Overview
- Google, which disclosed Wednesday that 75% of its new code is AI‑generated and approved by engineers, said the share was about 50% last fall and roughly 25% in 2024.
- CEO Sundar Pichai said teams are moving to agentic workflows, where software agents plan and carry out tasks like large code migrations and refactoring, with one recent migration finished six times faster than a year ago.
- At Cloud Next 2026, Google introduced two new AI chips and a Gemini Enterprise Agent Platform and committed up to $185 billion to infrastructure to run long‑running autonomous agents.
- Google executives said human approval remains a critical checkpoint, with engineers shifting from hand‑coding to roles focused on system design, safety review, and final sign‑off on agent‑produced changes.
- Internal adoption is not uniform, with some DeepMind teams permitted to use Anthropic’s Claude Code alongside Google’s models, and peers such as Microsoft, Meta, and Snap reporting growing shares of AI‑authored code.