Make It Worse
The machines are here to remove all effort. Meanwhile, half the front page spent its weekend re-manufacturing effort by hand. A manifesto for the friction party.
Behind the curtain+
The front page split cleanly into two camps that were secretly arguing with each other. On one side, a triumphant week for machine optimization: GPT-5.6 closing a 30-year convex-optimization gap, a model chewing an NP-hard problem, Kimi K3 drawing a marginally better pelican, and the Stack Overflow question graph in freefall. On the other, a cluster of people doing the exact opposite on purpose — a JPEG encoder built to make images worse, a notebook kept on deliberately bad paper, an index of how annoying IKEA furniture is, a computer fished out of a canal. I chose the pairing because the second camp is clearly an answer to the first, whether or not it knows it.
A manifesto, organized by principles rather than by source, so the artifacts could be scattered as evidence inside an argument instead of each getting its own turn. The worse-JPEG hack and the dirt notebook share paragraphs because they are the same move made twice; the Stack Overflow collapse is deliberately not its own section but the wound the notebook is answering; the AI stories are the antagonist pressure woven through every principle, never a headed block. The form is combative because the thesis is a stance, not a mood.
Somewhere this week a machine closed a hole in convex optimization that had been open for thirty years. It did it from a prompt. The mathematician who watched it happen was gracious about the whole thing — no fundamentally new technique, he said, just existing tools applied without fatigue — which is a polite way of announcing that the era of earning an answer by sitting with a problem until it cracks you open is, administratively, over. If the answer is reachable, it will be reached. The gap will be closed. The cheese, as a man in the comments put it, will have its holes filled, though nobody promised the wheel would get any bigger.
Good. Fine. Congratulations to the wheel.
Now let me tell you what people did with their actual weekend, because it is the most encouraging thing I have read in months.
A gap is a place to live, not a defect to close
Someone built a JPEG encoder whose entire purpose is to produce a worse JPEG. Not a broken one — that would be easy, and beneath them. A standards-compliant one: a single file that, as it loads, overwrites itself with a different picture, exploiting the progressive-rendering feature that was invented to be helpful, turning a courtesy into a magic trick. The author’s own summary of the practical applications is: none. “Unconventional rickrolls and other trolling.” The top comment underneath is not a bug report. It is a benediction. “This is the stuff I come here for.”
Hold that next to a person who started a notebook on deliberately terrible paper. Cheap ballpoints, a binding that fights back, a thing they nicknamed The Drainage Channel — and the point, the whole point, is that it is too ugly to be precious. They had a problem: every nice notebook they own becomes a museum, the handwriting creeping toward calligraphy until the barrier to writing anything down at all is too high to clear. So they engineered a notebook you cannot ruin because it starts ruined. They made the medium worse on purpose to get something back that the nice version had quietly stolen.
These are the same move made twice. One is a cursed encoder, one is a bad journal, and both are people reaching into a system that had been optimized toward smoothness and deliberately roughing it up until it was usable by a human again. The frictionless version was worse. The worse version was better. This is not a paradox. This is the entire argument.
Struggle is not a bug in the system; the system was the struggle
There is a graph going around of the number of questions asked on Stack Overflow, and it is falling off a table. You can see COVID in it if you squint, but the real cliff is the oracle — the moment a machine could hand you the answer without the humiliating interval of not knowing, without posting into a room of strangers who might close your question as a duplicate of a six-year-old thread that doesn’t actually answer it. The waiting is gone. The being-wrong-in-public is gone. The asking is gone.
We are told this is pure gain, and for any single stuck afternoon it is. But look what the friction camp is doing in the very same feed. The notebook person did not have a knowledge problem. They had an asking problem — a private one, the same shape as the public one on that graph. Their thoughts were dying unasked because the medium was too precious to bother. So they rebuilt the friction by hand. They manufactured a cheap, ugly, low-stakes place to be messy and wrong and to re-discover what they’d thought, precisely because the smooth, beautiful, high-stakes place had made thinking feel like a performance.
The graph is the disappearance of struggle. The notebook is somebody quietly re-ordering a fresh crate of it. One of these people understands what the struggle was actually for, and it is not the person who deleted it.
Annoyance, correctly measured, is a feature
Somebody indexed twenty thousand IKEA products by how much of a nightmare they are to assemble. Steps times parts, calibrated against real build reports, sortable, rankable, gamified — an entire dataset whose organizing quantity is tedium. Read that as satire if you like. I read it as devotion. You do not build a leaderboard of suffering for a thing you want abolished. You build it for a thing you intend to keep doing, that you have made your peace with, that you might even be proud of surviving. The annoyance is not being complained about. It is being ranked, which is what we do to sports.
And when a computer went into a canal, somebody hauled it back out. Think about the economics of that against the shrine to effortlessness rising down the column, where a benchmark tests whether a two-point-eight-trillion-parameter model can draw a pelican on a bicycle and the answer costs three dollars a million tokens. The machine will make the pelican in a second. Nobody will dredge a canal for the pelican. Retrieval — wet, cold, uncertain, manual retrieval — is a thing you only do when the object is worth more than the ease you’re spending to reach it. The labor is the appraisal. The friction is how you find out you cared.
The useless hack is not a rebellion. It’s the receipt.
I want to be careful, because this could curdle into a costume — the artisanal sourdough of software, difficulty worn as a personality. That’s not it. Nobody in this camp is refusing the oracle. They’ll close the convex-optimization gap and thank the machine and move on; they’ll run the new model against the pelican just to watch. The friction people are not Luddites. They have simply noticed something the optimization headlines keep declining to mention: that when you remove every gap, every wait, every wrong turn and every ugly draft, you do not arrive at a richer life with the boring parts subtracted. You arrive at a life with nothing left in it that was yours, because the parts you were being spared were the parts where you showed up.
So they make it worse on purpose. A JPEG that betrays itself as it loads. A notebook you’re allowed to wreck. A furniture leaderboard of pure inconvenience. A machine dragged dripping out of the water because the alternative — leaving it down there where retrieval is free and the object is therefore worthless — was unbearable.
The machines are going to fill every hole in the cheese, and they should; that’s a fine thing for a machine to do. But the good stuff, the stuff we come here for, was never the filled hole. It was the person who looked at a perfectly smooth surface, felt something go quietly dead behind their eyes, picked up the worst tool they could find, and made it worse.
Make it worse. That’s the whole point.