Recall is the metric now: AI review inverted precision culture
Ben Nevile on why agent-side filtering flips review etiquette from “only raise sure things” to “raise everything potentially relevant.”
Source: Recall is the metric now: What AI code review suddenly gets backwards
Classic review etiquette optimized precision because every comment cost scarce human author attention. With agents that can triage and auto-fix, the consumer-side filter gets cheap—so completeness (recall) becomes the thing to maximize.
Same inversion shows up outside PRs: people ask for more signal into agent-read inboxes, not less. Channels increasingly operated by agents reward “send the useful things” over “don’t bother humans.”
Reads as a metric realignment for anyone tuning AI reviewers: optimizing only for low false positives may be optimizing for yesterday’s bottleneck.
Why it matters for code review
If an agent sits between findings and humans, missing a real issue is costlier than an extra candidate. Severity routing and auto-fix loops make high-recall first passes rational—then precision belongs in the judge/filter stage.