Google CEO: 75% of All New Code at Google Is Now AI-Written

Google CEO Sundar Pichai announced at Cloud Next 2026 that 75% of all new code written at Google is now AI-generated and approved by engineers — up from 50% last fall and roughly 30% a year ago. The figure, disclosed in an official Google blog post tied to the conference, represents one of the most concrete public disclosures by a major technology company about the scale at which AI has displaced human developers as the primary authors of new software.

The shift has happened faster than most inside the industry predicted. Just twelve months ago, Pichai noted that AI was writing "well over 30%" of new code at Google. By late 2025 that had climbed to "nearly half," and now three-quarters of every line added to the company's codebase originates from an AI model rather than a human developer. All AI-generated code still requires engineer approval before deployment, meaning humans retain the final gate — but their role has fundamentally changed from primary author to critical reviewer.

Pichai framed the announcement as a productivity multiplier rather than a sign of impending layoffs, emphasizing that expanded technical output creates more possibilities rather than fewer jobs. "The important metric is engineering velocity," he said. Google has publicly stated it plans to hire more engineers, not fewer, arguing that AI assistance amplifies the value of skilled talent. The path to this point was not without internal friction: reports surfaced of a temporary ban on using Gemini for coding inside Google, later reversed after co-founder Sergey Brin escalated concerns directly. The episode underscored how fraught the organizational transition has been even at the company building the tools.

The announcement comes against a backdrop where AI-assisted coding has become standard practice across the software industry. GitHub Copilot, Anthropic's Claude Code, and Google's Gemini Code Assist are now embedded in the daily workflows of millions of developers worldwide. What makes Google's disclosure notable is its scale and specificity: a company with over 180,000 employees, running some of the world's most complex software infrastructure, now relies on AI to produce the majority of its new code. Pichai cited a specific productivity win: agent-assisted workflows completed a complex internal code migration six times faster than a comparable project a year earlier.

One key metric Google has not disclosed is the acceptance rate for AI-generated code — the percentage of AI-authored suggestions that pass engineer review without modification. That figure would be the most direct measure of AI reliability as a production-grade collaborator. Without it, the 75% headline, while dramatic, leaves open the question of how much rework AI-generated code still demands before it ships. As the industry races to claim similar milestones, the quality of what AI produces — not just the quantity — will ultimately determine how far this transition goes.