Google has confirmed a monumental shift in its engineering culture, revealing on Wednesday, April 22, 2026, that 75% of all new code at the company is now generated by AI and subsequently reviewed by human engineers.
This milestone was announced by CEO Sundar Pichai during Google Cloud Next 2026 in Las Vegas, signaling the start of what Google calls the “Agentic Era” of software development.
1. The Rapid Acceleration of AI Coding
The jump to 75% represents a staggering increase in internal AI adoption over the last two years.
- April 2026: 75% AI-generated.
- Late 2025: 50% AI-generated.
- October 2024: 25% AI-generated.
- 2023: ~15% AI-generated.
2. From “Copilots” to “Agentic Workflows”
Pichai emphasized that Google is moving beyond simple code completion (where AI suggests the next line) to Agentic Workflows. In this new model, engineers act as “orchestrators” for autonomous digital task forces.
- Project Antigravity: This internal “agent-first” platform allows AI agents to plan, write, and test applications with minimal human intervention. Google used this to build the native Swift prototype for the Gemini macOS app in just a few days.
- 6x Faster Migrations: A complex internal code migration that would have taken months was completed six times faster by utilizing specialized agents to handle the heavy lifting of refactoring.
- Customer Zero Strategy: Google is using its own massive codebase as a laboratory to battle-test the Gemini Enterprise Agent Platform before offering it to global cloud customers.
3. The Changing Role of the Human Engineer
Despite the high percentage of AI involvement, Google maintains that human engineers remain the ultimate “gatekeepers” and “architects.”
- Review and Approval: Every line of AI-generated code must still be approved by a human developer.
- Shifting the Focus: Engineers are being pushed up the “abstraction ladder.” Instead of writing boilerplate code or database connection handlers (the “toil”), they focus on system architecture, edge case reasoning, and security verification.
- Performance Reviews: AI adoption has become so central that some Google DeepMind teams now have specific AI-usage goals integrated into their formal performance reviews.
4. The “Internal Rivalry” Context
Interestingly, while Google is pushing its own Gemini-based tools, reports from April 2026 suggest some internal friction. Some teams at Google DeepMind have reportedly been permitted to use Anthropic’s Claude Code for specific tasks, leading to a complex internal environment where multiple frontier models are being used to build the future of Google’s infrastructure.
