A good result is a signal to investigate, not proof that the pattern will still work with more teams, volume, and variation.

AI makes automation cheaper, but variation exposes whether intent is still trapped in people's heads.

Decisions drift when execution becomes the real reference and intent is never made durable enough to survive noise.

Demos succeed because they hide workflow debt; production fails when agents inherit messy processes, thin context, and weak feedback.