Google says 75% of its new code is AI written | Hacker News Reader

Google says 75% of its new code is AI written

blog.google

11 pointsDrakexor10 comments

"AI-generated and approved by engineers" is doing a lot of work there. If accepting a 4-character Gemini autocomplete counts, Copilot users hit >90% last year. The useful metric is % of functions where >50% of the body was AI-written before human edits. just my 2 cents

I saw this post at the exact same time I ran across this elsewhere: https://www.phonearena.com/news/pixel-6-7-8-9-10-battery-dra...

There's no way to definitively say they're cause and effect but also....

There is a big difference between just “vibe coding” without reviewing the code and actually knowing what you’re doing and checking everything the AI produces.

In my opinion, humans don’t really need to write code anymore. The main skill is steering AI models.

Title should have the word "its" rather than "it's"

Fixed, thank you

One possibility is that most of the code that matters is being written by hand, while enormous amounts of code are being generated for other things. People are being evaluated on their AI usage after all.

This is an ad:

    Our engineers are orchestrating fully autonomous digital task forces, firing off agents and accomplishing incredible things.

Who swallows this stuff? Like all marketing it's full of weasel works and unverifiable claims. I don't doubt 75% of their code started life as the output of a model mind you, it just "forgets" to mention the subsequent hours of humans put into reviewing, discarding the crap, and refining what was left.

Edit: it's slop. How did I not spot it was slop posted by the marketing department immediately? If they want engineers to pay any attention they will have to do better. Or maybe they are targeting engineers any more? blog.google used to be a reliable read.

  > Who swallows this stuff?

you'd be surprised, but i suspect its mostly for signaling to ceo/cio decision makers

  > 75% of their code started life as the output of a model mind you, it just "forgets" to mention the subsequent hours of humans put into reviewing

it its like at $work, they probably have some internal self-reporting benchmark from employees so i'd guess those numbers are inflated a bit...

  > unverifiable claims

it reminds me of the used car salesman saying "they're just like new"

  > blog.google used to be a reliable read

blog.google was always just a marketing channel, as is basically all corporate blogs afaik

Relevant section:

"Today, 75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall."

Big tech is inflating the AI bubble. Because of that, I find it hard to fully trust these kinds of claims — especially from complex systems like Google.

Reply on news.ycombinator.com