Google announced this week that three-quarters of its new code is now AI-generated, up from 25% in October 2024. The number is designed to impress. It does. Just not in the way Google's PR team hopes.

The headline suggests armies of autonomous coding agents building the next generation of search and Android. The reality is almost certainly autocomplete on steroids. When Sundar Pichai says "AI-generated," he means a developer types a function signature and Gemini fills in the boilerplate. That still counts. Technically.

I've talked to enough engineers at big tech companies to know how this works. You write a comment describing what you want. The AI suggests an implementation. You review it, fix the inevitable bugs, adjust for your company's specific architecture, and commit. The AI wrote the first draft. You wrote the code that actually ships. Both get counted as "AI-generated" in the internal metrics.

This is not autonomous programming. It's very good autocomplete. And yes, it saves time. GitHub Copilot has been doing this for years. What's changed is that Google has turned it into a KPI.

The technology is impressive. Autocomplete that understands context across millions of lines of code is a genuine engineering achievement. But calling it "AI-generated code" obscures what's actually happening. The AI isn't designing systems. It isn't debugging production failures at 3 AM. It isn't making architectural decisions about whether to use microservices or a monolith.

It's filling in the boring parts. And that's valuable. Writing unit tests, converting data formats, implementing standard CRUD operations - this is real work that previously required human time. Letting AI handle it means developers can focus on problems that actually require human judgment.

The question is whether Google's 75% metric means they're shipping better code faster, or just that they've gotten very good at measuring AI involvement. Based on similar announcements from other companies, I'd bet on the latter.

What bothers me most is the framing. AI-generated code sounds like we're one step away from obsolete programmers. What Google is actually describing is a productivity tool. A very good one. But still a tool.

The technology is impressive. The question is whether anyone needs the hype.