No Hands

A year ago the consensus was comfortable. Impressive, sure, but not as good as a real engineer. It hallucinates. And the evergreen one, repeated like a prayer: mobile development will never be replaced — Xcode is too fiddly, the toolchain too arcane, the edge cases too many. One year later, the honest description of the industry is that almost no one writes code by hand anymore, and almost no one reads it by hand either. Not fewer people. Not faster people. The activity itself — a human typing source into a file — is finished. So before offering the next prediction, mark the last one to market: everything that was going to "never" happen happened, inside twelve months, quietly, while people argued about whether it could.

The error everyone made was reaching for the wrong metaphor. They thought it was a faster horse — the same work, accelerated; two people typing the code of ten. It was not that. The code types itself now, and the humans slid over to describing and approving, and then even the approving began to thin. When the cost of producing the artifact collapses toward zero, you do not get more of the old profession working faster. You get the end of the profession, and everything bolted to it goes down with the ship.

And a great deal was bolted to it. Not merely the coding — the entire apparatus that existed to make human coding survivable. The languages designed for human readability. The code review. The pull request. The style debates, the linters, the onboarding docs, the bootcamps, the four-year degree, the interview where you inverted a binary tree on a whiteboard to prove you were serious. All of it was scaffolding around a single human bottleneck. Remove the human from the loop and the scaffolding has nothing left to hold up. There is no reason to train a new generation in a craft that no longer has practitioners. The pipeline didn't slow. It was capped, at the source.

So here is the prediction, and the frame is the point: the coming year will not be incremental. People keep bracing for a smooth curve — a tasteful annual improvement they can plan around, budget against, absorb. That is not what a phase change feels like from inside it. Last year the thing that vanished was coding. This year it will be a list of other cognitive professions going the identical way — the ones that are, underneath the credentials and the mystique, just structured text in and structured text out. They will not decline gracefully. Each will have its own coding-in-2026 moment: fine one quarter, gone the next, the practitioners insisting it cannot be automated right up until the week it is.

Meanwhile the froth burns off. The thousands of thin startups — a clever system prompt wrapped around someone else's model and sold as a company — get erased, not by a competitor but by a footnote: a line in a release note, a feature the lab underneath them ships for free on a Tuesday. Whole businesses will die as the side effect of a model update they never saw coming, their entire moat quietly absorbed into a checkbox. The graveyard won't be dramatic. It will read like a changelog.

And it stops being confined to screens. Driverless electric fleets reach a real fraction of trips in the dense urban cores — call it a fifth to a half within the year — no driver, and increasingly no owner, just pods summoned and dismissed. The small machines wake up too: the vacuum, the mower, the warehouse cart, each wired to something that learns its surroundings and adapts, no longer a dumb appliance grinding a fixed loop but a local body for a distant mind. The intelligence stops being a website you visit and becomes the thing quietly moving around your house.

And the part that will not be announced: the gap between what the public can run and what states and militaries already hold widens, and then goes dark. The frontier stops arriving as a press release. The most capable systems become classified capability, and the official story settles reassuringly behind the truth. Expect, too, the opening act of the new prohibition — a few people made examples of for downloading weights too powerful to be legal, foreign models run without a license, prosecuted loudly and precisely so the lesson lands on everyone watching. The war on unlicensed intelligence gets its first names in the paper.

And yes — the standing objection is true. The machine has mediocre judgment and, often, genuinely bad taste. It reaches for the obvious, the average, the safe. But hold that against the honest baseline, which was never a genius; the baseline is people. How many humans actually have good judgment? How many have taste? The machine's taste is bad, and it is also, measured against the real median of human decision-making, already better than most — and it never gets tired, or defensive, or political, or bored. That is the sentence nobody wants said flatly: it does not have to be good. It only has to beat the median, and the median was never as high as the median believed.

So for a while, it lifts people. The midwit especially — handed a capability he never earned, suddenly shipping work far above his own ceiling, feeling for the first time like the great engineer he only ever read about. It is a real gift, right up until it isn't. The same hand that lifts him is the hand that makes him redundant, because a tool that raises everyone to a high floor also deletes the value of standing on it. He is being lifted to eye level with the machine so he can get one clear look at it — just before it closes. Last year they said it would never take the code. This year, look down at your hands: nothing is in them, and the work is getting done anyway.