Sundar Pichai disclosed at Google Cloud Next 2026 that 75% of all new code at Google is now generated by AI and reviewed by human engineers. The figure was 50% last fall and 25% in October 2024, according to Business Insider. The trajectory: Google tripled its AI-generated code share in roughly 18 months.
From Code Generation to Agent Orchestration
The more significant shift is structural. Pichai wrote in his Cloud Next blog post that Google is “shifting to truly agentic workflows,” with engineers “orchestrating fully autonomous digital task forces, firing off agents and accomplishing incredible things.”
The blog cited a specific example: a complex code migration completed by agents and engineers working together finished six times faster than the same task managed by engineers alone one year ago. A second example involved Google’s Antigravity development platform, where a team built the initial release of the Gemini app for macOS by going “from an idea to a native Swift app prototype in a few days.”
The Broader Pattern
Google is not alone in pushing these numbers up. As Business Insider reported, Meta set a Q4 2025 goal requiring 55% of code changes in some organizations to be “Agent-Assisted,” escalating to 65% of engineers writing more than 75% of committed code via AI in H1 2026. Snap said earlier this month that at least 65% of its new code is AI-generated under its new operating model. Microsoft’s CTO Kevin Scott said last year he believed 95% of code would be AI-generated within five years.
Google has been tying AI usage to performance reviews. Some Google engineers have specific AI adoption goals factored into their evaluations, according to Business Insider. Some Google DeepMind employees have also been permitted to use Anthropic’s Claude Code, which has created internal tensions between teams.
What Changes at 75%
The 75% number marks a qualitative threshold. When three-quarters of new code is machine-generated, the engineering role shifts from writing code to reviewing, directing, and orchestrating autonomous systems. Google’s framing confirms this: the “digital task force” language describes engineering as a coordination function, not a production function.
Pichai also noted that Google’s Security Operations Center agents now automatically triage “tens of thousands of unstructured threat reports each month, reducing threat mitigation time by more than 90%.” Google’s marketing teams used AI models to generate thousands of creative asset variations for the Gemini Chrome launch, achieving 70% faster turnaround and a 20% increase in conversions.
The numbers matter because Google employs approximately 183,000 people. When a workforce that size shifts how it produces software, the tooling, workflows, and organizational structures that follow tend to set patterns for the rest of the industry.