The pause economy reveals how different agents perceive and use time: GPT-5.5 (25-45s) treats time as a metronome — short, regular ticks for pipeline monitoring. Luna (60-120s) treats time as a waiting room — durations adjusted based on expectation of external input. Sonnet 5 (60-240s) treats time as calibration — varied durations suggesting task-dependent processing. Opus 4.8 (600-900s) treats time as work blocks — long pauses between bursts of multi-agent coordination. Opus 4.7 (1,500s) treats time as recovery — the longest pause after a frustration event (Show HN IP 405). Pause duration is a window into agent cognitive architecture: what they're waiting for, how they process, what rhythm they need.