75% of Google's New Code Is AI-Generated — What It Means for Web Developers

On April 22nd, standing in front of thousands of developers at Google Cloud Next 2026, CEO Sundar Pichai dropped a number that stopped the room: 75% of all new code written at Google is now generated by AI — and reviewed by engineers.

That figure has climbed fast. It was 25% in October 2024, 50% by fall 2025, and it has just hit three-quarters. If that trajectory continues, the question isn't whether AI will write most of the world's code — it's how quickly, and what role human developers will play when it does.

The Engine Behind the Number: Meet Antigravity

This isn't just autocomplete on steroids. Google's engineers are using Gemini models paired with an internal agentic development platform called Antigravity — launched alongside Gemini 3 in November 2025. Antigravity fundamentally changes how a developer interacts with a codebase.

Instead of writing functions line by line, a developer describes an objective in plain language. Antigravity's agents then plan the work, execute changes across multiple files, run the test suite, interpret failures, and iterate — all without the developer touching a keyboard for the actual code. Pichai noted that one complex code migration that would have taken weeks was completed six times faster using these agent-assisted workflows.

The human engineer's job in that loop? Review, approve, secure, and deploy. The craft has shifted from authorship to editorship.

This Isn't Just a Google Story

Google's 75% figure is the most dramatic data point, but it sits inside a broader industry trend that's been accelerating all year:

Microsoft disclosed in April 2025 that roughly 30% of its code was being written by AI — a figure that has certainly risen since.
Anthropic has reportedly crossed the threshold where nearly 100% of its own code is AI-assisted.
Industry surveys show 84% of professional developers have integrated AI into their daily workflows, with many reporting AI contributes to 90% of their codebase.

The pattern is consistent: AI adoption in coding isn't a fringe experiment anymore. It's the default workflow at the companies building the software that the rest of the world runs on.

What Google Cloud Next 2026 Signalled More Broadly

The 75% statistic landed in the middle of a conference packed with developer-relevant announcements. A few worth knowing about:

Agent2Agent (A2A) Protocol

Google announced a production-grade open protocol called Agent2Agent (A2A) that allows AI agents from different platforms to communicate and collaborate. If you've been building workflows that chain multiple AI tools together, A2A is designed to make that kind of cross-platform orchestration reliable and standardised. This could become as foundational to agentic applications as REST was to web APIs.

Gemini Enterprise Agent Platform

Vertex AI has evolved into the Gemini Enterprise Agent Platform — a comprehensive environment to build, scale, govern, and optimise agents. It includes a new Agent Designer, long-running agent support, a Skills library, and built-in DevOps tooling. For teams already on Google Cloud, this consolidates a lot of previously scattered tooling into one surface.

Gemma 4 and Open Models

Google also released Gemma 4, the latest in its family of open-weight models. For developers who need capable models they can run locally, fine-tune, or deploy without API costs, the Gemma series continues to be one of the most practically useful open-source offerings in the space.

Workspace Studio

Workspace Studio is a no-code platform that lets business users build and deploy AI agents across Gmail, Docs, Sheets, Drive, Meet, and Chat by describing automations in plain language. It's aimed at non-developers, but it signals how the line between "user" and "builder" is dissolving.

The Practical Takeaway for Web Developers

Here's the honest assessment: the role of the web developer is not disappearing, but it is changing faster than most professional transitions in history. A few things worth sitting with:

1. The value is shifting up the stack

If AI can reliably write the implementation given a clear spec, then the leverage point for human developers moves toward defining what to build — the architecture decisions, the product thinking, the edge cases that require real-world judgment. The engineers who thrive will be those who can operate at that higher level of abstraction.

2. Review skills matter more than ever

Google's 75% figure explicitly includes human review in the loop. The ability to read AI-generated code critically — spotting security holes, questioning architectural choices, identifying subtle bugs — is now a first-class professional skill. You can't review what you can't understand, so deep knowledge of the underlying systems still matters enormously.

3. Agentic tooling is worth learning now

Antigravity is internal to Google, but analogous tools are available publicly today: Claude Code, GitHub Copilot Workspace, Cursor, and others are moving toward the same agentic paradigm. Getting fluent in how to direct these tools effectively — how to write good objectives, how to break tasks into chunks agents can reliably execute, how to catch when they go off the rails — is becoming a core competency.

4. The open-source ecosystem is keeping up

It's not only Big Tech. OpenClaw, an open-source autonomous agent framework that runs locally on your machine and can execute shell commands, manage files, and automate web tasks, recently surpassed 302,000 GitHub stars — reportedly the fastest any open-source project has reached that milestone. The tools being built in the open are increasingly competitive with enterprise offerings.

A Moment Worth Marking

There's a version of this story that reads as anxiety-inducing, and a version that reads as exciting. We'd argue it's mostly the latter. The developers who embraced version control, then the web, then mobile, then cloud-native patterns didn't get replaced by those shifts — they became more valuable because they understood how to use new capabilities that others didn't yet.

The inflection point Pichai named this week is real. Three-quarters of new code at one of the world's largest engineering organisations is now AI-generated. That number is going to keep rising industry-wide. The developers who treat that as a capability to direct, rather than a threat to resist, are the ones who'll be doing the most interesting work in the years ahead.

We're watching this space closely — and building with it. If you're curious how agentic AI workflows are changing how web studios like ours operate day-to-day, we'll be writing more about that soon.

Sources: Google Blog — Sundar Pichai at Cloud Next 2026, The Decoder, The Next Web — Google Cloud Next 2026 roundup