Definition
The AI code review bottleneck is the shift of scarce developer capacity from writing code to reviewing agent-generated pull requests as coding-agents increase commit and PR volume.
Key Points
-
2026-10-09: Harvard/Jellyfish study (700+ firms): agents +30% LOC, +23% PRs, no significant Jira throughput gain; review time +49%, change-request rate ~2×, comments/PR +35%; AI review tools still minority of comments (2026-10-09-harvard-jellyfish-ai-bottleneck-paper)
-
2026-08-12: Funding cluster validates bottleneck thesis — coderabbit (45M), rwx ($12M) (ai-code-validation-funding-wave)
-
2026-08-14: 54% of agent PRs not merged — review capacity remains bottleneck even at 46% success (2026-08-14-claude-code-daily-maintenance)
-
TED CTO Andy Merryman: AI raised productivity but PRs grew too large for reviewers (2026-08-02-github-changelog-stacked-pull-requests)
-
stacked-pull-requests are a platform response: small dependency-ordered layers for parallel review
-
Secondary coverage cites GitHub projecting ~14B commits in 2026 (COO Kyle Daigle / Latent Space) — attribute if used (2026-08-02-github-stacked-pull-requests-public-preview)
Related
- agentic-change-management
- coderabbit
- blacksmith
- rwx
- jellyfish
- coding-agents
- stacked-pull-requests
- ai-code-review-era
- github
- agent-production-tooling-2026