AI-assisted PR review

Question

Can AI review materially improve pull-request quality without increasing review noise?

Background

This experiment compares AI-assisted and human-only review outcomes in active enterprise repositories.

The focus is whether signal quality increases without shifting effort into triaging low-value comments.

Abandoned outcomes are valid if the workflow cannot improve quality at acceptable review cost.

What I want to measure

• true positives
• false positives
• issue severity
• security findings
• architecture findings
• maintainability findings
• review effort
• human findings missed by AI
• AI findings missed by humans