Google 75% AI-generated code claim
- Google has not said 75% of its code is AI-generated. The public figure is 30% of new code at Google, with human review still required. - The hiring shift people are noticing is narrower and more concrete: Google has discussed tougher anti-cheating interview formats as candidates use AI off-camera. - That matters because AI is changing software work fastest in drafting and iteration, but verification, debugging, and judgment still look stubbornly human.
The viral claim is overstated. Google has not publicly said AI now generates 75% of its internal code. The number Google itself has put on the record is 30% of new code written at Google, and even that comes with an important qualifier — engineers still review and approve it. So the real story is not “Google handed coding to bots.” It’s that big tech is normalizing AI-assisted development fast, while hiring is scrambling to figure out what skill it should test now. ### Where did the 75% number come from? Turns out the cleanest public Google number is lower. In August 2025, Google said AI helps generate 30% of new code written at the company. Earlier, Sundar Pichai had told investors that more than 25% of new code was written by AI. Those are meaningful numbers, but they are not 75%, and they are about *new code*, not all code across Google’s internal repositories. ### So what is actually changing? The workflow is changing first. Google says AI now helps not just with code generation, but with reviews, testing, migrations, and bug handling. In the same company write-up, Google tied that broader assistance to an estimated 10% increase in engineering velocity and said 12% of duplicate bugs are automatically handled by an AI agent. Basically, the machine is showing up across the software pipeline, not just in autocomplete. ### Why does the hiring angle keep coming up? Because interview formats were built for a pre-ChatGPT world. CNBC reported in March 2025 that Google was weighing changes to coding interviews as candidates increasingly used AI tools off camera during remote screens. Pichai even raised the possibility of returning to more in-person interviews. That is a very different claim from “Google no longer trust a raw remote coding exercise to measure individual skill cleanly. ### Does that mean debugging matters more now? Probably yes — but that part is still more inference than policy memo. Once AI can draft a decent first pass, the bottleneck shifts to checking whether the code is correct, safe, maintainable, and actually solves the problem. Google’s own language hints at this. Its engineers stay “in the driver’s seat,” and AI-written code still needs review and approval. ### Is Google alone here? Not even close. Microsoft said in April 2025 that 20% to 30% of some code in its repositories was probably written by AI. Meta’s Mark Zuckerberg said he expects maybe half of development on some internal AI work could be done by AI within a year. The industry trend is obvious: leaders are racing to raise the share of machine-generated code, even if the numbers vary and the definitions are fuzzy. ### What’s the catch? Trust. Google’s 2025 DORA research found AI use in software development is now widespread, with 90% adoption among surveyed professionals and more than 80% saying productivity improved. But trust lagged badly — only 24% reported high trust, while 30% said they trusted AI “a little” or “not at all.” That gap is the whole story in one statistic. People use AI constantly, but they still don’t want to bet the release on it uninspected. ### So what should readers take away? The flashy number is shaky. The underlying shift is real. Google’s public position is that AI drafts a growing slice of code, humans still validate it, and hiring has to adapt to a world where candidates can lean on models during interviews. If that keeps spreading, the premium skill won’t just be writing code fast from scratch. It’ll be knowing when the machine is wrong.
