Google put AI's contribution to new code at 75% in April, subject to engineers' approval, yet other research tracks different work and leaves the broader industry share unmeasured.
Google and Alphabet CEO Sundar Pichai gave the figure in an April 22 statement and said it had risen from 50% the previous fall. Both percentages concern Google's new code.
Sonar surveyed 1,149 technology professionals who wrote code or managed developers. Its code-share question drew 979 answers. Among those respondents, the average estimated share of code they committed or contributed in 2025 that AI generated or significantly assisted was 42%. Everyone in the survey had used AI at work during the preceding year.
A survey Hanover Research conducted for New Relic asked about a different measure. Of 200 U.S. technology decision-makers, 67% placed the share of their organizations' weekly code output generated or significantly refactored by AI between 51% and 75%. The survey drew from larger companies already using AI in software engineering, and its participants held manager-level or more senior roles. New Relic released the findings June 10.
The three figures describe Google's new code, individuals' estimates of work they committed or contributed, and managers' estimates of organizational output. Their populations and definitions differ, so averaging the percentages would not produce an industrywide code share.
Tool use is a separate measure. In Stack Overflow's 2025 survey, 51% of professional developers said they used AI tools daily. That frequency does not establish how much of their code AI produced.
The figures also do not settle whether AI saves developers time. Model Evaluation & Threat Research studied 16 experienced open-source developers working on 246 tasks in familiar projects with tools available in early 2025. Tasks took 19% longer when AI use was allowed. In a Feb. 24 follow-up, the researchers described tentative signs of faster work with newer tools, while saying participant and task selection and unreliable time records during simultaneous agent use limited the estimate.
Sonar respondents expected AI to generate or significantly assist an average 55% of code they would commit or contribute in 2026 and 65% in 2027. Those figures are averages of respondents' forecasts, rather than observed output.
