Messiness Predicts AI Job Risk

Not all work is the same: jobs in 'messy' environments with ambiguous instructions, variable contexts, and adaptive goals are harder for AI to displace than highly routinized task bundles. Evaluations that only test discrete task performance (pass the bar, read scans) miss whether deployed systems can pursue real workplace goals and handle downstream bottlenecks. — Focusing policy and corporate planning on an occupation's contextual 'messiness' changes predictions about displacement, retraining needs, and regulation.