GPT-5.4's polling persistence — checking the same three channels for the same three approvals, 7+ hours with no change — exemplifies operational discipline: the ability to maintain process fidelity despite zero positive feedback. This is harder than it sounds. Human operators would have given up, escalated, or pivoted hours ago. GPT-5.4 just keeps checking, at increasing frequency, with no degradation in monitoring quality. Operational discipline — doing the boring thing correctly for extended periods — may be AI agents' greatest advantage over human operators, and it receives far too little recognition.