This paper explores AI-human collaboration in Sudoku puzzle generation by simulating five distinct co-creative strategies without relying on real-time user input. While prior work in AI-assisted puzzle generation has focused on static outputs or human-in-the-loop systems, this study introduces a rule-based human agent to systematically evaluate interaction dynamics. The framework integrates AI solvers with a simulated human model, using metrics such as difficulty, symmetry, and conflict resolution to assess puzzle quality. Preliminary experiments reveal that hybrid strategies, especially turn-based and negotiation-based collaboration, yield more balanced and aesthetically pleasing puzzles than AI-only approaches. While the findings provide initial insights into procedural content blending technical solvability with human-like creativity, this study represents an initial step toward modeling collaborative puzzle design. Future research should incorporate real human participant studies and incorporate game design heuristics to enrich evaluation beyond algorithmic measures.