No Shortcut

In 1957, Isaac Asimov published a short story called Profession, about a world where careers are installed by neural download in an afternoon. A few people don’t get the procedure — they end up learning the slow way, from books, in something the state calls a House for the Feeble-Minded. They turn out to be the only people the society can’t replace. I read it years ago. I haven’t been able to stop thinking about it since the agents started writing my code.

The protagonist, George Platen, has spent his life preparing for Education Day — the day at eighteen when the analyzer reads your brain and decides which profession-tape it can take. George’s brain can’t take any, and he’s shipped off to something the state calls a House for the Feeble-Minded — the name a system gives a place when it wants you to be ashamed to be there. What he eventually learns is that the House isn’t where the failures go but where the originals go — the only minds capable of producing the new tapes that everyone else, a generation later, will download in an afternoon. The system runs on them, and cannot say so out loud, because if it did every ambitious eighteen-year-old would try to fail on purpose.

The genius of Asimov’s setup is that he removes the choice. George doesn’t decide to struggle. He’s selected by failing. The shame is part of the test — it filters out anyone who would game the path on purpose, which means the only people who end up doing the deep work are the ones who couldn’t avoid it.

Our version is harder. The shortcut is on every laptop. The analyzer says yes to everyone. Nobody is going to fail you out of the easy path — there is no House to be sent to, no shame to be ashamed of, no system quietly routing the originals around the tapes. You have to choose, every morning, to do the slow thing when the fast thing is right there and works.

What is the slow thing, when the agent can write the function? It is not the typing. The typing is what the agent took. The slow thing is the work that comes before the typing and the work that comes after — knowing what you’re trying to build, knowing why this piece of it exists, knowing which seams in the system you can move and which ones will tear if you touch them. Naming the problem under the problem the user described. Sitting with a design long enough to see what it forecloses. None of that is slower because of an agent. It is just, now, more obviously the only work left.

What you ship without it is code that works for reasons you can’t articulate, systems whose seams you can’t see because you never had to find them, a career that consists of fluently narrating decisions you didn’t make. It will look fine for a while. It will pass review. The agent’s outputs are good. What you lose is the ability to be the person someone calls when the agent can’t.

Asimov wrote that in 1957, before there were programmers. The story is older than every framework I use. The tools change. The thinking doesn’t — and the thinking is still yours.