Abstract: With the growing application of generative artificial intelligence in translation, the traditional “translator–text” interaction model underpinning translator behavior theory faces unprecedented challenges. Using English translations of Chinese classics as illustrative cases, this article advances the thesis of a paradigm shift in translator behavior: translators now design, adjust, and optimize prompts to guide AI’s generative processes. This is not merely a technical operation but a new form of translator behavior with theoretical significance. The argument is developed across three dimensions: first, from the perspective of Frame Semantics, prompts externalize cultural cognitive frames into machine-readable instructions; second, from the perspective of Relevance Theory, prompts enable translators to optimize the balance between cognitive effects and processing effort; third, from a phenomenological perspective, prompts constitute the generative field of meaning, positioning translators as designers of emergent meanings. The study demonstrates that the design, implementation, and reflection of prompts form a complete behavioral chain in the AI era, redefining the translator as a “translation architect” who bridges human cognition and machine intelligence through metalinguistic agency. This paradigm shift not only responds to the challenges posed by technological transformation to the translation profession, but also offers a new theoretical framework for understanding and advancing translator competence in the age of AI.