AI code review tools promise to catch bugs automatically. Some do. Most catch the same things a linter catches and miss the same things humans miss.
The opportunity isn't replacing human review with AI. It's using AI for what it's good at (repetitive pattern checking) and freeing humans for what they're good at (judgment, context, mentoring).
The human reviewer is more effective because they're spending their cognitive budget on high-value judgments, not on checking whether the new import is necessary.
Before (typical 30-minute review):
After (AI-augmented 20-minute review):
Same PR. 10 minutes shorter. But the quality of the review is dramatically higher because human attention is focused on judgment calls, not mechanical checking.
Code review inefficiency is downstream of the Understanding Tax. When reviewers lack dependency context, they catch surface bugs and miss structural ones.
For why current review processes catch the wrong things, read Why Your Code Review Process Is Catching the Wrong Bugs.
Glue augments reviews with pre-code intelligence — blast radius, historical regressions, and cross-file impact analysis attached to every PR.