TL;DR
Anthropic recently published When AI Builds Itself, an essay explaining how AI is increasingly helping build the next generation of AI. Today, more than 80% of Anthropic's production code is written by Claude, and engineers are shipping around eight times more code than they were in 2024.
I went into the essay carrying the same quiet anxiety I think many developers have right now. I came out feeling less scared.
Not because the numbers are small. They aren't.
It's because after reading the entire essay, I realized the thing AI is getting better at is not the thing that makes developers valuable. Execution is getting faster and cheaper. Judgment, the ability to decide what's worth building, whether a result actually makes sense, and when to question an answer instead of accepting it, is not. That distinction gets lost in most of the conversations happening online, and once you see it, the whole essay reads differently.
