This practice design case study documents a portfolio-based assessment that treats artificial intelligence (AI) as bounded scaffolding in university language education. It adopts a critical stance on “artificial intelligence,” noting that much current functionality is better described as artificial imitation and is subject to hallucinations, bias, and uneven language coverage. The intervention targets B1 learners of German at a UK university who were asked to write a 500-word essay through a three-stage sequence: (1) idea generation and planning; (2) dictionary-mediated drafting (e.g., Collins/Oxford, LEO, Linguee, Reverso); and (3) transparent use of LLM-enabled tools for revision (predictive text, DeepL Write), with process documentation and reflective commentary throughout. Wholesale English to German machine translation was prohibited to preserve authorship and process visibility. The case will be framed within context as routines for students of German at said institution explicitly develop pragmatic competence (register, genre conventions, speech acts) in authentic tasks such as staff email correspondence, and extend to genre-specific homework with style/tonal calibration. The discussion links the design to AI literacy, employability-relevant skills (editing, verification, register control), and the need for ethically grounded assessment and sustained professional development so that policy, pedagogy, and student agency remain aligned.