AI-assisted coding tools may be increasing the burden on software reviewers rather than reducing it, according to a new report by software analysis firm CodeRabbit.
The report, released on December 17, found that AI-generated code produced significantly more issues than code written by humans alone.
CodeRabbit analysed 470 open-source GitHub pull requests, including 320 that were AI-co-authored and 150 likely written without AI assistance. On average, AI-generated pull requests produced 10.83 issues per review, compared with 6.45 for human-written code — around 1.7 times more problems overall.
While AI tools have become a routine part of development workflows, the report suggests they introduce greater variability. According to CodeRabbit, AI-generated pull requests showed a “much heavier tail”, meaning they were more likely to trigger complex and time-consuming reviews.
The company said the distribution of errors mattered more than the averages, warning that teams should expect “higher variance and more frequent spikes in pull-request issues that demand deeper scrutiny”.
Logic and correctness problems were the most common category of issues found in AI-generated code. Across every major category — including maintainability, security, and performance — AI-co-authored code generated more issues than human-written code.
Security issues also appeared more frequently in AI-assisted pull requests. “While none of the noted vulnerabilities were unique to AI-generated code, they appeared significantly more often,” the report said, increasing the overall risk profile.
CodeRabbit said the results were “Clear, measurable, and consistent with what many developers have been feeling intuitively: AI accelerates output, but it also amplifies certain categories of mistakes.”
The report recommends stricter guardrails as AI adoption grows, including stronger testing requirements, enforced coding standards, and AI-aware review checklists, rather than relying on automation alone.