hyperfitting humans
you can run a profitable education company while stalling society. that is the failure mode we refuse.
the thesis
at scale, every education system gets pressured to pick a metric that approximates human formation, then optimize that metric. the metric is never the thing. once governance is watching the proxy and the people doing the work cannot feel whether the formation is real, the system hyperfits to the proxy and the formation rots underneath it. the company can keep growing while the humans it touches get worse off.
this is the trap. naming it is the only way to refuse it.
hyperfitting, defined
in machine learning, hyperfitting is what happens when you pick a benchmark, train hard against it, and watch the model get great at the benchmark and worse at the underlying capability. perfect on the test. useless on the real task.
the same thing happens with humans, just slower and with worse stakes.
the receipts
three places this pattern shows up cleanly:
youth empowerment nonprofits. the metric is internships placed. the internships are often bad. but the funder asks for a number, and the number is what gets reported, and the number is what determines next year's grant. nobody on the board can feel whether a kid was actually formed by their summer. so the program optimizes the number, and the formation drifts.
megachurches. the metric is people saved. on a sunday, two hundred hands go up. the church reports two hundred saved. but how many of those people are actually walking with the Lord six months later? how many got discipled? the count of raised hands is not the count of disciples. the proxy is comfortable to report. the actual outcome requires being present in someone's life for years. the comfortable number wins, and the discipleship work shrinks.
y combinator and MRR. founders sign up for a subscription product once, sometimes for a single month, to make the chart go up before a raise or a batch interview. it gets reported as recurring revenue. it is not. it is a one-time spike of one-time subscriptions that churn the moment the deck is delivered. the metric got hyperfit. the underlying business did not exist. customers feel the inflation later and resent it.
each example involves well-intentioned people. that is the point. hyperfitting is not a moral failure. it is a structural one. the structure forces it.
why scale forces it
three structural reasons scaled education systems cannot avoid this trap on their own:
- governance gets reduced to numbers. boards, funders, and accreditors cannot sit in a room with one student and feel whether formation is happening. they can only read reports. so the reports become the reality the system is optimized for.
- the proxy is the only thing the system can feel. the people running a hundred-cohort program cannot adjust like a benevolent dictator who knows each kid. they have to standardize. standardization requires a metric. the metric becomes the master.
- funding cycles are shorter than formation cycles. real human formation takes years. budgets and reporting periods are quarters. the system optimizes for what it can show this quarter, which is rarely the thing that compounds over a decade.
industrial education systems are not bad because the people in them are bad. they are bad because at a certain scale, the structure stops being able to feel the thing it claims to produce. this is why homeschools and microschools keep picking up traction. they are small enough that the formation is felt directly. no proxy required.
what a well-formed human looks like
a well-formed human is one who can do the role they are called to with strength. the test depends on the role.
if the calling is to be a stay-at-home mother, that is a beautiful and serious vocation. commercial viability is not the test. presence, patience, and household formation are the test.
if the calling is to create economic value in the world, the test sharpens. the cleanest available proxy for whether someone is well-formed for that role is whether they can earn honorable money in the economy without it being charity. earning money this way is a holistic test. it requires taste, discernment, follow-through, the ability to read a room, the ability to deliver, the ability to keep going. it is the least-worst proxy we have for intelligence and formation in a commercial role. that is why we use it.
a system that produces a "certified applied AI graduate" who cannot earn honorable income in the new economy has not formed them. it has hyperfit them to a credential.
what this means for imagos
imagos does not run scaled, standardized education. we do not want to be the org whose metric becomes the master. we will not chase the certificate-per-head subsidy that pays a thousand dollars for every "applied AI graduate" we credential, because the moment that subsidy exists, the system is structurally pushed toward hyperfitting.
we are also clear that the world will figure out applied AI education without us. people are smart. they adapt. local mentors, microschools, communities of practice, and open networks will decentralize this work the way they always do. we do not have to be load-bearing on the educational floor of the imagination economy. that floor will get raised by many hands.
what the world cannot rely on others for is the pegasus model. assembling the right human unicorn, the right applied AI engineer, and the right movement builder around a real cultural leader is rare. that geometry is hard to get right. when it works, it 1000xs an industry. that is where our leverage compounds and where our attention goes.
so the work splits cleanly. raising the floor of applied AI literacy is a public good others are well-positioned to own. incubating sovereign cultural enterprises, where the formation of the founder is felt directly by us at every step, is the work only a configuration like ours can do.
the purest win-win
for anyone who loves education, mentorship, executive coaching: the purest, lowest-noise version of this work is helping a founder become well-formed enough, and strategic enough, and able to lead a small enough team well enough, to actually build a scalable business.
that engagement has all the upsides of education and none of the hyperfitting. the formation is felt directly. the proxy and the outcome are the same thing. the founder either creates real value or they do not. there is no metric to game. no certificate to inflate. no class to fake your way through. the test is whether the company they build, with the team they assembled, with the strategy they made real, actually works.
this is why imagos is the soul of the human factory. the factory part is real: output, efficiency, infinitely scalable products, agentic systems. but soul is what completes the human at the center. without it, you get a productive company that hollows the founder out. with it, you get an enterprise that elevates everyone it touches, starting with the leader.
we are not against education. we are against hyperfit education. the discipline is to stay close enough to the human to feel whether the formation is real, and to refuse any metric that would let us stop checking.
the rule
at imagos: when something we do can be measured by a clean number, ask whether the number is the thing or a proxy for the thing. if it is a proxy, name what it is approximating, and name the failure mode where the proxy gets gamed. then build the practice that keeps us in contact with the underlying reality the proxy was supposed to point at. the day we cannot feel it, we have already lost it.
related reading: the pegasus moment, the discipline, talent and agent.