Explained how ephemeral CI runners, cache compression tax, and shared compute silently drain engineering hours and how a persistent runner architecture fixes the issues.