Abstract: When advanced planning and scheduling (APS) models for production-logistics are applied under multi-source dynamic disturbances, changes in decision variables, modifications to objectives, and additions/reductions of constraints are inevitable. Such structural changes diminish the applicability of the original models or even render them ineffective, leading to frequent demands for model reconstruction. Addressing the challenges of high task complexity and difficulty in rapidly reusing knowledge, this study proposes large language model and agentic AI-driven reasoning methods for modelling of production-logistics APS. From three dimensions—reasoning architecture, domain knowledge, and reasoning capability, this study respectively proposes a multi-agentic AI collaborative reasoning architecture for modelling of production-logistics APS, a multi-step knowledge augmentation method for the production-logistics optimisation modelling domain, and a context engineering-driven enhancement method for agentic AI thinking and reasoning capabilities. This achieves orderly modelling reasoning, augmented modelling knowledge, and strengthened modelling units. Numerical experiments demonstrate that the proposed methods exhibit significant advantages in task completion rate and success rate for modelling tasks. They provide reliable assistance for APS participants addressing the need for optimisation model reconstruction in complex environments. This study offers an innovative perspective and approach for AI-assisted lean production-logistics management within smart manufacturing systems under Industry 4.0.