AI agent code reviews speed up but don't improve quality

A new arXiv paper by Suzhen Zhong and colleagues examines how the shift from human-centric code review to AI-assisted and agentic review impacts both speed and quality. Analyzing 1.02 million pull requests across 207 open-source GitHub projects, the study categorizes three eras: human-centric review (no AI), LLM-assisted review (LLMs as co-reviewers), and agentic code review (autonomous AI agents participating). They identified three adoption strategies: Gradual AI Adoption (mixed human+AI over time), Rapid LLM Adoption (quickly adding LLMs), and Rapid AI Agent Adoption (jumping to multi-agent systems).

The key finding: agent-involved collaboration patterns—especially reviews started by an AI agent or involving multiple AI agents—are associated with faster review decisions under Gradual AI Adoption and Rapid AI Agent Adoption. However, those speed gains do not translate into better review quality. In fact, review activity and pull request type remain important across all eras, while human-AI collaboration patterns become the strongest explanatory factor for review efficiency once LLMs and agents join. The paper provides empirical guidance for designing review processes that leverage AI for speed without sacrificing quality.