Google CEO Pichai says AI code share now ‘meaningless’

Google CEO Sundar Pichai said the question of how much of Google's code is written by AI no longer has a useful answer, because AI is now involved in nearly all of the company's engineering work. In a Big Technology Podcast interview with Alex Kantrowitz published October 20, he was asked where the figure stands now. He had once put it at 25%. "That number in our world today is meaningless in the sense that every part of what we now touch is influenced by AI deeply," he said.officechai.com

The comment marks a change for Google. The company has used that single percentage for about two years as its main evidence that its own AI tools work. Pichai said Google now relies on other measures, chiefly how engineers use Gemini 4 Argon, its newest flagship model.

New metrics replace the percentage

"People are using Argon to really go tackle complex projects, including projects which affect millions of lines of code," Pichai said in the interview. He said that in the four weeks since Argon launched, Google saw "a 30% increase in CLs," which is Google's term for code changes submitted. Over the same period, rollbacks (changes that had to be undone) fell by about a third. "People are writing code more productively and better code with it. We are tracking it with newer metrics now," he said.bigtechnology.com

Google's official figures are a little different. When Argon went on sale to developers and businesses, Google said teams using it in its Antigravity coding tool raised agentic code submissions by more than 35% and cut rollbacks by nearly a third, Reuters reported. The company's cloud blog repeated those numbers.reuters.com+1

A figure that kept climbing

The shift comes six months after Google's percentage reached its highest level. In an April 22 blog post during Cloud Next 2026, Pichai wrote that "75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall." Pichai first said in October 2024 that AI was writing more than a quarter of Google's new code.blog.google+1

These figures have long drawn questions about what they actually measure. An August 2026 study by Harvard economists, as summarized by Techora, found that coding agents raised new lines of code by 30% and review requests by 23%. Review time rose 49%, however, and the increase in completed tasks was not statistically conclusive.techora.ru

Pichai said the change in how software gets written is already settled. "Nobody would want to code the way they were coding three years ago, or even the same time last year," he told Kantrowitz.bigtechnology.com