2026-06-28 · Architecture
The patch found the bug
Build a patch for a small pain, and sometimes it hands you a bug too big to have noticed. The bug here is ten thousand years old.
Agriculture was a mistake made the case from health — the weak version; any archaeologist will trade you cities and surplus for it. The stronger version is about property. Farming didn't win because it fed people better — per calorie of labor it lost to foraging for a long time. It won because a crop and a herd could be fenced, owned, and defended, and a deer in the forest could not. The flip wasn't more food. It was the birth of the thing that can be held. The vector turned from seeking to keeping.
That bug is in production today, at the top of the market. Alphabet: ~75% of revenue is advertising; Google Search alone is 56%. Meta: 98%. Google is a hunter by nature — it crawls the whole world outward, pure seeking — and monetizes by keeping: a fenced auction, a moat around the catch. The act of search itself, privatized and turned into a fence. "Data is the new oil" was never about value. It's the agrarian move on steroids: don't seek the new, hold the caught, and milk it.
The power sits on one thing — a monopoly on connection. Tim Wu's Cycle: every information industry is born open and dies a monopoly. Illich named it sharper — radical monopoly, when the need itself is reshaped so it can't be met except through the middleman's fence.
Decentralizing connection has been promised before, and it kept failing. "This time it's different" is the cheapest sentence in tech, so here's the difference. Every prior technology zeroed out distribution. The internet zeroed the cost of moving a bit — and firms got bigger, because distribution fell and judgment didn't. Someone still had to decide what's good, who to trust, what to buy. That judgment held managers at the top and middlemen in the center. AI zeroes out judgment. Not "the firm dissolves" — that's false; the labs themselves are concentrating like never before. The honest shape is an hourglass: the model layer monopolizes, the application layer scatters down. And under the neck, the hunters come back. Small bands are efficient again — not nostalgia, Coase: a firm grows only to the size where coordinating inside it is cheaper than the market. AI just handed that one advantage to everyone. The big aren't stupid. They're slow.
A liquid database doesn't fence a schema and pour the world in — it chases the world. Memory is a place, not a log. The thing eats GitHub projects in five minutes — absorbs someone else's pipelines and data and moves on. It can't become a fence: open, forkable, and a market on top of it can't be enclosed. Nothing to monopolize, by construction. Point yours outward — stand up a local model, the open-source gap to the frontier is down to months, and absorb what you need. The core is open, MIT.
The singularity isn't the machine learning to devour better. It's the moment the vector turns back outward — from keeping to seeking. Hunters again. Only this time the prey is the new.