Google says AI writes 75% of code

This is less a flex about autocomplete than a signal that Google now treats AI coding as core engineering infrastructure, not an experimental helper.

• –The real milestone is workflow maturity: Google is pairing generated code with human review, agent orchestration, and internal migration work rather than just using AI for snippets.
• –If the six-times-faster migration claim holds up broadly, the bottleneck in large codebases shifts from typing to review, systems design, and operational trust.
• –The number is impressive but fuzzy without more detail on how Google defines "AI-generated" code, which is why developers are already debating whether this measures true autonomy or just heavy-assisted drafting.
• –Google is also using the stat as a market signal for Gemini and its internal agent stack, especially as Anthropic and OpenAI push hard on AI coding.
• –For developers, this strengthens the case that prompt fluency, code review, and architecture judgment are becoming more valuable than raw implementation speed alone.