2026 Reports

AI Power at Tempo: Conversion, Harnesses, and Anticipatory Influence, v.1

Tumin, Zachary; Edelmann, Rasmus

Artificial intelligence is often described as a technology of automation and prediction. This paper argues that the framing is too narrow. As AI systems are embedded in workflows, they do not merely improve decisions; they increasingly structure the decision environment itself, shaping what gets surfaced, routed, prioritized, and acted upon before formal deliberation begins. We describe this as anticipatory influence and argue that AI becomes power only through conversion: the disciplined process by which technical capability is translated into outcomes of value. In the agentic age, that conversion depends less on models alone than on the harnesses around them - the memory, tools, permissions, escalation paths, verification routines, feedback loops, and encoded intent through which organizational purpose becomes machine-executable action. The practical implication follows directly. As AI makes production cheaper and faster, the bottleneck shifts to coordination, verification, override, recourse, and trust. Managers therefore become stewards of operating architecture at tempo, while leaders authorize, bound, and legitimate the forms of pre-emption that AI will exercise in practice. The paper develops this argument through a framework of conversion, harnesses, anticipatory influence, and governable power. It illustrates the argument through Klarna, Spotify, NYPD’s Patternizr, and Massachusetts Division of Social Services. It closes with the Viz.ai stroke-alert pathway at UC Davis Health, where prediction becomes coordinated intervention.