“Decolonial AI” treats artificial intelligence as both technical object and sociotechnical subject. Algorithms, datasets and applications cannot be separated from the institutions, value systems and power relations that shape their production and deployment. The paper’s central intervention is to transform decolonial theory from retrospective critique into foresight: historical patterns of extraction, hierarchy and epistemic exclusion become tools for anticipating future harms before they are stabilised in technical infrastructures. Three tactics organise the proposal—critical technical practice, reverse tutelage and the renewal of affective and political communities. These moves challenge the assumption that ethical correction can be added after design. Instead, impacted communities and suppressed knowledge traditions must alter research agendas, pedagogies and standards from within. Methodologically, the essay joins political history to technical governance, widening the aperture through which AI is understood. It bridges decolonial thought, computer science and technology ethics by defining justice not as bias mitigation alone, but as the redistribution of authority over what systems are built, whose worlds they encode and who bears their risks.