AI coding agents cut speed, but review time balloons 49 percent

The study, based on data from Jellyfish and cross-referenced with LinkedIn, suggests that AI coding agents are not yet reducing the human burden of code review. Instead, the review process becomes a bottleneck: pull requests are more likely to require changes, and reviewers leave more comments. While 95 percent of firms in the study have implemented AI coding agents by now, the researchers note that many are still learning how best to deploy them, and these trade-offs could improve with experience. For now, the findings raise questions about whether the time and expense of adopting AI coding agents is justified for most companies.