A strong international reform agenda has been established around the idea that benefit recipients must be ‘activated’ to find jobs. This approach, which has found support across the political spectrum in times of affluence and austerity, rests on previously contested assumptions about human motivation, choice, action and personal responsibility. This article considers the largely untested assumptions within UK welfare-to-work policies and marketised employment services, which are designed to control and modify behaviour through compulsion and incentives. It examines those assumptions in relation to conceptualisations of human agency drawn from social policy literature. A gap is identified between accounts of agency grounded in the lived experiences of social actors (policy-makers, front-line workers and service users) and hypothetical models of individual agency (e.g. ‘rational economic man’) which have been more influential in policy design. It is argued that scope exists for understandings of agency to encompass the motivations, intentions and actions of all social actors involved in the policy process. This highlights the power dynamics of context creation, the universal potential for malevolence and the weight of moral significance. Conceptual and empirical insights point towards understanding the enactment of agency as relational, dynamic, differentiated, interconnected, interdependent, intersubjective and interactive.