News · 2026-06-23
The AI That Now Writes Most of Its Maker's Code
Anthropic, the company behind the Claude assistant, just published an unusually candid look inside its own engineering and the headline number is hard to ignore: as of May 2026, more than four out of every five lines of code the company ships are now written by Claude itself, not by its human engineers. You can read the company's full essay, called When AI builds itself, and the coverage it drew from Tom's Hardware and VentureBeat.
A little background helps. Two years ago this share was in the low single digits. The shift came after Anthropic released Claude Code, a tool that lets the model read a whole codebase, make changes, run tests, and fix what breaks, all on its own. The human role quietly flipped. Engineers used to be the authors and the machine was the helper. Now the machine is the author and the engineers are the editors who approve, reject, and steer. Anthropic reports its typical engineer now ships roughly eight times as much code in a quarter as a few years ago, not because people type faster, but because they spend their day reviewing the model's work instead of writing it.
The simplest way to picture this is a newsroom where a tireless junior writer drafts every article and the senior editors only sign off. The volume goes way up. But here is the catch that makes the eighty-percent figure less impressive than it sounds: a draft that a human has to check, fix, and approve is not the same as a writer you can leave alone. Most of those lines still pass through a person. So on its own, this number measures effort the machine saves, not work it can be trusted to do unsupervised.
The results buried deeper in the essay are the ones worth your attention, because they are about taste rather than volume. Anthropic ran a recurring test where the model is asked to choose the best next step in a research project, then compared its choices against its own scientists. Late last year the model was basically a coin flip against the humans. By spring 2026, an unreleased internal model was picking the better direction clearly more often than its own researchers did. Choosing what to work on next was supposed to be the part that stayed human longest. That is the part that moved.
There was an even sharper demonstration. Anthropic handed its own agents an unsolved problem in AI safety and let them work it start to finish with no human in the loop. An earlier version closed only a small slice of the gap to human experts. The spring model closed almost all of it. Anthropic is careful to frame this not as a stunt but as evidence that the missing ingredient, which it calls judgment, is filling in.
Why does this matter beyond one company's bragging rights? Because it is the clearest first-party signal yet that the labs at the frontier believe a feedback loop is forming, the one where AI helps build better AI, which then helps build better AI again. The company even tracks how long a task an AI can handle before a human has to step in. A couple of years ago that was a few minutes of work. By early this year it had stretched to a full workday. Independent researchers have measured the same trend climbing on a steady curve, in a widely cited study on how long the tasks AI can finish keep getting longer. If that line keeps bending the way it has, the gap between an assistant and a colleague keeps shrinking.
Here is the honest caveat, and it is a big one. Almost every dramatic figure in the essay comes from an unreleased internal model that no outsider can test. A company telling you, with its own measurements, that its own product is becoming powerful enough to be concerning is exactly the kind of claim that deserves outside verification before anyone treats it as settled fact. It can be sincere and self-serving at the same time. Anthropic itself adds the line skeptics will want to remember: it says plainly that this is not full self-improvement yet, and that such a future is not inevitable. The volume number is real and checkable. The judgment numbers are the interesting ones, and they are still taking the company's word for it. For the longer arc this fits into, see our earlier story on the model that could rewrite itself but held back, and our primer on what recursive self-improvement actually means.