The Week It All Went Free
For three years Tomas sold access to intelligence that the world found scarce, and it made him a modest, defensible living. Then four labs reached parity in a span of weeks, the price of thinking fell toward nothing, and the moat he had spent his thirties digging filled in overnight. A quiet story about the week value itself stopped holding still.
On Monday morning Tomas was a man who had built something, and by Friday night he was a man who had built a wrapper.
The distinction had not changed. Only the world's opinion of it had.
He made his coffee the way he always did, ground beans the night before so the noise would not wake the apartment, and he sat down at the desk by the window where, three years earlier, he had written the first version of Margin in a single feverish weekend. Margin was not complicated. He was honest enough with himself, on his good days, to admit that. It took the frontier models — the expensive ones, the ones whose intelligence had been scarce and metered and rationed like a precious metal — and it wrapped them in something a small accounting firm could actually use. It read messy invoices. It reconciled them against ledgers. It flagged the discrepancies a tired bookkeeper would miss at four in the afternoon. The cleverness was not in the intelligence. The intelligence belonged to the labs. The cleverness was in the wrapping, in knowing exactly which questions to ask and in what order, in the four hundred lines of orchestration that turned a general genius into a specific tool.
And the business — this was the part he had been proud of, in the way you are proud of a thing nobody else thinks to be proud of — the business was the arbitrage. He paid the labs for raw intelligence by the token. He charged his customers a flat monthly fee for the answers. The margin lived in the gap between what intelligence cost him and what certainty was worth to a firm that could be ruined by a misfiled number. For three years that gap had been wide and warm and his.
He checked the dashboard out of habit. Eleven hundred and forty paying accounts. Recurring revenue that, after the model bills and the hosting and the one contractor in Lisbon who handled support tickets, left him enough to live without an alarm clock he resented. Not wealth. Sufficiency. He had told himself for years that sufficiency was the harder, rarer thing, and he had half believed it.
The first lab announced on Monday. He read it the way everyone read these things now, as one more press release in a season of press releases — a new frontier model, top of the benchmarks, the usual chart with the usual bars. He noted the price. It was lower than he expected. Not alarmingly. Just lower.
The second lab announced on Tuesday.
This was the part that, looking back, he could never quite reconstruct as a single moment, because it was not one. It was a texture. The second model matched the first on every benchmark that mattered and undercut it on price, and the language in the release was not the confident language of a leader but the clipped, hungry language of someone in a fight. By Tuesday afternoon the first lab had cut its price to match. By Tuesday evening there was a thread, fourteen hundred replies deep, where people were posting the same difficult reasoning problem to both models and arguing about which had answered more elegantly, and the honest answer, the one nobody wanted to type, was that you could not tell. They had arrived at the same place from different directions, like two travelers who had never met agreeing on the shape of a mountain.
Tomas ran his own evaluation suite that night, the private set of three hundred gnarly invoices he had collected over the years, the ones that broke things. He ran them against both new models. They both passed. They both passed faster than the model he was paying for, the one Margin was built on, the one that had been the best in the world fourteen months ago and was now, he understood with a strange cold clarity, simply old.
The third lab announced on Wednesday.
By then the pattern had a shape, and the shape was a race to the bottom of a well that had no bottom. The third model was not better. That was the thing he kept turning over. It was not better than the other two. It was the same — the same parity, the same uncanny agreement on the shape of the mountain — but it was faster, and it was cheaper, and the release barely mentioned capability at all because capability had become the table stakes, the thing you were assumed to have, like a car being assumed to have wheels. The whole announcement was about speed. Tokens per second. Latency in milliseconds. They had stopped competing on how smart the thing was because the thing was as smart as it needed to be, as smart as the others, as smart as smart got this season, and the only ground left to fight over was how fast it could be that smart and how little it could cost.
He did the math on Wednesday night because he could not stop himself. The price of the intelligence inside Margin — the cost he had built his entire margin on, the scarce expensive thing he resold as certainty — had fallen by a factor he had to write down twice to believe. A query that had cost him real money in January now cost a fraction of a cent. By the trajectory of the week it was heading toward something he could only describe as functionally free. Not free. Nothing was free. But cheap enough that the gap he lived in, the warm wide gap between cost and value, was closing like a wound healing over.
The fourth lab announced on Thursday, and Tomas did not even read it carefully. He knew what it would say. Parity. Speed. A price that made the others look greedy. By Thursday the press releases had stopped feeling like news and started feeling like weather.
The emails began Thursday afternoon.
They were polite. That was almost the worst of it. The accounting firm in Cleveland that had been with him since the second month wrote to say they had been experimenting, that one of their junior people had wired up the new free-tier model to their ledger over a weekend, that it did "basically what Margin does," that they were so grateful for everything but they were sure he understood. He did understand. That was the trouble. He understood completely. The junior person in Cleveland had done in a weekend what had taken Tomas a feverish weekend to do three years ago, except now the intelligence underneath was free and fast, and the four hundred lines of orchestration that had been his moat were the kind of thing the free model would write for you if you asked it nicely.
His moat had not been breached. It had simply evaporated, the way a puddle evaporates, leaving behind the faint outline of where water used to be.
He spent Thursday night doing the thing he was good at, the thing that had always saved him before. He went looking for the next gap. There was always a next gap. That had been the unspoken faith of his whole working life, the faith that had let him quit a salaried job and live without an alarm clock: that human ingenuity could always find the next thin warm seam between what something cost and what it was worth, and you could live in that seam, and when it closed you would simply find the next one. He had done it three times before in smaller ways. He believed in the next gap the way other people believed in God or compound interest.
So he looked. He sat at the desk by the window until the coffee went cold and then made more, and he looked for the thing that was still scarce. And what he found, somewhere around three in the morning, was a feeling rather than an answer. Because every gap he could think of — every clever orchestration, every domain wrinkle, every specialized workflow that a small business would pay for — was a gap that depended on intelligence being scarce. And intelligence was no longer scarce. It was the cheapest input in the world now, cheaper than electricity, cheaper than attention, racing toward the price of nothing at the speed of four desperate companies trying to outrun each other off a cliff. Whatever clever thing he built this weekend, the free model could build by next weekend, because the cleverness lived in the intelligence and the intelligence belonged to everyone now, instantly, for free.
He understood, sitting there, that the problem was not that he had failed to build something valuable. He had built something valuable. It had been genuinely valuable for three years. The problem was structural and it was permanent and it was quiet: value itself would not hold still anymore. Anything he could make valuable through ingenuity would be made worthless through commoditization on a timescale shorter than the time it took to make it. The seam closed faster than a person could climb into it. "Just build something valuable" had been the only advice anyone ever gave, the only advice that had ever been true, and it was now a sentence with the bottom cut out of it, because value had become a thing that the machinery dissolved on contact, like sugar in hot water.
By Friday the dashboard read four hundred and ninety accounts and falling, and the falling did not feel like a crisis. It felt like a temperature dropping. He could see the rest of it from here, the way you can see the end of a road from a hill: the firms would all wire up the free models themselves, the way you stop buying ice when everyone gets a refrigerator. There was no villain in it. No one had attacked him. Four companies had simply competed so hard that the thing he sold had become air, and you cannot charge for air, and there was nothing to be angry at because anger needs a direction and this had come from everywhere at once.
He did not cry and he did not rage. He was, he thought, clear-eyed about it, and he took a thin bitter pride in that, the last pride available to him. He had built something. It had been real. It had fed him for three years and that was three years more than most clever ideas ever managed. He told himself this and it was true and it did not help, because the help he wanted was not the help of having done a good thing once. The help he wanted was a future, and the future he could see was just this week again, repeating: build something valuable, watch the value collapse toward zero, build again, watch again, each cycle a little faster than the last, the seam closing a little sooner, until the time to build exceeded the time the thing stayed worth building and the whole proposition quietly stopped being possible. Not with a catastrophe. With an asymptote.
On Friday night he sat by the window with the cold coffee and watched the city, which did not look any different. Somewhere out there eleven hundred firms were getting smarter and freer by the hour, and not one of them would ever know his name, and the intelligence that had been a precious metal on Monday was running through the wires now like water through a city that had forgotten it was ever a desert.
He opened a new file. The cursor blinked in the white. He had done this a thousand times, and the blinking had always meant the same thing — possibility, the next gap, the next warm seam. He kept his hands on the keys. He could think of three things to build. He knew, with the cold clarity the week had given him, that all three would be free by the time he finished them, and that he would build one anyway, because building was the only thing he knew how to be, and a man has to do something with his hands while the value drains out of the world.
He did not type. He sat there with his hands on the keys and the cursor blinking, and the cursor did not care whether he typed or not, and outside the intelligence kept getting cheaper, and the night went on being exactly as long as it was going to be.