Google: 75% new code AI
- A social post claims Google now generates about 75% of its new code using AI, reframing engineers as reviewers. - The post suggests entry-level 'writing' roles are shrinking while engineers act as editors and reviewers. - The claim circulated widely on X, portraying internal shifts toward AI-assisted development workflows at large tech firms. (x.com)
Google says artificial intelligence now generates 75% of its new code, with engineers reviewing and approving the output. (blog.google) Chief Executive Officer Sundar Pichai made the statement in a post tied to Google Cloud Next on April 22, 2026, saying the share was up from 50% “last fall.” (blog.google) That is a sharp jump from October 29, 2024, when Pichai told investors that “more than a quarter” of Google’s new code was being generated by AI and then reviewed and accepted by engineers. (abc.xyz) Google has spent the past year turning that workflow into products. Gemini Code Assist for individuals and Gemini Code Assist for GitHub became generally available on May 20, 2025, with Google pitching them as tools for writing, editing and reviewing code inside common developer tools. (blog.google) The company’s own language around the job has also shifted from writing to checking. In the April 22 post, Pichai said Google engineers are “orchestrating” autonomous agents, while the 2024 earnings call described humans as the people who review and accept machine-generated code. (blog.google) (abc.xyz) Google’s research arm has been arguing that these tools change speed more than they eliminate oversight. A Google Cloud post on DORA’s April 24, 2025 report said a 25% increase in AI adoption was associated with higher individual productivity, faster code review and faster approval speed. (cloud.google.com) Google has also framed the tools as standard equipment rather than niche experiments. In February 2025, the company launched a free preview of Gemini Code Assist for individuals with up to 180,000 code completions per month, and in May 2025 it said the service was generally available. (blog.google 1) (blog.google 2) What Google has not published in these statements is a breakdown of what counts as “new code,” how much of that 75% is autocomplete versus larger generated blocks, or how the share varies across teams. The public claim, as stated by Google, is about code generation volume, not about fully autonomous software releases. (blog.google) The result is a clearer picture of what the company says changed inside Google between late 2024 and April 2026: AI moved from writing more than one-quarter of new code to writing three-quarters, while engineers stayed in the approval loop. (abc.xyz) (blog.google)
