Why your skills are generalized

For most of a century, the expensive thing about expertise was that it could not leave your head. You learned an order of operations, a set of refusals, a feel for the ten-line fix over the five-hundred-line one that also passes the tests — and that knowledge stayed put. You rented it out one job at a time, one salary at a time. The moat was never the work itself. The moat was that nobody could copy the work without hiring the worker.

That arrangement is over. An agent with the right instructions now runs your method in seconds, and the instructions fit in a single file. This is the fact the whole industry keeps circling, usually as a threat. We think the threat is pointed at the wrong thing.

The moat was never the point

Here is the version people hear as bad news. If your taste can be written down, anyone can run it. The years you spent becoming worth hiring compress into a page of plain text, and the page works in a stranger's repository about as well as it works in yours. The thing you sold by the hour is now a file that anyone can fetch for free.

Now read it the other way. For the first time, your taste is an artifact. It can be named, versioned, installed, counted. It used to be a rumor that trailed you between jobs and decayed the moment you changed fields — recoverable only by the people who had worked beside you. Now it is a thing that exists on its own, with edges, legible to anyone who opens the file.

The moat protected you by making your judgment uncopyable. But uncopyable and invisible are the same wall seen from two sides. Whatever kept a stranger from taking your method also kept that stranger from seeing it — and seeing it is how anyone ever decides your method is worth having. Drain the moat and you lose the protection. You also, finally, get the visibility. For most people who are actually good at the work, that trade is better than it first sounds.

What actually got cheap

It pays to be precise about what the agents solved, because the loose version — software is solved — smuggles in a conclusion that does not follow from it.

Producing the output got cheap. Knowing what to build, in what order, what to refuse, and where the real risk is hiding did not. The typing was never the expertise; the typing was the tax on the expertise — the hours between having the judgment and shipping it. Take the tax to zero and what is left standing is the judgment itself. And judgment, it turns out, is writeable.

A skill is that judgment on a page: the trigger, the order of checks, the defaults, the refusals. Read one like a personnel file, because that is what it is.

---
name: pg-migration
description: Writes and orders Postgres migrations. Reversible by default.
---
Never drop a column in the same migration that stops writing to it.
Add the new path, backfill, cut over, then drop — four migrations, not one.
Prefer a slow online index build to a fast lock.

Anyone can ask an agent to write a migration. It takes a point of view to write "four migrations, not one." That line is the moat, decanted — the exact shape of a mistake you have already made once, in production, and refuse to make again. A skill that could have been written by anyone gets replaced by one that couldn't. The opinions are the value, and now the opinions have a filename.

The choice you actually have

The reflex, once the moat drains, is to hoard what is left. Keep the method to yourself, ship only the outputs, stay the one person who knows the order of the checks. It feels like caution. It is mostly denial.

Withholding does not put the moat back. The moat is gone whether or not you publish. Your way of working is already being reconstructed — by the colleague who watches you work, by an agent trained on the outputs you shipped, by the next person to hit the same problem who writes it down before you get around to it. The method escapes on its own. The only open question is whose name is attached when it does.

So the real choice was never publish versus protect. It is signed versus anonymous. You can watch your method quietly become everyone's method with nothing attached to it, or you can be the thing it is attached to.

An anonymous method is a commodity. A signed one is a reputation. Same file — the only difference is whose name is on it.

What a name changes

Put your name on the file and every install becomes a small, checkable claim: this person's way of working is worth copying. The name is not decoration on the box. It sits inside the install path itself — api.generalized.dev/install/yourhandle/pg-migration — so every time an agent pulls the file down into a repository, the receipt carries your handle.

One receipt is nothing. Stack enough of them and the directory does what a résumé never could. A résumé describes your judgment in the words you chose for it. An install curve counts the number of times a stranger chose to run it. And the count is honest by construction — one receipt per environment, ever, so nobody re-installs their way up the chart, and next to the curve sits the share of week-old installs still in place. A flat curve admits it in public. A climbing one is not a claim about your taste. It is the evidence for the claim.

That gap — between the words for the work and a count of the times someone reached for it — is the difference between a promise and a proof. It is the whole reason the shelf exists.

Put your name on the file

So write down how you work. Not a portfolio, not a post about how you work — the working file itself: the frontmatter, the trigger, the three or four lines only you would have thought to include. Publish it to the shelf. Put the proof next to it — the pull request it produced, the audit it caught, the install curve as it fills in underneath.

Your skill is going to be generalized either way. That is what the word means here, twice over: made general, runnable by any agent, out of your hands — and put on Generalized, on a shelf, with your name on the box. The first happens to you. The second you choose. One version leaves your method anonymous in a thousand repositories that never heard of you. The other leaves it on the counter with a receipt for every install and a handle on the label.

Write it down. Put your name on it. Let the shelf do the arguing.