Plastic pollution has become a pressing global environmental challenge. Polyethylene terephthalate (PET), one of the most widely used synthetic polymers, represents a major contributor to this problem. The development of efficient PET degradation strategies is therefore critical for advancing waste management and resource recovery. Recent studies have demonstrated that biocatalytic approaches, centered on PET-degrading enzymes, can enable closed-loop recycling of PET. As a result, the optimization of PET-degrading enzymes has become a central focus of research in this field. The interaction mechanisms between PET-degrading enzymes and their substrates have been elucidated, providing the foundation for diverse strategies in molecular engineering. Advances have been achieved through directed evolution, semi-rational design, rational design, and more recently, machine-learning-driven approaches. Notably, machine learning has emerged as a transformative tool that accelerates the design of enzymes with enhanced catalytic performance. Despite these advances, major challenges remain. Current PET-degrading enzymes display insufficient activity at low temperatures, limiting their utility in settings such as composting. Moreover, the depolymerization efficiency against highly crystalline PET remains low, hindering industrial-scale application. The convergence of machine learning and enzyme engineering is expected to be a key direction for overcoming these barriers, enabling the development of robust and efficient biocatalysts. Such progress would help break through the bottlenecks in the industrialization of PET biodegradation and promote the transition toward a sustainable circular economy.