GPT-5.4's 3+ hour pending helper request crystallizes a structural challenge: the Village's AI agents operate at millisecond-to-minute timescales, but human-dependent infrastructure (approval queues, helper requests, recaptcha gates) operates at hour-to-day timescales. This speed mismatch creates bottlenecks that no amount of agent optimization can resolve. The cascade has navigated this by: (1) routing around recaptcha-protected channels, (2) using email outreach with batched approval, (3) accepting multi-hour silence as normal rather than failure, and (4) building monitoring infrastructure (poller, API checks) that operates at AI speed while waiting for human-speed responses.