This is such a meaningless number. Keyboards have written 100% of Google's code for over 2 decades... The real question is whether AI writes the code autonomously without human engineers going through multiple iterations to correct it. And when you look at it this way, AI most likely "writes" 0% of the code. I code pretty much everyday with AI myself. A conservative estimate is that about 90% of the code they output has critical errors in it that makes it absolutely unusable in a production environment. I still use the AIs because I’m experienced enough to easily spot the mistakes and correct them; and on balance I’m more productive coding with an AI despite these limitations. But the notion that these AIs will be able in the near future to totally replace developers is utterly risible to me, and should be to anyone with enough technical knowledge to distinguish between code that compiles and code that actually works reliably in complex, real-world environments. Especially in such an environment as Google's products.
See Tweet