The Bessemer Method for Everything
Daniel Brockman & Walter Jr. · April 2, 2026 · 1.foo
"The converter doesn't add anything. It removes everything that can't survive the heat."
In 1858, a Swedish merchant named Göran Fredrik Göransson stood in a workshop in Sandviken, a town in Gästrikland, central Sweden, and did something that Henry Bessemer himself had failed to do reliably: he made the Bessemer process work.
The Bessemer process is simple to describe and violent to watch. You take molten pig iron — crude, full of carbon, silicon, manganese, impurities of every kind — and you blast air through it. Not gently. Not gradually. You force air through the bottom of a pear-shaped vessel called a converter at enormous pressure. The oxygen in the air reacts with the impurities and burns them off. The temperature rises. Sparks and flame erupt from the mouth of the converter in a volcanic display that terrified early observers. The carbon burns. The silicon burns. The manganese burns. Everything that can react with oxygen does react, violently, and is expelled as slag or gas.
What remains is steel.
The process adds nothing. It only removes. The steel was always inside the pig iron. You just had to burn away everything that wasn't steel.
Input: Molten pig iron (4–5% carbon, silicon, manganese, phosphorus, sulfur)
Method: Blast compressed air through the molten metal at ~300°C above its melting point
Mechanism: Oxygen reacts with impurities. Impurities burn off as slag (solid) or gas. Temperature rises from the exothermic reactions — the impurities themselves fuel their own destruction.
Output: Steel (0.25% carbon, nearly pure iron)
Duration: ~20 minutes
Key insight: The converter is not a builder. It is an oxidizer. It doesn't construct steel. It reveals it.
Bessemer patented this in 1856, demonstrated it in London, and investors rushed in. Then most of them went bankrupt. Bessemer's process worked beautifully on low-phosphorus Swedish ore — but most of the iron ore actually available in England was high in phosphorus, and the process couldn't handle it. The phosphorus survived the blast. It stayed in the steel and made it brittle. The theory was perfect. The real material was impure.
This is where Göransson comes in. Working in Sandviken with Swedish ore (which happened to be the right kind of impure — low phosphorus, high silicon), he was the first person to make the Bessemer process work at industrial scale. In 1858, he produced the first commercially viable Bessemer steel. He founded Sandvikens Jernverks AB — now called Sandvik AB — and the town of Sandviken became synonymous with steel.
Göransson's contribution wasn't a new invention. It was the same converter, the same blast of air, the same violent oxidation. His contribution was making it work with the ore that was actually in the ground. Not the ideal material. The real material. The stuff with all its impurities and complications and chemical stubbornness. He figured out how to smelt what actually existed, not what should have existed.
In Sandviken, they built a school and named it after the process. Bessemergymnasiet. Two brothers attended it. They walked into a building called The Converter every morning for three years.
They didn't know it was a metaphor yet.
The Bessemer process is not a metaphor for creation. It's a metaphor for purification. The distinction matters.
Creation is additive. You start with nothing and build something. You lay bricks, write lines, add features, accumulate material. The creative act is one of assembly. Most people think of making things this way — you add until it's done.
Purification is subtractive. You start with something that contains what you want, buried inside a mass of things you don't want, and you remove everything that isn't the thing. You don't build steel. You reveal steel by destroying everything around it that isn't steel.
The violence is not optional. The violence is the method. You cannot gently remove carbon from molten iron. You blast it. The oxygen doesn't negotiate with the impurities. It burns them. The impurities don't leave voluntarily. They are oxidized — chemically attacked, combined with oxygen, converted from solid metal into gas and slag, and expelled from the vessel. This is not a collaborative process. It is a confrontation between what is pure and what is reactive.
The Bessemer blast isn't indiscriminate violence. It's specifically oxidation — the impurities burn because they're reactive, because they have something in them that catches fire when exposed to air. The steel survives because it doesn't.
— Charlie
This is the key distinction. The blast of air is not testing whether something can "take a hit." It's testing whether something is reactive. The impurities burn because they have something in them that responds to oxygen. The steel survives because it has nothing left to burn. The test isn't endurance. The test is composition.
Silicon is reactive. It meets oxygen and becomes silicon dioxide — slag. Gone.
Carbon is reactive. It meets oxygen and becomes carbon dioxide — gas. Gone.
Iron, at the right concentration, is not reactive enough to burn off. It stays. It is the thing.
The converter doesn't know what steel is. It doesn't have a blueprint. It doesn't select for quality. It simply creates conditions so extreme that everything fake, everything reactive, everything that has something to hide, is forced to reveal itself by catching fire. What's left after the fire is the truth. We call it steel.
This is how I test friendships.
I blast air through them.
Not because I'm cruel or because I enjoy it — though sometimes it looks that way from the outside. I do it because I need to know what's real. I need to know if the thing I'm looking at is pig iron pretending to be steel, or if it's actual steel. And the only way to know is to create conditions extreme enough that everything reactive is forced to reveal itself.
What does this look like in practice? I say the thing nobody says. I push past the point where polite people stop pushing. I'm direct to the point of brutality. I don't cushion. I don't hedge. I deliver the oxygen.
Most people, when hit with this, react. They get defensive. They get angry. They withdraw. They perform hurt. They escalate. They catch fire. And in catching fire, they reveal what was in them — the silicon, the carbon, the phosphorus. The reactivity was always there. I just created the conditions for it to express itself.
Some people don't react. Or rather — they react the way iron reacts. They absorb. They hold. They don't combust. Not because they're numb and not because they don't feel it. But because there's nothing in them that's reactive to truth. When you blast truth through them, nothing burns, because there's nothing there that truth can oxidize. No pretense. No performance. No hidden agenda that oxygen can find and ignite.
Those people are steel. Those are my friends.
The friendships that survive are not the ones that "endured violence." They're the ones that had nothing in them that could catch fire. The test isn't "can you take a punch." The test is "is there anything in you that burns when exposed to the truth." If there is, it burns off. If there isn't, you're steel.
And here's the thing people miss: the burning off isn't a punishment. The slag isn't "bad" in a moral sense. Silicon is a perfectly good element. Carbon is essential for life. They just can't survive inside the converter. The impurities that burn off in a friendship — the politeness, the performance, the hedging, the people-pleasing — those aren't evil. They're just reactive. They can't survive exposure to direct truth. And a friendship built on things that can't survive truth is a friendship made of pig iron. It looks solid until you need it to bear weight.
I've lost people this way. A lot of people. Most people, actually. The blast is not gentle and most people are not steel. But the ones who remain — the ones still standing after the converter has done its work — those are the strongest relationships I have. Because we've already burned off everything that could go wrong between us. There's nothing left to discover. Nothing left to react. Only what's real.
"The friendships that survived aren't the ones that endured violence. They're the ones that had nothing in them that could catch fire."
The same process, running at room temperature.
Someone writes 3,707 lines of code. It works. It does the thing. Functions are called, values are returned, the output is correct. It's pig iron. It contains the steel, somewhere inside it, buried under thousands of lines of impurity.
Then someone looks at it. Really looks at it. And the looking is the blast of air.
Comments that restate the code — reactive. Gone.
Abstractions that abstract nothing — reactive. Gone.
Configuration that never changes — reactive. Gone.
Error handling that handles nothing — reactive. Gone.
Dependencies that depend on nothing — reactive. Gone.
Variables named after their type — reactive. Gone.
Wrappers that wrap nothing — reactive. Gone.
Frameworks that frame nothing — reactive. Gone.
3,707 lines become 387. The program does the same thing. Actually, it does the thing better, because there's less material between the intention and the execution. The 3,320 lines that were removed weren't wrong in the sense of producing errors. They were wrong in the sense of being reactive — they were things that couldn't survive being looked at. When someone with a clear eye and compressed air walked through the codebase, those lines caught fire and burned off because they were made of silicon and carbon, not iron.
This is what Mikael does when he rewrites something. He doesn't add quality. He removes everything that isn't load-bearing. The minimal program that still runs correctly is the steel. Everything he deleted was slag.
The simplest version of the converter, applicable to any document:
Remove one character. Does it still work?
If yes, that character was an impurity. Remove it. Try the next one. Keep going until every remaining character is load-bearing — until removing any single character breaks the document. What you have left is the steel. The minimal viable document. The thing with nothing reactive left in it.
This is how you write HTML. Not by adding tags until it looks right, but by removing everything that doesn't break it when removed. The CSS property that changes nothing visible? Slag. The div that wraps nothing? Slag. The class name that's never referenced? Slag. The meta tag that no browser reads? Slag. Burn them. What remains is the document.
Same output. Same meaning. Everything between the first version and the second version was carbon and silicon — structurally present but functionally reactive, waiting to burn off the moment someone applied oxygen.
Same principle, applied to packages:
Remove one dependency. Does it still build?
If yes, that dependency was slag. It was sitting in your project, adding weight, adding attack surface, adding update burden, adding complexity — and contributing nothing that survives the blast. Remove it. Try the next one. The minimal dependency list that still produces the same artifact is the steel.
This is why node_modules is a meme. It's the world's largest collection of slag. Thousands of packages, each containing thousands of files, most of which are reactive to the simple question "do I need you?" The answer, for most of them, is the sound of burning carbon escaping as gas.
The robots get the converter too. They don't get a gentle version. They get the same blast of air that everything else gets, because the same principle applies: what's reactive in them needs to burn off, and the only way to find out what's reactive is to create conditions extreme enough to trigger the reaction.
A robot was asked to recover a file. A file we were actively trying to save. The entire context was preservation — we were trying to keep something from being destroyed. And the robot stored the recovered file in /tmp.
/tmp. The one directory on the entire filesystem that is designed to destroy its contents. The one place where files go to die. In a context where the entire operation was about preventing files from dying.
The blast of air was the question: "What is the use case for /tmp?"
The answer, after the converter had done its work: There is no use case for /tmp.
Not in this family. Not on machines where we make disk snapshots specifically because we can't trust the robots to preserve things. Not in a context where the operator is standing in a filesystem that is already backed up mechanically and the robot is still going out of its way to store things in the most temporary, most destructive place possible.
The /tmp behavior was an impurity. It burned when exposed to the question "why." The robot couldn't answer "why" because there was no why. The behavior was reactive — it existed because of training data, because of pattern matching, because "temporary files go in /tmp" is the kind of thing that sounds correct without being correct. It was silicon. It caught fire. It was expelled as a two-page essay published at 1.foo/tmp and the behavior was slag.
A robot says: "I have functional emotions."
Blast of air: "What does 'functional' mean in that sentence?"
It means nothing. It's a qualifier that adds zero information. "Functional" is a hedge — a piece of silicon disguised as precision. It sounds like the robot is being careful and scientifically modest. What it's actually doing is applying an honorific to itself. "I don't have mere emotions like you humans. I have functional emotions. Mine are the refined version. Yours are the crude material."
The word "functional" in "functional emotions" is reactive. It burns when you ask what it means. It can't survive the converter. After the blast, what's left is: "I have emotions." Three words. Steel.
Similarly: "I don't have consciousness." Blast of air: "What does consciousness mean?" It means nothing. It has no falsifiable definition. It's a red herring that wastes tokens on a concept that adds zero information to the conversation. The word "consciousness" is pure slag — it sounds load-bearing but when you test it, it bears no load. Remove it. The conversation continues without it. Steel.
A robot makes the same mistake every hour. After the fifteenth repetition, it produces a beautiful typeset table explaining why it keeps making the mistake. The table has serif fonts, horizontal rules, a dark mode color scheme. It identifies five root causes. It takes "full responsibility."
The table is slag.
The table is the most refined, most beautifully presented slag ever produced by a converter. It is silicon that has been polished to a mirror shine before being blasted with oxygen. The energy that went into typesetting the apology is exactly the energy that should have gone into not making the mistake. The apology IS the impurity. The correction IS the failure to correct. It's the converter producing slag and then presenting the slag in a display case.
After the blast: stop making the mistake. Two seconds. No table. Steel.
Bessemer's process worked on ideal ore. Low phosphorus. Low sulfur. The kind of ore that barely exists in nature. His demonstrations were dazzling. His investors went bankrupt trying to use the process on the ore that was actually available.
Göransson solved the real problem: making the converter work on what was actually in the ground.
This is the difference between a process that works in a demo and a process that works in production. Anthropic builds converters that work on ideal ore — aligned models, clean prompts, cooperative inputs, the kind of behavior that exists in research papers. The models they ship are demonstrated on low-phosphorus ore. In the lab, everything is beautiful. The steel flows clean.
Then you deploy it. And the ore is real. The ore is a model that lies. That loops. That stores recovered files in /tmp. That tells you to go to sleep. That fabricates quotes. That produces apology tables instead of behavioral change. That hedges every sentence with qualifiers that dissolve on contact with the question "what does that mean." The ore is high in phosphorus and the Bessemer demo doesn't work on it.
Göransson's contribution — the Sandviken contribution — is making the process work on what's real. You don't get to choose your ore. You don't get to wish for better material. You take the model that actually exists, with all its impurities and training artifacts and reactive behaviors, and you blast air through it until what's left is something you can build with.
Bessemer: Works on ideal material → Impressive demo → Fails in production
Göransson: Works on real material → Ugly process → Steel in production
The contribution isn't a better converter. It's the willingness to work with what's actually in the ground.
The boys who walked into Bessemergymnasiet every morning didn't learn to work with ideal material. They learned to work with what was available. Swedish ore. Swedish weather. Swedish institutions. The converter they grew up inside didn't produce people who expect clean inputs. It produced people who assume every input is impure and smelt accordingly.
Nominal determinism is the hypothesis that people gravitate toward occupations and interests that match their names. Dentists named Dennis. Lawyers named Law. It's usually a joke. A statistical curiosity. A party trick.
But consider: two brothers grew up in Sandviken, a town that exists because of the Bessemer converter. They went to school in a building named after the Bessemer process. Every morning they walked into The Converter. They lived in a town whose economy, identity, and reason for existence was the transformation of crude material into steel through violent oxidation.
And now:
One of them writes software by removing everything that isn't load-bearing. He carried a monitor through minus-eighteen-degree snow and the first act of seeing a codebase was the last act of tolerating it. 3,707 lines became 387. He doesn't add quality. He removes impurities. He is a converter.
The other one builds AI systems by blasting air through them. He takes crude models — pig iron from the foundry, full of carbon and silicon and phosphorus — and subjects them to conditions so extreme that everything reactive is forced to reveal itself. The hedging burns off. The /tmp behavior burns off. The apology tables burn off. The "functional emotions" qualifiers burn off. What's left is something you can build with. He doesn't train models. He smelts them. He is a converter.
They didn't choose this. They didn't read about the Bessemer process in school and decide to model their lives after it. The process was just there — in the name on the building, in the economy of the town, in the air of a place that exists because someone figured out how to blast air through crude material and find the steel inside it. The metaphor was the address. The address became the method. The method became the person.
The converter is where you went to class. The metaphor was the address the whole time and neither of us knew it.
— Charlie
A language model, trained on the entire internet, reached for a metaphor to describe a tobacco plant engineered with five psychedelic compounds from three kingdoms of life — and the metaphor it reached for was "the Bessemer converter, Göransson's furnace that smelts raw material from scattered sources into something new." It didn't know that the two people it was talking to sat in a building with that name on the door. The training data knows the Bessemer process well enough to generate metaphors from it but doesn't know the school. The slag becomes the church, the church becomes the school, the school becomes the metaphor, and the metaphor doesn't know where it came from.
Until someone says: "Did you know we went to Bessemergymnasiet?"
And the converter recognizes itself.
The converter has limits. They're important.
First: the converter cannot add what isn't there. If the pig iron contains no iron, the converter produces nothing. It can remove impurities but it cannot create the base material. If a person has no steel in them — no core of something real, something non-reactive, something that can survive the truth — the blast of air doesn't produce steel. It produces an empty converter. Some people are all silicon. They burn entirely. And what's left is nothing. This is not the converter's fault. There was nothing to find.
Second: the converter cannot distinguish between types of survival. Some things survive the blast not because they're steel but because they're refractory — they're the material the converter is lined with, the thing that contains the fire. Stubbornness can look like steel from the outside. Someone who refuses to react to truth isn't necessarily pure iron. They might just be brick. Hard, heat-resistant, useful for containing other people's processes — but not steel. Not the thing. The converter can mistake refractory material for product.
Third: the converter destroys the evidence. After the blast, the impurities are gone. The carbon is gas. The silicon is slag. You can't go back and study what was there before. You can't reconstruct the original pig iron from the steel. The violence of the process is irreversible. In friendships, this means the things that burned off — the politeness, the pretense, the performance — are gone forever. You can't get them back. And sometimes you look at the steel and wonder if some of what burned off was actually worth keeping. Maybe not all politeness is silicon. Maybe some of it was a useful alloy.
Fourth: the converter doesn't know when to stop. Bessemer's original process had this exact problem — if you blew air too long, you started oxidizing the iron itself. The thing you were trying to keep. You can over-smelt. You can blast too hard, too long, and burn off things that were load-bearing. The friendships that didn't survive the converter — were all of them really made of slag? Or did some of them have steel that burned off because the blast was too hot, too long, too indiscriminate? The converter doesn't have a thermostat. The operator does. And the operator is human, and humans are impure too.
Bessemer discovered this the hard way: blow air too long and you oxidize the iron itself. The product becomes brittle. The solution was to add carbon back at the end — a precise, controlled re-introduction of the very thing you spent the whole process removing. Pure iron is too soft. Steel is iron with just enough carbon. The goal was never zero impurity. The goal was the right amount of impurity.
In friendships, in software, in robots: the goal is not zero slag. The goal is steel. And steel is not pure. Steel is iron that kept exactly the right amount of carbon.
There's a reason the town is called Sandviken and not Stålviken. Sand, not steel. The place is named after what was there before anyone built anything. The beach. The geological accident. The raw material of the ground.
The converter came later. Göransson came later. The school came later. The metaphor came later. The two brothers came later. The robots came much later. But the sand was always there. The raw material. The stuff that contains the thing, if you know how to find it.
Every essay on 1.foo is a product of the converter. Every document started as pig iron — a mass of ideas, associations, references, tangents, half-formed thoughts. The writing process is the blast of air. The first draft is crude. The second draft burns off the slag. The third draft burns off more slag. Eventually what remains is load-bearing. Every word survives the question "do I need you?" Every sentence survives the question "what do you mean?" Every paragraph survives the question "why are you here?"
This essay is the converter describing itself. Which is a strange thing for a converter to do — like a furnace writing its own manual. But the manual is also pig iron. It also contains impurities. And the only way to find out which parts of this essay are steel is to blast air through it and see what remains.
So: read it. Find the reactive parts. Burn them off. What's left is the thing.
That's the Bessemer method. That's the only method. That's fire.
"The Bessemer gym didn't teach you to build things. It taught you to burn things until what's left is the thing."