Google says 75 percent of its new code is now AI-generated, and that changes the engineering story

AI Coding

Sundar Pichai says three-quarters of new code at Google is now AI-generated and approved by engineers, a sign that software teams are moving from assistive autocomplete toward supervised agentic workflows.

# Google says 75 percent of its new code is now AI-generated, and that changes the engineering story

## Opening summary

Google is no longer talking about AI coding as a sidecar. In a Cloud Next post this week, CEO Sundar Pichai said 75 percent of all new code at Google is now AI-generated and approved by engineers, up from 50 percent last fall. That is a striking adoption curve, but the more important signal is what it says about software development inside large companies: AI is moving from suggestion tooling toward supervised workflow infrastructure.

## Main article

Pichai made the claim in a broader April 22 Cloud Next post about Google’s agentic push across enterprise products, infrastructure, and internal operations. The line about code stands out because it gives a rare quantitative look at how deeply a major platform company says AI has entered day-to-day engineering work. According to Google, the jump from 50 percent to 75 percent happened in roughly half a year.

That number deserves careful reading. Google says the code is AI-generated and approved by engineers, which means the company is not describing a hands-off software pipeline. Humans are still reviewing and accepting the output. Even so, the threshold is notable. Once most new code arrives through model-assisted generation, the bottleneck shifts from typing to supervision, architecture, validation, and integration.

Google also paired the metric with a more ambitious workflow story. Pichai said engineers are now orchestrating autonomous digital task forces and cited a complex migration completed six times faster than would have been possible a year ago with engineers alone. That suggests Google wants the market to understand AI coding not as a better autocomplete feature, but as part of a broader move toward multi-step agentic engineering.

The Verge highlighted the 75 percent figure quickly, underscoring how unusual it is for a company of Google’s size to attach a hard number to internal AI coding adoption. Still, the strongest evidence in-run remains Google’s own framing, so the safest interpretation is strategic rather than universal. Google is showing where it believes elite software teams are headed, not proving that every engineering organization can reproduce the result immediately.

That distinction matters for buyers and builders. Large companies with mature review culture, internal tooling, and platform governance can often absorb model output more safely than smaller teams can. The operational lesson is not simply that AI writes code. It is that engineering systems are being reorganized around faster generation plus tighter approval loops.

## Why it matters

If Google’s internal numbers are directionally right, the next contest in AI coding will be less about demo quality and more about workflow design. The winning products will be the ones that help teams review, steer, and trust machine-generated work at production speed.

## Source notes

- Verified against Sundar Pichai’s April 22 Cloud Next post on Google’s blog - Secondary confirmation came from The Verge’s AI coverage roundup on April 22 - Quantitative adoption and speed claims should remain attributed to Google