General AM

Three-quarters of large enterprises are investing in digital twins, and the global market is on track to hit $49.47 billion in 2026, headed for $328.51 billion by 2033 at a 31.1% compound annual growth rate. The board has heard the pitch, the budget is approved, and somewhere in your organization a proof of concept is already running on a single pump or a single production line.

About 85% of those initiatives never make it out of the pilot. They produce a promising demo and a slide deck of green dashboards, then they stall. The virtual replica works beautifully on the one asset it was built for and never scales to the other 4,000.

A reliability engineer at a mid-sized manufacturer told a story this spring that should worry every leader who signed off on an AI maintenance budget. Her plant had rolled out a predictive maintenance (PdM) platform vendors promised could forecast failures 6 to 12 months out with “over 85% accuracy.” Eight months in, her most experienced technician had started ignoring the alerts. Not because they were always wrong — but because too often, he’d pull a pump, tear it down, and find nothing wrong.

That technician wasn’t being difficult. He was doing math the model’s marketing brochure never showed him.

When autonomous maintenance acts on safety-critical assets, accountability stops being philosophical and becomes a procurement criterion. Five tests separate accountable autonomy from unmanaged liability.