The liability stack

The liability stack is the layered structure of legal and professional responsibility that determines where AI actually gets adopted. It explains a pattern that capability cannot: the same organization that lets AI write its documents will not let AI touch its decisions — because the two sit on different layers of the stack.

Healthcare is the cleanest illustration. Per the AMA's 2025 survey, 66% of US doctors use AI — but overwhelmingly for documentation, billing codes and discharge notes; only 31% for clinical decisions, and 82% call liability coverage critical for adoption. AI penetrates the paperwork layer freely because errors there are cheap and reversible. It stalls at the decision layer because a wrong answer that changes what happens to a human body must have an owner — and that ownership is unsettled.

The general law: AI adoption follows settled liability, not available capability. Markets where responsibility for machine error has a clear owner (or a price, via insurance) absorb AI quickly; markets where it does not, stall — however good the models get. Reading the liability stack of an industry predicts its AI adoption curve better than reading its benchmarks.