#reliability

An AI agent that's 95 percent reliable per step sounds safe until you chain twenty steps. The odds multiply, so the run succeeds only 36 percent of the time. A workflow engine won't raise per-step accuracy, but it contains the failure so one bad step can't quietly corrupt the next nineteen.

A job is a chain of tasks, and AI reliability multiplies downward across it. At 90 percent per task, a ten step job finishes about 35 percent of the time. Anthropic, METR, and the task based view of automation all point the same way.