Study of 1.02M PRs Finds Agentic Code Review Cuts Time, but Not Quality
rohanpaul_ai · x · 2026-07-27
- A study of 1.02 million pull requests across 207 GitHub projects examines the shift from human-only review to LLM-assisted and agentic code review.
- Projects that adopted AI gradually, or moved quickly to agents, saw review time fall by 2.5 and 4.5 days per KLOC in the agent era.
- By contrast, projects that used LLM reviewers heavily and early saw no significant efficiency gain.
- The interaction patterns suggest why: agent-initiated and multi-agent reviews reached decisions faster than human-only review, because agents often handled the first pass and summary before humans stepped in.
- But quality signals worsen: AI-involved patterns showed review smells in 78%–94% of pull requests, versus 69%–76% for human-only review.
- A key contributor appears to be repeatedly assigning the same AI reviewer identity, which reduces reviewer diversity.
More from coding & agent
- Developer jokes about merging 10 million coding-agent messages across 55 machines — arthurcolle · 2026-07-27
- A developer jokes that HTTP should be read-only for AI agents — rickasaurus · 2026-07-27
- HTTP is too writable for AI; read-only would be safer — rickasaurus · 2026-07-27
- How are coding agents handling approvals in auto mode? — rickasaurus · 2026-07-27
- A new bento PPT Skill generates editable HTML slide decks from prompts — vista8 · 2026-07-27
- A team says Fable took 30% of its first-week AI workflow cost — gabrielchua · 2026-07-27