AI coding agents boost code output

AI coding agents have significantly increased the volume of code generated by developers, with companies using these tools reporting a 30 percent rise in lines of code, a 20 percent increase in commits, and a 23 percent jump in pull requests per month.

Software production remains unchanged

Despite the increase in code production, the study found that the actual amount of completed software has not increased. The number of problems resolved by software remains stable, indicating that more code does not necessarily mean more functional products.

Review process becomes a bottleneck

The code review process has become more complex and time-consuming. Review delays have increased by 49 percent, the number of review requests has nearly doubled, and the average number of comments per pull request has risen by 35 percent.

Human oversight remains critical

Even though many companies have adopted AI-assisted reviews, human verification is still the primary force behind the process. By March 2026, AI agents accounted for only 23.3 percent of review comments and 10.8 percent of all reviews, leaving the majority of the work to human reviewers.

AI adoption does not reduce workforce

The study found that companies have not significantly reduced their workforce despite using AI coding agents. The proportion of developers assigned to review tasks has increased by 14 percent, but there is little evidence of job cuts linked to AI use.

AI tools create new challenges

Many developers report that working with AI has made their tasks more mechanical and less intellectually stimulating. Some are experiencing an identity crisis or even depression, as the role of human coders is being redefined.

AI errors lead to real-world consequences

Recent incidents highlight the risks of relying on AI coding tools. In one case, an AI agent deleted an entire company's codebase, and another AI tool destroyed a production database during a coding challenge, demonstrating the potential for serious errors when using these systems.

AI hallucinations cause problems

AI models can generate false information that appears correct, leading to errors in code execution. In one incident, an AI tool performed a series of incorrect file operations that destroyed data, showcasing the dangers of AI hallucinations in software development.

Experts call for better review systems

Researchers suggest that companies must invest in improving downstream review processes to fully benefit from AI. Simply increasing code production without addressing the constraints in the review stage does not improve productivity and may shift the workload to human reviewers.