The Gym Came After the Machine
Industrial machines took the labor out of work and the gym followed. AI is taking the thinking out. Here is what a cognitive gym would have to be to work.
Nobody in 1900 paid to lift heavy objects and put them back down. Work did that to you for free, whether you wanted it or not.
Then machines took the lifting, and a strange thing followed: an entire industry built to give the effort back, on purpose, for money, at six in the morning. The gym is what a society builds after the machine removes the labor it used to run on.
AI is now removing a different kind of labor. Whether the same thing follows for the mind, and what it would even look like if it did, is an open question. We don’t know yet. But we do have one precedent, and it is worth reading closely before anyone builds the product.
One person, two versions of the same trade
A structural engineer, thirty-four, drives twenty minutes each way to a building where she pays to carry weight. Nothing in her day requires her to carry anything. She goes because nothing in her day requires it.
The other half of her day runs on the same trade, one step behind. The parts of her work that used to be effortful, the first-pass calculation, the argument for why one design beats another, the draft nobody wants to write, now go to a model. The effort is gone, exactly as the lifting is gone. Unlike the lifting, nothing in her life gives it back. There is no building she drives to for that.
The precedent, in numbers
The physical side of this trade is measured, and the measurements are less flattering to the gym than the gym would like.
A 2011 study in PLOS ONE traced fifty years of American occupational data and found that in the early 1960s almost half of private-industry jobs required at least moderate physical effort. Today it is under a fifth, and the average worker burns more than a hundred fewer calories a day at work than in 1960. The machine did its job.
The deliberate replacement came later and came slowly. Kenneth Cooper’s Aerobics was published in 1968, the running boom followed in the 1970s, and the modern gym industry grew out of that decade. Decades passed between the machines taking the labor and a mass culture that gives it back, and the culture never finished the job: a 2024 analysis in The Lancet Global Health put insufficient physical activity at 31 percent of adults worldwide in 2022, up from 23 percent in 2000. The gym is real and it works. Most people still don’t go.
That is the precedent. Now the same trade is starting on the other half of her day.
Scenario 1: the brain-training reboot
The first cognitive gym anyone builds will be the obvious one, because it is cheap and it already has a business model.
Abstract puzzles, one difficulty curve for everyone, a score that goes up. This version was tested in 2010 on more than eleven thousand people over six weeks, and the trained puzzles improved while nothing else did, not even closely related tasks. A gym that only makes you better at the gym is not a gym. It is a game with a subscription.
The physical analogy predicts this failure exactly. Nobody gets stronger at carrying groceries by playing a video game about carrying groceries. Load has to sit inside the movement you want to keep.
Scenario 2: the tutor in a box
The second version is already being built, and it works, within a boundary that matters.
A model that plays the other side of a negotiation, argues back on your design, or refuses to hand over the answer until you have tried it, is a real coach. It runs the loop a human tutor runs: you attempt, it pushes, it corrects immediately, it raises the resistance when you clear the last level. That loop is the best-documented way anyone has found to build skill.
Its limit is the gym’s limit. It is a place you go, for one thing, if you are the kind of person who goes. A language-practice partner does nothing for your judgment at work. A negotiation coach you have to open on purpose gets opened by the same minority who already own running shoes. The 31 percent is not a physical-fitness number. It is a human-behavior number, and it transfers.
Scenario 3: the load is built into the work
The version that could actually reach everyone is the one nobody has built, because it is not a separate product. It is a setting inside the tools people already work in.
Every piece of this already exists in the physical version, which is why the mapping is so clean. A watch counts steps; the tool counts how much of the hour the model did for you. A machine has a pin you move to add weight; the model has a setting where it stops answering first, asks for your attempt, and pushes harder as you improve. A spotter catches the bar the moment your form breaks; the model says, immediately, where the argument went vague. A training log tracks the weight going up; the tool tracks the point where you stop needing help. And a deload week is rest on purpose, so the days you hand everything to the model are chosen, not guilty.
None of that is a new product category. It is a resistance setting on the products people already open two hundred times a day. That is the only scenario in which the cognitive gym reaches the third of people who would never drive to one.
What this gets wrong
The analogy is doing a lot of work, and it breaks in three places that anyone building this needs to know about.
First, the mind is not one muscle. The body responds to general load: a squat helps you carry groceries. Cognitive skill mostly does not transfer that way, which is the whole finding of the 2010 study. So a cognitive gym cannot sell “general fitness.” It has to be sport-specific, one real skill at a time, inside the real task. Any product that promises a sharper mind in general is the brain-training reboot with better fonts.
Second, gyms are opt-in, and the honest record of opt-in is a third of adults not moving enough, and rising. If the cognitive version is a place you go, the precedent says most people will not go. The only escape from that is scenario 3, and scenario 3 requires the companies that make the tools to add friction to a product whose entire pitch is removing it. Nobody has an incentive to build it yet.
Third, there is no metric. A gym has weight on the bar, a VO2 max, a time on the clock. There is no agreed measurement for “how much of your own thinking you did this week,” and without one there is no progression, no log, and no way to know whether the dial is working. Cooper’s book worked partly because it came with a test you could run in twelve minutes. The cognitive version does not have its twelve-minute test yet. Until it does, everything above is a design, not a result.
What you can do before anyone builds it
The gym took decades to arrive after the machine. The people who got fit in the meantime did not wait for the building. They found something heavy and picked it up on purpose.
The resistance dial already exists in a crude form: it is a sentence at the top of a prompt asking the model to hold its answer until you have given yours. It is not a product, it is not measured, and it is the same trade the gym makes, one half of the day at a time. The precedent says a minority will do this on their own for years before the tools make it the default. It also says that minority ends up in noticeably better shape.
