessive surveillance over a body that was fine. The maturity of the layer is measured by restraint. By when it chooses not to speak. By how much signal it absorbs and silences in order to deliver only what changes a decision. A good layer is a quiet layer.

And third, the risk of the AI learning the average again, just disguised as personalization. If the models are trained mostly on data from one slice of the population — those who can afford expensive wearables, a narrow demographic — "personalization" will be, in reality, that group's average applied to everyone, with an individual veneer. Bias doesn't disappear when you add AI. It hides better. Human engineering only delivers on the promise if the reading is genuinely of your body, and not of the dataset's average body projected onto you. This is a data and design problem, and it has a solution — but only if it's treated as the central risk that it is, and not as a detail.

One revolution behind

Medicine is, today, the trillion-dollar industry that most resembles what it was fifty years ago in its fundamental logic: reactive, episodic, of the average, opaque between contacts. Every other major area of life has already gone through the transition from event to infrastructure. Communication became a continuous layer. Money became a continuous layer. Computation became a continuous layer. Health is the last great frontier still stuck in the campfire model, and it's stuck not for lack of technology — the technology already exists, it's on the wrists of millions of people — but because of institutional inertia, because of wrong incentives, and because of a reading layer that hasn't yet matured enough to be invisible.

This transition is going to happen, and it's going to happen through the same door every infrastructure comes in: from below, through consumption, without asking the old system's permission. It won't be the hospital that installs the next phase of health. The hospital is the top of the old stack, and the top never leads the swap of the foundation. It will be the layer that forms silently on the wrist, on the ring, on the sensor, on the mattress, and in the AI that learns to read all of it as a single, continuous system, yours. When that layer matures, episodic medicine won't be abolished — it will be demoted. It will be left with what it was always good at: trauma, the emergency, surgery, acute failure. The auto shop keeps existing for when the car crashes. But predictive maintenance, continuous tracking, the anticipation of the deviation before it becomes a diagnosis — that part, which is the larger part of a whole life's health, migrates to the layer.

The body was always a system with invisible architecture, reorganizing itself in layers no one could read. The difference in our decade is that, for the first time, there exists a receiver equal to the signal. The next phase of health won't be written by an enlightened doctor nor by a clever app. It will be written by the layer — the AI that reads the body continuously, anticipates the deviation, personalizes the correction and, when it's mature, disappears into life the way electricity disappeared into the wall. Health becomes a background condition. And the generation that grows up with this will look at our episodic medicine the same way we look at bloodletting and candles: as the best that could be done in the dark, before someone turned on the light.