A review queue becomes a production bottleneck when agent-assisted developers generate changes faster than qualified reviewers can assess them. Six developers producing 48 pull requests daily overwhelm two reviewers who can clear 20, creating a backlog that reaches 140 by Friday. section 10.5
Unchecked queue growth creates predictable failure modes:
- Reviewers replace deliberate assessment with diff skimming, test checks, and file-path inspection. section 10.5
- Trivial changes compete with architectural changes, burying systemically risky decisions among low-risk work. section 10.5
- A review process can become theater: it signals oversight while failing to verify intent, conventions, or subtle coupling. section 10.5
The innocent three-line change
An agent-generated checkout change altered currency rounding. Although the diff looked harmless, approval missed a floating-point precision error that later surfaced in production. section 10.5
Figure 10.5 Approval Queue Paradox.

Visualizes the approval queue paradox: faster code production can reduce review quality when reviewer capacity remains fixed. section 10.5
