Large Language Models (LLMs), such as ChatGPT, are increasingly applied across disciplines for their adaptability and reasoning capabilities. However, their use as autonomous agents in simulations addressing sustainability and recycling remains underexplored. This study introduces a hybrid NetLogo–Python simulation where GPT-based agents replace traditional rule-based logic within a dynamic recycling environment. This study contributions are threefold: (1) the integration of LLM-based agents into a resource exchange simulation framework, (2) comparative scenariobased experiments assessing behavioral differences between GPT and rule-based agents, and (3) the implementation of visual analytics to enhance outcome interpretability. Results show that GPT agents enable context-aware, adaptive decision-making, though with trade-offs in predictability and controllability. Accordingly, the study explores emerging opportunities, highlights key challenges, and provides recommendations to guide future research. This approach advances the integration of LLM-Agent into agent-based models and offers new directions for decision support in complex sustainability and circular economy systems.