Despite advances in theory, often driven by feminist ethicists, research ethics struggles in practice to adequately account for and respond to the agency and autonomy of people considered vulnerable in the research context. We argue that shifts within feminist research ethics scholarship to better characterise and respond to autonomy and agency can be bolstered by further grounding in discourses from the social sciences, in work that confirms the complex nature of human agency in contexts of structural and other sources of vulnerability. We discuss some of the core concepts and critiques emerging from the literature on women and children's agency in under-resourced settings, highlighting calls to move from individualistic to relational models of agency, and to recognise the ambiguous, value-laden, and heterogeneous nature of the concept. We then draw out what these conceptual shifts might mean for research ethics obligations and guidance, illustrating our analysis using a case vignette based on research ethics work conducted in South Africa. We conclude that if research practices are to be supportive of agency, it will be crucial to scrutinise the moral judgements which underpin accounts of agency, derive more situated definitions of and responses to agency, and enable people and participants to influence these based on their own experiences and self-perceptions.