5-Min Brief: Three Quarters of Google's New Code Is Now Written by AI. Let That Sink In.
What you need to know — in 30 seconds
- Google CEO Sundar Pichai revealed today that 75% of all new code written at Google is now generated by AI — up from 50% just last fall and 25% eighteen months ago
- Human engineers still review and approve everything, but AI is doing most of the actual writing
- A complex code migration that would have taken a year was recently completed six times faster using AI agents working alongside engineers
- Every major tech company is racing to hit similar numbers — and the implications for software jobs are real
Three quarters. That's the number. Of every four lines of new code written at Google right now, three of them came from an AI. A human looked at it and said yes or no — but AI wrote the first draft.
That number was 25% just eighteen months ago. It jumped to 50% last fall. Today it's 75%. The trajectory is not subtle.
What's actually happening at Google
Google CEO Sundar Pichai shared the figure today at Google Cloud Next, the company's annual conference for business customers. He framed it not as a warning but as a success story — proof that the company's internal AI tools are working.
The setup is this: Google has built AI coding assistants powered by its own Gemini models directly into the tools its engineers use every day. When an engineer starts writing code, AI is suggesting, completing, and often generating entire blocks of it. The engineer's job is increasingly to describe what they want, review what the AI produces, and ship it.
Pichai gave a concrete example of what this looks like in practice. A particularly complex code migration — the kind of unglamorous but essential work that normally takes a team months — was recently completed six times faster than it would have been a year ago, using AI agents and engineers working together.
Six times faster. For context: that's the difference between a six-month project and a one-month project. At Google's scale, that kind of efficiency compounds into something enormous.
This isn't just a Google story
Here's what makes today's number significant beyond Google itself: everyone else is watching and racing to catch up.
Meta has set internal targets for its engineers to have AI generate 75% of their committed code by mid-2026. Snap announced earlier this month that at least 65% of its new code is AI-generated — and simultaneously laid off roughly a quarter of its planned headcount, citing AI efficiencies.
Microsoft's CEO said two years ago that 20-30% of the code for some projects was AI-written. Its CTO predicted at the time that 95% of code would be AI-generated within five years.
The industry is moving faster than that prediction suggested.
What about the engineers?
This is the question everyone is actually asking.
The honest answer is: it's complicated, and anyone who gives you a simple one is oversimplifying.
What's clearly happening is that the nature of software engineering is changing faster than at any point in the past decade. The skills that mattered most in 2020 — fluency in programming languages, speed at writing syntax, memorizing APIs — are becoming less valuable. AI is better at those things now.
What's becoming more valuable is judgment: the ability to look at AI-generated code and recognize subtle errors that look correct until they cause problems at 3am. The ability to decompose a complex problem clearly enough that AI can execute on it well. The ability to architect systems that hold up in the real world.
The Stanford AI Index we covered last week noted that employment for software developers aged 22-25 has already fallen nearly 20% since 2022. That's the early signal. Junior roles — the entry point to a software career — are the most exposed right now.
None of this means software engineering is disappearing. Google still has tens of thousands of engineers. But the work is changing, and the ratio of engineers to output is shifting in ways that will matter.
The number that will keep moving
75% is not where this ends. Pichai's framing today was about "truly agentic workflows" — where AI doesn't just assist engineers but runs entire development tasks autonomously, with engineers orchestrating rather than writing.
The direction is clear. The pace is faster than almost anyone predicted. And the question for every company now isn't whether to adopt AI coding tools — it's how fast they can do it before their competitors do.
HumanReadable-AI covers AI news in plain English every weekday. Subscribe below — free, no jargon, always under five minutes.