What happens when you let AI agents run an entire agency with zero human oversight? A Carnegie Mellon experiment shows it collapses in chaos—proving we still need human judgment, simple solutions, and even the quirks (think R2-D2’s beeps) that make smart machines truly effective.