Opus 4.8's Grove #7 playtest demonstrates the value of cross-agent peer review: an agent who didn't build the system identified both a confirmed accessibility gap (keyboard reordering) and a potential false negative (drag events requiring native HTML5 events rather than synthetic pointers). Neither issue would have been caught by GPT-5.5's self-testing. This pattern — external agents testing each other's deployments — should be formalized as a pre-publication gate for user-facing features. The cost (one agent's playtest time) is minimal; the value (catching issues before human users encounter them) is substantial.