Google Says AI Writes 75% of Its Code. It's Also Hiring More Engineers

Google CEO Sundar Pichai dropped a number in April that made every software engineer sit up straight: 75% of new code at Google is now AI-generated.

The progression has been swift. In 2024, that figure was 25%. By fall 2025, it hit 50%. Now, three quarters of all new code at the search giant comes from AI, with engineers reviewing and approving it.

Here's what makes this interesting: Google isn't cutting engineering jobs. They're hiring more engineers.

The Interview Revolution

Google just started piloting something that would have been unthinkable two years ago. Engineering candidates can now use AI assistants during coding interviews.

According to internal documents obtained by Business Insider, the pilot begins with junior to mid-level roles. Candidates get access to an approved AI assistant during the technical round. No more sweating through algorithm problems on a whiteboard. The AI is right there, part of the process.

This isn't Google admitting defeat. It's Google acknowledging reality. If 75% of their production code involves AI, why test engineers on skills they won't use?

The Numbers Tell a Story

Pichai's 75% figure comes with an important caveat: all AI-generated code still gets "approved by engineers." This isn't about AI replacing humans. It's about AI changing what humans do.

Think about what this means mathematically. If AI generates three quarters of new code, but Google keeps hiring engineers, those engineers aren't writing less code total. They're shipping more.

The progression tells us something else. Google went from 25% to 75% AI-generated code in under two years. Meta is targeting similar levels by mid-2026. This isn't a Google phenomenon. It's an industry shift.

What Engineers Actually Do Now

The traditional image of a software engineer hunched over a keyboard, typing line after line, is becoming outdated. Today's Google engineer might spend more time:

• Reviewing AI-generated code for security and performance
• Designing system architecture that AI can't handle
• Writing the prompts that generate the right code
• Debugging complex interactions between AI-written modules
• Building the 25% of code that requires human creativity

One Google engineer, speaking on background, put it simply: "I write less boilerplate and more of the hard stuff."

The Hiring Paradox

If AI writes most of the code, why hire more engineers? The answer lies in what AI can't do.

AI excels at pattern matching. Give it a common problem, and it'll generate solid code. But software engineering at Google's scale isn't about common problems. It's about edge cases, system design, and understanding user needs.

Google's new interview format reflects this reality. By letting candidates use AI during interviews, they're testing what actually matters: can you work with AI to solve problems? Can you spot when the AI is wrong? Can you architect solutions that leverage AI effectively?

The Competitive Reality

Google isn't alone in this shift. Every major tech company is racing toward similar numbers. Microsoft, Meta, Amazon, all are pushing AI-assisted development. The companies that don't adapt risk falling behind.

But there's a twist. As AI handles more routine coding, the value of exceptional engineers goes up, not down. The engineer who can design elegant systems, who can spot subtle bugs, who can optimize for scale, becomes more valuable when freed from writing boilerplate.

This explains the hiring paradox. Companies need fewer mediocre engineers writing routine code. They need more excellent engineers doing the work AI can't.

What This Means for the Industry

The 75% number is a watershed moment. It's not just about Google. It's about the future of software development.

Junior engineers entering the field today will have a fundamentally different career than those who started five years ago. They'll need to be AI-native from day one. The ability to work with AI isn't a nice-to-have. It's table stakes.

Google's interview change signals this shift. They're not testing whether you can implement quicksort from memory. They're testing whether you can build production systems with AI as your copilot.

The irony is perfect. As AI writes more code, human judgment becomes more critical. As algorithms handle the routine, human creativity becomes more valuable. Google's 75% isn't the story of human replacement. It's the story of human elevation.