Artificial intelligence (AI) and machine vision have advanced optical quality inspection by enabling automated detection of surface anomalies with high precision and speed. Nonetheless, several unresolved challenges limit deployment in industrial environments, including reliance on extensive defect datasets, limited adaptability to rare or unseen defect types, and the lack of closed-loop mechanisms to connect defect identification with corrective actions. This paper proposes a novel end-to-end framework that integrates vision, language, and action (VLA) to address these gaps and transform inspection results into actionable remediation. The framework introduces four capabilities: (1) customized vision backbones with confidence quantification, (2) continual learning to sustain performance across new defect classes, (3) natural language and touchscreen interfaces for on-floor data labeling, and (4) AI agents to infer root causes and recommend corrective measures. Unlike conventional inspection methods that stop at defect detection, the proposed system closes the loop from image acquisition to remediation guidance. The approach is validated in collaboration with industry on flexible electronic display inspection, a task characterized by reflective surfaces and low defect-to-surface ratios. Results demonstrate accurate defect detection, robust generalization to rare defect types, and successful validation in collaboration with an industry partner on flexible electronic display inspection. These outcomes confirm the practical viability of the proposed end-to-end framework and demonstrate its potential to redefine optical quality inspection by transforming raw visual data into actionable remediation within real manufacturing environments.
