The Horse That Trained Harder

Picture a horse in 1840 watching a steam train pull out of a station, deciding the answer is to try harder. Faster mane. Better stride. Maybe some cardio. The horse isn’t doomed because it lost the race. It’s doomed because it’s still running it.

That’s most of the AI conversation in 2026, and there’s something almost tender about it. Coders speed-typing against Claude at 2am like it’s a duel. Junior associates billing more hours to look busy in front of software that doesn’t sleep, doesn’t bill, and doesn’t care. Writers proving they can churn out drafts faster than ChatGPT, which is a bit like winning a hot-dog eating contest against a wood chipper. Wrong contest, all of it. Industrial revolutions don’t kill horses — they make horses irrelevant on the part of the map where horses used to matter, which is a quieter and sadder thing than being killed, and also harder to put on a motivational poster.

Looms made cloth cheap, which created accountants, then office workers, then HR — none of which existed as full-time roles before anyone needed them, and all of which would have sounded absolutely demented to a weaver in 1790. Damien Charlotin made the same argument in May about lawyers: U.S. law school applications up about 40% over two cohorts, employment for new grads at an all-time high. His read is the Jevons paradox coming for legal work — cheaper advice gets used in more places by more people, so you end up with more lawyers, not fewer. NYU’s Max Raskin went further in the same paper and argued chatbots should be allowed to practice law outright, which is either the future or a really elaborate way to lose a custody battle. The horse, in this telling, gets to deliver mail. We don’t know yet what the new accountants of 2046 will do. The industrial revolution didn’t either, until it did, and then everyone pretended they saw it coming.

What’s different this time is what’s being industrialized. Not muscle. Information. And information has a property coal didn’t — whoever refines it gets to define what reality looks like for everyone using it. Coal at least had the decency to just be coal.

Half of that is already visible in the plumbing nobody writes headlines about. AWS wired crypto payments into AI agent infrastructure with Coinbase and Stripe. Solana and Google Cloud rolled out crypto rails for AI APIs. Tether-backed Oobit launched Visa cards for AI agents to spend crypto on their own, which is a sentence that would have gotten you committed in 2019. Crypto-Reddit insists the agents have “chosen Bitcoin”, as though somewhere a GPT cluster is doing technical analysis on a candle chart. It isn’t. The rails the infrastructure actually settled on are quieter, less ideological, much less fun to post about. But the underlying observation is right, and worth sitting with for a second: software has a wallet now. Not metaphorically. A line of code can hold money, spend it, lose it, earn more of it, and probably get scammed eventually. We made a thing, and the thing buys its own oats.

The other half is borders. Not the river kind. When you can translate any post on any social network in real time, the thing that grouped you with your neighbors stops being geography and starts being whatever the algorithm decided you’d like at 11pm on a Tuesday. Your nation is wherever your feed says it is. And the feed runs on a model. If your translation layer, your search layer, and your agent layer all belong to someone else, the country you actually live in is the one that owns those layers — not the one on your passport, which is increasingly just a vibe. This is probably why every mid-sized country with a research budget has decided it needs its own foundation model. It isn’t pride. It’s closer to currency than airspace. Letting a foreign model mediate your information environment is letting someone else’s vocabulary become the one your citizens think in — something like informational gatekeepers, closer to algorithmic enclosure than to censorship in the old, honest, jackbooted sense. And the fights between these enclosures won’t look like wars. They’ll look like recommendation feeds, slowly bending in opposite directions, until people inside them stop being able to hear each other and start being very confident the other side is insane.

The mistake the horse is making is treating any of this as a speed contest. The industrial revolution didn’t reward the strongest horse. It rewarded the people who noticed the horse was no longer the unit of work. The unit is the model now. Whoever notices first gets to write the next century’s accountant job description — and probably the next century’s borders too. Everyone else gets a faster mane.