This paper argues that there is little direct evidence for consistent livelihood understanding and analysis-informing humanitarian practice in Afghanistan in the past. The need for this is now greater than ever. The dynamics of the chronic conflict in Afghanistan has been poorly understood, not least in terms of its effects on livelihoods. Aid practice has been driven by simplified stories about the country reinforced through short-term humanitarian based programming that has emphasised delivery and paid little attention to learning. The result has been a monotonous landscape of interventions. Three case studies – on opium production, an economic blockade in Hazarajat, and carpet production in Northern Afghanistan – illustrate the complexity of livelihoods and the dynamics of power relations and the relevance of this understanding to programming. Key lessons drawn from these studies include: the recognition of embedded knowledge of nongovernmental organisations (NGOs) that have worked long-term in specific locations and the need to build learning explicitly into their programming; the poor understanding of the resilience of livelihoods; the need to recognise the legitimacy of ‘illicit’ activities; the dynamic nature of conflict and power relations and the poor conceptualisation of vulnerability within the livelihoods framework.