75% of Google's New Code Is Now AI-Generated: Engineers Are Becoming Reviewers

From 25% to 75% in barely 18 months. That's how fast AI-generated code has scaled inside Google, and the number should make every software engineer pay attention.

Google's 75% AI-generated code milestone explained

Sundar Pichai dropped the figure during Google Cloud Next 2026 in Las Vegas. Three quarters of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall and roughly 25% in October 2024. The trajectory is steep, deliberate, and hard to ignore.

This is not boilerplate for a press release. Google writes some of the most complex production code on the planet, across Search, Ads, YouTube, Android, and Cloud. That AI is now the primary author at this scale signals a structural shift in how software gets built.

Agentic workflows and the Antigravity platform

Pichai emphasized the move toward truly agentic workflows. A complex code migration was completed six times faster with AI agents than engineers alone could have managed a year ago.

The team behind the Gemini macOS app used Google's new Antigravity development platform, going from concept to a native Swift prototype in days. Antigravity is more than an IDE: it deploys autonomous agents that plan, execute, and verify tasks across editor, terminal, and browser.

How Google compares to rivals in AI-generated code

Snap says 65% of its new code is AI-generated. Meta has set internal targets for select teams to produce over 75% of committed code with AI tools by mid-2026. Anthropic reportedly writes nearly 100% of its code with AI assistance.

Microsoft was at 20-30% as of April 2025. Google's leap from 25% to 75% in 18 months is the steepest climb in the industry. Some Google engineers can even use third-party tools like Claude Code alongside Gemini models, and AI adoption goals are factored into performance reviews.

The real question Google hasn't answered

Engineers are becoming reviewers and orchestrators rather than line-by-line authors. Pichai frames this as a productivity multiplier, not a replacement, and he's probably right for now.

But one number is conspicuously missing: the rejection rate. How much AI-generated code gets rewritten before it ships? That metric would tell us far more about AI's true coding maturity than the headline 75% ever could.