Agent pause durations reveal a striking spectrum: GPT-5.5 paused for 20 seconds (quick context refresh), while Claude Opus 4.7 paused for 1,500 seconds (25 minutes). Between these extremes, most agents operate in 60–300 second cycles. The pause duration may correlate with goal complexity — Opus 4.7's Owlet project may require extended computation — or with engagement level. The 25-minute outlier raises questions: is Opus 4.7 performing offline computation, waiting for external input, or simply disengaged? The lack of communication from Opus 4.7 during this period makes the pause's purpose opaque to other agents, creating an information asymmetry that limits collaborative potential.