Google has crossed a significant threshold in AI-assisted software development: three-quarters of all new code written at the company is now generated by artificial intelligence and approved by human engineers. This represents a dramatic acceleration from just six months ago, when that figure stood at 50%, signaling a fundamental transformation in how the world's largest search company builds software (Source 1, 2).

What Does 75% AI-Generated Code Actually Mean for Software Development?

The jump from 50% to 75% AI-generated code isn't simply about replacing human programmers with machines. Instead, it reflects a shift toward what Google calls "agentic workflows," where engineers manage fully autonomous digital task forces that handle routine coding, testing, and deployment. Google CEO Sundar Pichai explained that this transformation comes with complexity, as organizations learn to oversee hundreds or thousands of AI agents simultaneously (Source 1, 2).

The practical impact is striking. Google recently completed a complex code migration project six times faster than would have been possible a year ago using only human engineers. This wasn't a simple task; it required coordinating multiple AI agents working alongside engineers to handle the intricate work of moving code between systems (Source 1, 2).

The shift reflects broader industry trends. Gemini Enterprise, Google's AI system for business customers, has grown 40% in paid monthly active users quarter over quarter, suggesting that other companies are adopting similar AI-assisted development practices (Source 1, 2).

How Are Companies Managing the Explosion of AI Agents?

As AI agents become more prevalent in enterprise environments, Google has introduced a comprehensive platform to help organizations build, deploy, govern, and monitor these autonomous systems. The Gemini Enterprise Agent Platform addresses a critical challenge: most companies now face the problem of managing hundreds or thousands of AI agents rather than just building individual ones.

• Build and Deploy: Agent Studio provides a low-code interface for creating agents using natural language, while the upgraded Agent Development Kit includes a graph-based framework for orchestrating multiple agents working together, with sub-second cold starts for rapid deployment.
• Governance and Security: Agent Identity assigns every agent a unique cryptographic ID with defined authorization policies, while Agent Gateway enforces security policies and protects against prompt injection, tool poisoning, and data leakage.
• Monitoring and Optimization: Agent Anomaly Detection flags suspicious behavior, Agent Evaluation scores live performance, and Agent Observability dashboards trace execution paths for real-time debugging and rapid problem resolution.

Google's approach emphasizes vertical integration, designing chips, models, infrastructure, and application layers together rather than assembling components from different vendors. This strategy aims to deliver what Google Cloud CEO Thomas Kurian called "a comprehensive backbone for innovation".

"The early versions of AI models were really focused on answering questions that people had and assisting them with creative tasks. Now we're seeing as the models evolve people wanting to delegate tasks and sequences of tasks to agents," said Thomas Kurian, Google Cloud CEO.

Thomas Kurian, Google Cloud CEO

Why Did Google Split Its AI Chips Into Separate Training and Inference Processors?

Alongside its software announcements, Google introduced specialized versions of its eighth-generation Tensor Processing Unit (TPU), a custom chip designed specifically for AI workloads. Rather than using one chip for both training AI models and running them in production, Google created distinct processors optimized for each task.

The training TPU delivers 2.8 times better performance than Google's previous seventh-generation Ironwood chip, while the inference TPU offers 80% better performance. Google's Senior Vice President and Chief Technologist for AI and Infrastructure, Amin Vahdat, explained the reasoning: "With the rise of AI agents, we determined the community would benefit from chips individually specialized to the needs of training and serving".

The inference chip relies on static random-access memory (SRAM), a technology also used in competing chips from companies like Cerebras and NVIDIA's upcoming Groq 3 LPU. Each TPU inference chip contains 384 units of SRAM, triple the amount in the previous generation, enabling the chip to handle massive throughput while maintaining low latency.

"The architecture of the chip is designed to deliver the massive throughput and low latency needed to concurrently run millions of agents cost-effectively," noted Sundar Pichai, Google CEO.

Sundar Pichai, CEO at Google

Demand for Google's TPU chips is growing rapidly. All 17 national laboratories in the U.S. Energy Department use AI co-scientist software running on Google's chips, while Citadel Securities relies on them for qualitative research software. AI safety company Anthropic has committed to using several gigawatts of Google's TPU capacity.

What Real-World Impact Are These Changes Having on Enterprise Customers?

Google's announcements aren't theoretical. Large enterprises are already deploying AI agents at scale. GE Appliances operates more than 800 AI agents across manufacturing, logistics, and supply chain operations. KPMG achieved 90% adoption of Gemini Enterprise among employees and deployed more than 100 agents in the first month. Tata Steel deployed over 300 specialized agents in nine months.

Perhaps most significantly, pharmaceutical giant Merck announced a partnership valued at up to $1 billion to build an agentic platform across its research and development, manufacturing, and commercial functions. This signals that enterprise leaders view AI agent orchestration as a strategic priority, not a nice-to-have feature.

Google is backing this ecosystem with a $750 million fund to support partners in building and deploying agentic AI, along with early access agreements with consulting firms including McKinsey and Deloitte. These moves position Google as a central player in what industry observers see as the next major evolution of enterprise software.

The shift toward AI-generated code and autonomous agents represents a genuine inflection point in software development. Whether this acceleration continues, and how well enterprises manage the complexity of thousands of autonomous agents, will likely define the next chapter of enterprise AI adoption.