Why “developer intelligence” is mostly training — and where raw aptitude still shows up

Call someone a natural and you have told a story that makes the hours disappear. I have watched that story get applied to teenagers who had a computer in the house and a parent who did not panic at a broken install. I have also watched it get withheld from people who were slower at the first midterm and then, three years later, were the ones you wanted on the incident. The word we use for the gap is intelligence. Most of the gap is training you did not watch.

I already tried to be careful about what actually predicts a strong developer — the slice-versus-career problem, the limits of IQ talk, the way “genes” gets used as a door. This is the other half: why I still treat developer intelligence as mostly a training outcome, and the few places I think raw aptitude still peeks through without becoming a myth.

What looks like a mind is often a library

The staff engineer who “just sees” the race is not performing a miracle. They have a library of races. The first time, they did not see it. They shipped it. Someone yelled. The second time they hesitated. The twelfth time it looks like insight. Insight is what we call a match against a private corpus.

That is not a slogan against thinking. It is how expertise looks in every field that has been studied without romance: chess, music, medicine, firefighting. Pattern inventory plus retrieval under time pressure. Software is not exempt because we like to imagine we are doing math on a whiteboard. Most paid work is recognizing a shape and knowing which tool has bitten you before.

When I say intelligence is mostly training, I mean this library. You cannot download it in a week. You also cannot be born with last year’s outage. The people we call gifted usually got more contact with the shapes, earlier, with less shame. That is training plus luck. It is not a different species of mind.

Training is not hours. It is being proven wrong

A year of tutorials can leave you fluent and thin. A year of owning a service that fails in production will make you look smarter than you feel. The difference is not grit as a moral quality. It is the structure of the feedback. Wrongness that is cheap and frequent builds the library. Wrongness that is rare and catastrophic builds fear. A lot of “talent” talk is just who got the cheap wrongness early — a sandbox, a patient mentor, a CS program that let you fail a compile in public.

In 2026 the cheap wrongness is in flux. Agents finish the line. Ticket queues shrink. The old junior curriculum is thinner. If companies only hire the already-trained, they will keep discovering “naturals” who are people who trained somewhere else. That is not a finding about minds. It is a finding about who paid for the practice.

I am not anti-tool. I am anti-confusing a finished line with a learned one. If you never hold the state yourself, you do not add the card to the library. You add a feeling that you could. The feeling is not the same retrieval on Saturday night.

Where aptitude still shows up — without a mythology

I will not pretend every starting line is identical. Some people hold more moving parts before the representation collapses. That shows up in early courses, in competitive programming, in the first weeks of a new abstraction. It is real. It is also a poor predictor of who you want in year five, and it is a terrible excuse to stop teaching.

Aptitude still shows, for me, in three narrow places:

Ramp on a new formal system. A type system, a memory model, a logic puzzle with a sharp rule set. Speed-to-first-competence differs. The ceiling after a year of honest practice differs less than people think, unless the work stays at that formal edge — compilers, some kinds of protocol work, some kinds of math-heavy graphics. Most product engineering does not stay there.

Tolerance for a certain kind of frustration. Not “grit” as branding. The willingness to stay with a state that is ugly for an hour without turning it into a personality problem. Some of that is temperament. A lot of it is whether you were allowed to look lost. Temperament is not a gene-for-coding. It is closer to how you handle being stupid in public. That can be trained, slowly, if the room is not cruel.

The first sort, not the last. Screens that look like puzzles will keep selecting the people who are fast at puzzles. That is aptitude-plus-recent-practice for a slice. It is not “developer intelligence.” If your company runs that screen and then wonders why seniors feel different from new hires, you built that difference on purpose.

What I do not treat as aptitude: knowing the framework last season, having a loud Slack presence, finishing the agent’s thought. Those are fashion and access.

Why the “natural” story is so sticky

It flatters the people who already won. It excuses the people who do not want to build a practice path. It is simple enough for a thread. Training stories are slower and they implicate managers: you have to leave the ugly work with someone, you have to review, you have to let them fail a small thing. That is more expensive than a myth.

It also matches how memory works. We remember the moment of insight. We forget the twelve ugly versions. The person themselves will tell you they “just got it.” Ask what they were doing at fourteen. Ask who paid for the machine. Ask how many times they broke a thing and were allowed to stay. The answers are usually ordinary and decisive.

What this means if you are trying to get strong

Stop waiting to feel like a natural. Put yourself where the system can prove you wrong twice a week. Own a slice. Write the postmortem. Do the review where you have to say the risk. Use the agent, then close it and explain the change without looking. If you cannot explain it, you borrowed a line. Borrowing is fine. Calling it intelligence is how you stay thin.

If you are slow at the first puzzle, that is information about the first puzzle. It is not a verdict. I have known too many people who were mid-pack in CS1 and later became the person who could hold the whole billing path in their head because they lived there. Living there was the mechanism. The mid-pack start was a footnote they still apologize for. They should stop.

A note on late starters

The natural story is especially cruel to people who arrive at twenty-eight, or after a different career, or after a language that is not the one the screen uses. Their library is not empty. It is filed under other names: operations, teaching, a trade, a science lab. Transfer is real and it is incomplete. They will look “less intelligent” on a week-one puzzle and more intelligent on a month-four ownership problem if you let month four happen. Most processes do not. Then the process publishes a finding about age or background that is actually a finding about week one.

If you are that person, do not spend a year proving you were a secret natural. Spend it building the software library in public: a finished ugly thing, a postmortem, a system you can narrate. The hallways will still say talent. You will have the cards.

What this means if you hire

If you want “intelligence,” buy training, not a story. Paid time on a real system. A reviewer who will be honest. A path that is not only the leftover tickets an agent now clears. If you refuse to pay for that, you will keep fishing in the pool of people trained by someone else, and you will call your fishing a talent strategy.

Keep a small place for the formal-speed slice if your work is actually that slice. Do not universalize it. And do not write genes in the interview notes. You are not running that study. You are choosing a practice environment. Choose it on purpose.

Developer intelligence, as people mean it in a hallway, is a trained library plus a little variance in how fast the first chapter goes. The hallway prefers a simpler myth. The work does not care. It cares whether the library has the card you need at 3 a.m. You get that card by being there when it is dealt — not by being told you were born holding it. Be there on purpose. Pay for other people to be there. That is the whole unromantic theory.