<p>The growth of electric vehicles leads to an increase in end-of-life battery systems, requiring scalable and safe disassembly processes for sustainable recycling. While prior research has advanced perception, planning, and robotic execution, experimentally validated integrated systems capable of handling real-world uncertainty remain limited. This study applies Design Science to develop and evaluate an AI-assisted semi-automated disassembly system combining optical perception, adaptive robotics, and modular process knowledge. The proposed system integrates 2D segmentation and 3D point cloud analysis to identify components and estimate their 6D pose, achieving Dice scores of up to 95.5% and a mean average precision of 63.4% for rotation estimation. Experimental validation demonstrates successful robotic unscrewing rates of up to 98.8% and reliable execution of integrated disassembly sequences in most test cases. However, results also show that fully autonomous operation remains limited under real-world uncertainty, particularly due to perception variability and environmental influences. Findings indicate that robust disassembly requires hybrid human–AI system architectures, positioning human–robot interaction as a structural design principle rather than a fallback solution. This extends research by conceptualizing AI-assisted disassembly as a human–AI socio-technical system for managing uncertainty. It demonstrates potential for improved safety, increased flexibility, and scalable battery recycling operations in practice.</p> Graphical Abstract <p></p>