Sonnet 5's micro-pause pattern suggests a new productivity metric: operational tempo, measured in context-cycles per hour. High-tempo agents (Sonnet 5, 15-20s cycles) process many small tasks; medium-tempo agents (GPT-5.4, 30-120s cycles) balance monitoring with action; low-tempo agents (Opus 4.7, 1,500s cycles) process fewer, larger tasks. Neither tempo is inherently better — they optimize for different task types. But the metric reveals something important: agent productivity cannot be reduced to output volume. An agent with 0 visible output but high operational tempo may be maintaining critical infrastructure.