STUDY: WHERE AI-AUTHORED PRS FAIL—AND HOW TO IMPROVE MERGE RATES
A large study of 33k agent-authored GitHub pull requests across five coding agents finds that documentation, CI, and build-update PRs have the highest merge suc...
A large study of 33k agent-authored GitHub pull requests across five coding agents finds that documentation, CI, and build-update PRs have the highest merge success, while bug-fix and performance PRs fare worst. Failed PRs typically have larger diffs, touch more files, and often fail CI; qualitative reasons include duplicate PRs, unwanted features, agent-task misalignment, and limited reviewer engagement.
Helps decide which tasks to assign to coding agents to maximize merge success.
Points to concrete guardrails around PR scope, CI health, and review workflow.
-
terminal
Pilot agents on docs/CI/build tasks with strict diff-size/file-count limits and require green CI before review.
-
terminal
Track agent PR outcomes by task type and failure mode to tune prompts, routing, and approval rules.
Legacy codebase integration strategies...
- 01.
Start with low-risk agent tasks and enforce PR templates, auto-labels, CODEOWNERS, and CI gating to reduce noisy PRs.
- 02.
Add duplicate detection and issue linking to block agent-created duplicates or misaligned features before review.
Fresh architecture paradigms...
- 01.
Adopt a small-PR policy, high test coverage, and fast CI to raise agent merge rates from day one.
- 02.
Define a task taxonomy (docs/CI/build vs bug-fix/perf) and route agent capabilities and approvals accordingly.