This article examines the translator’s role in sustained AI-assisted translation and argues that it cannot be adequately captured by phase-specific activities such as pre-editing and post-editing or by prompt engineering alone. The study draws on the author’s experience of translating an approximately 400-page book on philosophy of religion from English into Persian over three months with sustained assistance from a generative AI system based on a large language model. The workflow was cumulative, with project instructions, prior agreements, the original author’s Persian notes, previously approved chapters, and ongoing human revisions remaining available to shape subsequent translation activity. The article identifies three interconnected scales of translator involvement: project-scale norm setting and accumulation of approved decisions, chapter-scale reactivation and mobilization of relevant contextual resources, and paragraph-scale feedback in which human revisions informed subsequent translation. Across these scales, earlier decisions and interactional conventions were progressively carried forward, allowing the workflow to stabilize over time. The case therefore points to a broader conception of the translator’s role as organizing the contextual and interactional conditions of sustained AI-assisted translation.