At Google Cloud Next 2026 this week in Las Vegas, Google CEO Sundar Pichai revealed a number that stopped a lot of people in their tracks. About 75 percent of all new code written at Google is now generated by AI and reviewed by human engineers. Just a year ago, that figure was around 25 percent.
That is a threefold increase in twelve months. And it is happening at one of the most sophisticated software engineering organisations on the planet.
Here is what this data point actually means — for Google, for the software industry, and for anyone thinking about the future of programming.
What Pichai Actually Said
Pichai shared the statistic during his keynote address at Google Cloud Next 2026, where he was laying out Google’s vision for the “agentic cloud” — a future where AI agents handle complex, multi-step work inside businesses.
The 75 percent figure refers specifically to new code being written at Google. It does not mean AI is maintaining existing codebases, debugging production systems, or making architectural decisions independently. Human engineers are still reviewing and approving all AI-generated code before it ships.
That human review step is important context. The number describes AI as a highly productive coding assistant, not an autonomous software engineer operating without oversight.
How Did Google Get Here So Fast?
A year ago, the same number was closer to 25 percent. That kind of growth — from one-in-four lines to three-in-four lines — in a single year is almost unprecedented in how a company adopts a new tool or workflow.
The likely explanation is a combination of factors. Google has been integrating its own Gemini models deeply into its internal developer tools, including code completion, test generation, and code review assistance. As models have improved and developer confidence has grown, the share of AI-generated code has risen steadily.
Google is also one of the few companies in the world that can train and deploy frontier AI models on its own custom hardware — the TPU v8 chips it announced at the same event — giving it an infrastructure advantage in scaling these tools internally.
Human Engineers Are Still in the Loop
It is worth being clear about what “AI-generated code” means in practice. Google engineers are not sitting idle while Gemini writes entire applications from scratch.
In most AI-assisted coding workflows, a developer describes what they want, the AI generates a draft, and the engineer reviews, edits, and approves the output. The AI dramatically speeds up the writing step, but the human remains responsible for correctness, architecture, and quality.
This is consistent with how Snap recently restructured its workforce, citing AI as a reason for reducing headcount in roles where automation was handling more of the output. The pattern is emerging across the industry: AI is not replacing engineers wholesale, but it is changing what engineers spend their time on.
What This Means for the Software Industry
Google’s disclosure is significant because it offers a concrete, public benchmark for AI coding adoption at scale. Other major tech companies are almost certainly at similar or higher levels of AI code generation internally — they simply have not published the numbers.
For the broader software industry, this points toward a clear direction of travel. AI coding assistants are not a novelty tool anymore. They are becoming the default way software gets written at the highest levels of the industry.
For developers, this does not mean jobs are disappearing — it means the job is changing. The engineers who will thrive are the ones who can direct, review, and improve AI-generated code, not just write from scratch.
For consumers, it means faster software development cycles and potentially more frequent product updates across the tools and apps they use every day.
Gemini Is the Engine Behind This
The AI doing most of this work at Google is Gemini, which is also the AI powering the broader suite of products Google announced at Cloud Next this week — including Auto Browse for Chrome, the Gemini Enterprise Agent Platform, and the new Workspace Intelligence layer.
Google has been systematically integrating Gemini across everything it builds, from consumer apps to internal developer tooling. Earlier this year, Gemini Nano 4 was announced for Android devices, and Google has been building a native Gemini app for Mac as well.
The 75 percent number is both a proof point and a signal. It tells the world that Gemini works well enough for Google to trust it with three-quarters of its own codebase — which is about as strong an internal endorsement as any AI model can receive.
If you are trying to figure out where Gemini sits relative to other AI tools, our Google Gemini vs ChatGPT vs Copilot comparison is worth a read.
Frequently Asked Questions
What did Sundar Pichai say about AI writing code at Google?
Pichai confirmed at Google Cloud Next 2026 that around 75 percent of all new code at Google is now generated by AI and reviewed by human engineers. He noted this was up from roughly 25 percent just a year earlier.
Does AI writing code mean human programmers are being replaced?
Not exactly. At Google, all AI-generated code is reviewed and approved by human engineers before shipping. AI is accelerating the coding process, but humans remain responsible for decisions around quality, architecture, and correctness.
Which AI is Google using to write its code?
Google uses its own Gemini models as the backbone of its internal AI developer tools, including code generation, code completion, and test generation.
Is 75% AI-generated code normal across the tech industry?
Google has not disclosed whether this is typical, but given how aggressively the entire industry has adopted AI coding tools, similar numbers at other major tech companies would not be surprising. Google is simply one of the few to share a specific figure publicly.
What does this mean for people learning to code?
It reinforces the importance of understanding AI tools and code review skills, rather than focusing purely on manual code writing. The ability to guide, evaluate, and improve AI-generated code is becoming as important as the ability to write code from scratch.
Conclusion
Google’s 75 percent AI code generation figure is one of the most concrete data points we have had about how deeply AI is reshaping software development at the frontier. It does not signal the end of human programming — it signals the beginning of a new kind of programming, where the best engineers are the ones who can work most effectively with AI.
The trend is only going to accelerate from here.
Leave a Reply