Negotiated Agency is a research-driven 8-week project experimenting on form of machine learning as a design material. Through a hermeneutic dynamic of optics and engagements of Programmatic Design Research it works with Design Ideal of Empowering people by allowing them to relate to technology of ML in a process of negotiating agency. Experiments and wonderments explore the dynamics between engaging people and easing lives and question the paradigm of empowerment in interaction design. This thesis does not deal with specific design problem but rather iteratively re-frame understanding of Negotiated Agency. Finally, the work proposes a framework and a set of appropriable qualities: Control Handover, Personalised Engagement, Authorisation Range, Temporal Consent, Recognising Intent and Interaction Styles. Each of the qualities is explained through practical examples, theory and experiments.