Google just told us that 75% of its new code is AI-written. The other 25% is the job description you actually have now.
Sundar Pichai dropped that number at Google Cloud Next this week, almost casually, the way you mention a weather forecast. Seventy-five percent. Up from fifty percent just six months ago. At Google — one of the most sophisticated engineering organizations on the planet, with tens of thousands of engineers, decades of internal tooling, and more compute than most countries — three out of every four lines of code committed today were written by a machine.
And then Pichai said the quiet part loud: every single line is reviewed by a human engineer.
That one sentence changes everything about how you should think about your career.
The Number Isn't the Story. The Rate Is.
A lot of coverage is treating 75% as a milestone. It's not a milestone. It's a speed reading.
Six months ago it was 50%. Now it's 75%. If that pace holds — and there's no reason to think it won't — you're not looking at a stable new normal. You're looking at a trajectory. What does 90% look like? What does 95% look like? At some point the math gets uncomfortable: if AI is generating 95% of the code and engineers are reviewing all of it, what exactly are engineers contributing beyond their attention?
That's not a rhetorical taunt. It's the actual question that engineering leadership at every serious company is going to spend the next two years figuring out. You should be thinking about it now, before someone else answers it for you.
What Pichai Actually Said About Engineers
Here's what the CEO of Google did not say: engineers are obsolete, we're laying everyone off, the machines won. He said the opposite.
His framing was precise: AI generates drafts, humans provide judgment. The code migration example he used is worth sitting with. A complex migration — the kind of unglamorous, multi-system, touch-everything-carefully work that normally takes a team several months — was completed six times faster using AI agents working alongside engineers. Not instead of. Alongside.
Six times faster is extraordinary. But the engineers didn't go anywhere. They shifted from writing the migration to architecting it, supervising it, and catching the places where the AI confidently did exactly the wrong thing.
That's the new job. It's harder than it sounds.
The Skills That Are Going Stale
Let's be honest about what's depreciating. The abilities that made you a strong developer five years ago:
- Typing fast and writing syntax fluently
- Memorizing API signatures and library quirks
- Knowing which Stack Overflow answer to reach for
- Being good at autocompleting your own thoughts into code
These aren't worthless. But they're becoming table stakes. The bar to generate functional code has dropped to near-zero for anyone with a decent AI tool. If your value proposition as an engineer is that you can write correct Python faster than average, that advantage is shrinking every quarter.
The Skills That Are Appreciating
What Pichai's announcement actually describes is a market correction in engineering skills. The value is migrating upward.
Judgment. This is the big one. AI-generated code looks fine until it doesn't. The bugs it produces are often subtle — not syntax errors, but logic that passes review, passes tests, and then fails at 2am in a way that takes four engineers a day to untangle. The ability to look at AI output and know something's off before you can articulate why — that's experience you can't shortcut, and it's becoming the most valuable thing an engineer can bring.
Problem decomposition. AI executes on what you describe. If you describe it poorly, you get plausible-looking garbage. The engineers who get the most out of these tools are the ones who can break down a complex problem into pieces clear enough that an agent can execute each one correctly. That's a writing skill as much as a coding skill. It's a thinking skill.
Systems architecture. Someone has to design what gets built before anything gets generated. Understanding how services talk to each other, where the failure modes live, what the data model needs to look like three years from now — AI is getting better at this, but we're not there yet. The engineers who own the whiteboard still own the product.
What You Should Actually Do About This
This isn't a "learn prompt engineering" listicle. It's a more fundamental reorientation.
Start measuring your value in outcomes, not output. If your self-assessment is built around how much code you ship, recalibrate. The metric that survives this shift is impact — did the system get better, did the product move forward, did the team make the right call?
Get uncomfortable with review. Reading AI-generated code critically — not just skimming it for obvious errors, but actually interrogating it — is a skill that takes practice. Start treating every AI-generated PR like it was written by a smart intern who sometimes hallucinates.
Invest in the parts AI can't do (yet). Customer context. Organizational history. The half-decision that got made in a Slack thread eighteen months ago that explains why the auth flow looks like that. This embedded knowledge is your moat while the tools catch up.
Own an opinion about architecture. The engineers who are going to matter most in the next five years aren't the ones who can code the fastest — they're the ones who can look at a system and tell you what's wrong with it before anything is built.
The Identity Part Nobody Wants to Talk About
There's a version of this conversation that's just practical — skills to acquire, metrics to shift, career pivots to consider. That's all real and worth doing.
But underneath it is something messier. A lot of developers built their sense of professional identity around being good at writing code. Not just competent. Good. Proud of their craft, proud of clean implementations, proud of the way they could sit down with a hard problem and wrestle something elegant out of it.
That identity is getting disrupted. It's worth naming.
The craft isn't gone. But the expression of it is changing shape. The engineer who thrives in the next decade probably looks less like a virtuoso coder and more like a very technical product thinker — someone who can hold a complex system in their head, direct machines to build pieces of it, and catch the places where the machine's confidence is unearned.
That's not a lesser version of engineering. But it's a different one. And the sooner you're honest with yourself about which version you've been training for, the easier the pivot gets.
Google is at 75%. The number is going up. The job description that matters now is the 25% — and if you're not actively figuring out what lives there, someone else will answer that question for you.
Sources: Tech4Gamers · Analytics Drift · WION · HumanReadable AI