Industrial visual inspection operates at the intersection of perception, decision-making and physical consequences. Despite advances in vision and foundation models, industrial inspection remains fundamentally unsolved: inspected assets are domain-specific and evolving, labeled data is scarce, environments drift and inspection decisions must be made under uncertainty with safety constraints. We argue these challenges require a shift from single-pass perception to an agentic approach. Drawing on deployments including automated welding inspection, we identify the structural limits of classical pipelines and generic foundation models. We then introduce a human-centric, agentic physical AI framework that treats inspection as a closed-loop cognitive process, integrating memory, geometry-aware alignment, uncertainty-aware reasoning and safe adaptation. This architecture is device-agnostic and domain-specific, enabling inspection systems that scale across assets and environments while remaining interpretable, auditable and aligned with industrial practice.