The results of the "Survey on AI Utilization in Development (2026)" conducted by ITmedia Enterprise and Keyman's Net have revealed that AI-driven development by non-engineers is spreading within large enterprises.
AI development by non-engineers reaches 65% in large enterprises, while insufficient verification of generated code and delays in detecting issues remain challenges
This article is a translation. Read the Japanese original
In the survey, 65.0% of respondents from large enterprises using AI reported that development is being conducted by non-engineers. Furthermore, 50.0% of large enterprise respondents stated they are "using it across the entire company," indicating that while development by non-engineers is progressing, the scope of AI utilization is expanding across entire organizations.
However, there are concerns that governance and verification systems are not keeping pace with this expanding utilization. Among large enterprise respondents using AI in development, 55.0% responded that they "cannot grasp/do not know" when troubles or incidents requiring IT department intervention occur due to AI use.
Challenges also remain regarding code quality. The most frequent response among large enterprise respondents was that they "do not know the quality of errors and defects" because they do not verify AI-generated code. Additionally, 46.0% of large enterprise respondents cited the accumulation of "understanding debt (a state where the structure of the code cannot be grasped)" as a disadvantage of using code without understanding its intent or structure.
As a future measure, 56.0% of large enterprise respondents called for the "introduction of tools to automatically check and audit the quality and vulnerability of AI-generated code." Among IT department respondents, in addition to the introduction of quality check tools, the introduction of management tools capable of visualizing and limiting token consumption and usage costs was also listed as a top requirement by 64.3%.
Sources
- 大企業に勤める回答者の過半数が「非エンジニアによる開発を実施」 全社に行き渡ったコード、検証できているのか？【調査】 （ITmedia AI+、2026-09-30）