This chapter analyzes the shift from the linear source analysis–transfer–target rendering model to an open, iterative human–AI collaborative process in which large language models function as cognitive partners. It details how dynamic task decomposition, multi‑round iterative refinement, and prompt engineering reframe translation as negotiated meaning construction. Core mechanisms include the Translate–Estimate–Refine (TEaR) loop, standards‑first plus feedback refinement, multi‑version generation, simulated adjudication and version fusion, and error‑prevention/risk‑control prompts. Lexical and knowledge layers are addressed through terminology governance, named entity standardization, handling of culture‑loaded terms, emergent expressions, idioms, and inference‑oriented plus retrieval‑augmented prompts that integrate terminology tables and external knowledge bases for controllability. For structurally complex discourse the chapter contrasts structure‑oriented, meaning‑decomposition, stepwise (comprehend–decompose–reconstruct–polish), role‑based collaborative, and back‑translation–driven quality assurance prompting strategies. A Tang poetry case study demonstrates progressive literal grounding, semantic interpretation, and stylistic transcreation across target literary registers, exemplifying human–AI co‑creative practice. The chapter argues that translator competencies must expand to encompass prompt design, AI interaction literacy, evaluative reasoning, data and standards governance, and ethical accountability. It proposes a layered process architecture coupling automated generation, diagnostic evaluation, and human judgment to enhance accuracy, consistency, cultural adaptation, transparency, and innovation for sustainable improvement and continuous quality optimization.