Google AI Now Generates 75% of All New Code

Google AI Now Generates 75% of All New Code

1. Overview

This article covers Google's adoption of AI for code generation, the infrastructure supporting this transformation, and how Google is implementing AI across various domains including security and operations. It's intended for developers, technical leaders, AI professionals, and anyone interested in Google's AI strategy. No specific prerequisites are required beyond a basic understanding of AI and software development concepts.

2. Core Concepts

• AI-generated code: Code created by AI models and reviewed by human engineers

• Agentic workflows: Systems where AI agents work alongside or autonomously with engineers

• Gemini: Google's AI platform for the agentic era

• TPUs: Tensor Processing Units, Google's specialized hardware for AI workloads

3. Detailed Explanation

AI-Generated Code at Google

Google has significantly increased its use of AI for code generation internally. Currently, 75% of all new code at Google is AI-generated and approved by engineers, up from 50% last fall. This represents a substantial acceleration in AI adoption for software development.

Google has been using AI to generate code internally for a while, and the company is now shifting to "truly agentic workflows" where engineers orchestrate fully autonomous digital task forces. This approach has demonstrated clear benefits - a particularly complex code migration done by agents and engineers working together was completed six times faster than was possible a year ago with engineers alone.

The Gemini app on MacOS serves as an example of this approach, built with Google's agentic development platform Antigravity, which allowed the team to go from an idea to a native Swift app prototype in just a few days.

Google's AI Infrastructure

Google's first-party models now process more than 16 billion tokens per minute via direct API use by customers, up from 10 billion last quarter. To support this growth, in 2026, just over half of Google's overall machine learning compute investment is expected to go towards the Cloud business to benefit cloud customers and partners.

To handle the demands of AI agents, Google is introducing the eighth generation of its Tensor Processing Units with a dual chip approach:

• TPU 8t: Optimized for training, scales up to 9,600 TPUs and 2 petabytes of shared, high-bandwidth memory in a single superpod

• TPU 8i: Optimized for inference, connects 1,152 TPUs in a single pod, dramatically reducing latency

These new TPUs will be offered to Cloud customers as a core part of Google's selection of compute processors, alongside a portfolio of NVIDIA GPU instances.

Gemini and AI Agents

Google has firmly entered the "agentic Gemini era," with Gemini Enterprise serving as the end-to-end system for this new approach. Gemini Enterprise has shown impressive growth, with 40% growth in paid monthly active users quarter-over-quarter in Q1.

As organizations adopt AI agents at scale, the conversation has evolved from "Can we build an agent?" to "How do we manage thousands of them?" In response, Google is introducing its new Gemini Enterprise Agent Platform, which provides secure, full-stack capabilities to build, scale, govern and optimize agents with confidence.

Google engineers are using Gemini models for generating code.

AI in Security and Operations

While AI can increase security risks, Google's Cloud customers now have AI on their side to protect their organizations. Google is unveiling a range of new agentic solutions for threat detection as part of an AI-powered cybersecurity platform that combines Google's Threat Intelligence and Security Operations with Wiz's Cloud and AI Security Platform.

Google's Security Operations Center agents automatically triage tens of thousands of unstructured threat reports each month, reducing threat mitigation time by more than 90%. The company has also built and actively uses Gemini-based AI agents (like CodeMender) to find and fix critical software flaws.

In marketing operations, Google has demonstrated AI's value through the launch of Gemini in Chrome, where marketing teams used models to rapidly generate thousands of variations of creative assets. This approach led to 70% faster turnaround and a 20% increase in conversions, getting products to market faster and more effectively.

Google's Approach as "Customer Zero"

To be the best partner, Google positions itself as "customer zero" for its own technologies, using them internally to imagine, test, build and scale the best solutions for cloud customers. Examples include Bigtable, which powers many Google services, and TPUs, which have been important in training and powering Google's Gemini models.

This internal testing helps Google develop and refine technologies that benefit its cloud customers and partners, ensuring that solutions are practical and effective in real-world scenarios.

4. Common Questions

• How does Google ensure AI-generated code quality? AI-generated code at Google is approved by engineers before being integrated into production systems.

• What are the benefits of using AI for code generation? Google has demonstrated that AI-assisted development can complete complex tasks like code migration six times faster than with engineers alone.

• How does Google balance AI adoption with human oversight? Google employs a collaborative approach where AI agents work alongside engineers. Engineers orchestrate and approve AI-generated output, ensuring quality.