A metaphysical account of agency for technology governance.

The way in which agency is conceptualised has implications for understanding human-machine interactions and the governance of technology, especially artificial intelligence (AI) systems. Traditionally, agency is conceptualised as a capacity, defined by intrinsic properties, such as cognitive or volitional facilities. I argue that the capacity-based account of agency is inadequate to explain the dynamics of human-machine interactions and guide technology governance. Instead, I propose to conceptualise agency as impact. Agents as impactful entities can be identified at different : from the low level of individual entities to the high level of complex socio-technical systems. Entities can impact their surroundings through different , and more influential channels of impact lead to higher of agency. Technology governance must take into account different channels of impact in the contexts of use, design and regulation.