75% of Google's new code is now written by AI.
Not assisted. Not suggested. Written by it. Engineers review and approve. That is the job now.
Sundar Pichai dropped this at Cloud Next like it was a productivity win. The headlines agreed. "AI augments engineers!" "Humans still in the loop!" Everyone nodded because the alternative is a conversation nobody wants to have.
So I will have it.
The number was 30% in April 2025. 50% by last fall. 75% today. The trajectory does not flatten. Snap hit 65% and simultaneously cut a quarter of planned headcount. Google is already tying AI adoption to performance reviews. Companies do not do that because they want more engineers.
The engineers surviving this shift are the ones who already knew what good architecture looked like before the model suggested it. Who could tell when generated code was technically correct but conceptually wrong. That judgment came from years of writing bad code first.
We are about to stop producing those engineers.
A junior hire in 2026 spends their time reviewing model output instead of building the intuition that makes a senior engineer irreplaceable. The craft is in the doing. You cannot shortcut it by being a good approver at 22 and expect real judgment at 30.
The question is not "will engineers be replaced?" It is: what kind of engineers will exist in ten years, and how do they get there from here?
Anyone entering this field deserves that question asked honestly. Not reassurance that the machines are just tools.
They are excellent tools. And excellent tools have always changed who gets to call themselves a craftsperson.