Home / Technology / 81% of Dev Teams Merge Unknown AI Code
81% of Dev Teams Merge Unknown AI Code
22 Nov
Summary
- Over 80% of software teams admit to merging AI code they do not understand.
- Manual review and AI-on-AI code review show mixed results for verification.
- Few teams have incident response plans for AI-generated errors.

Recent research indicates that a vast majority of software development teams, precisely 81%, have integrated AI-generated code into their projects without fully understanding its intricacies. This practice contributes to a "verification gap," where the speed of AI code generation outpaces thorough validation.
Approaches to verifying this code are varied. While 55% of teams opt for manual reviews, and 51% involve dedicated QA teams, a notable portion utilizes a secondary AI model to assess the output. However, this "LLM-as-a-judge" method correlates with a higher rate of merging code that remains unverified.
Furthermore, the TestMu AI report highlights a lack of preparedness for AI-related failures. Only 13% of teams have established incident response provisions for AI-generated errors, suggesting a widespread vulnerability in current quality engineering practices.