The percentage of code written by AI at Google has gone from 25% to 75% in exactly a year.
Google CEO Sundar Pichai disclosed the latest figure in a blog post. “We’ve been using AI to generate code internally at Google for a while. Today, 75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall,” he wrote. The trajectory has been steep: in April 2025, Pichai had put the number at “well over 30%.” By late 2025, Google’s CFO was citing “nearly half.” Now it’s three-quarters.
This is not a vanity metric for Google’s PR machine. The company writes some of the most complex production code in the world, across Search, Ads, Cloud, YouTube, and Android. The fact that AI is now the primary author of new code at this scale — and that Google’s engineers are signing off on it — is a meaningful signal about where software development is heading.
Pichai has been careful to frame AI coding not as a replacement for engineers but as a productivity multiplier. On the Lex Fridman podcast, he said the more important metric isn’t the share of AI-written code but the increase in engineering velocity. He’s also said Google plans to hire more engineers, not fewer — the argument being that higher productivity expands what’s possible, which increases demand for talent.
That said, the cultural shift inside Google has not been frictionless. Co-founder Sergey Brin, who returned to an active role at the company as AI took off, had a public disagreement with internal teams over a policy that had quietly banned the use of Gemini for internal coding — Google’s own flagship model, prohibited from the very use case it was built for. The policy was eventually reversed after Brin escalated to Pichai.
Google is not alone in this shift. Microsoft’s Satya Nadella disclosed in April 2025 that roughly 30% of Microsoft’s code was being written by AI — a figure that is likely higher now. But Google’s 75% figure shows how quickly AI has disrupted coding. Over at Anthropic, nearly 100% of the code is now written using AI.
The one number worth watching alongside the 75% is the approval rate — how much of that AI-generated code actually passes engineer review. Pichai’s framing, that the code is “approved by engineers,” implies the bar is being cleared, but Google has not disclosed how much AI-generated code gets rejected or rewritten before it ships. That figure would tell us more about the true state of AI as a coding collaborator than the headline percentage alone.