The most rigorous data point comes from ShiftMag, which analyzed 4.2 million developers between November 2025 and February 2026 and found that AI-authored code accounts for 26.9% of all production code — up from 22% the prior quarter. This is a direct measurement of code actually written by AI tools, not a self-reported estimate.
Higher figures circulating online — such as 42% from Sonar's State of Code Developer Survey and 41% from Netcorp — come from developer surveys asking about code that is "AI-generated or assisted." That broader definition includes code merely reviewed, suggested, or lightly edited with AI help, which inflates the share well beyond what AI tools independently produce.
AI coding tool adoption is genuinely widespread: a 2026 Developer Ecosystem Research Group survey of 15,000 developers found 73% of engineering teams use AI coding tools daily. But high adoption rates don't translate directly into high code-generation share — developers still author the majority of production code themselves, with AI playing a supporting rather than primary authorship role.