The use of artificial intelligence (AI) in sport is often presented as a technical advance, yet its coaching value depends on a simpler question: can the result be explained in movement terms? This review revisits current AI applications in sport from a biomechanical perspective. It considers computer vision, markerless motion capture, wearable monitoring, injury-risk modeling, explainable AI, coordination theory, muscle synergy research, and sport-specific movement studies. The central argument is that a useful sport AI system should not stop at prediction. It should show how the athlete produced the performance through segment coordination, joint loading, postural control, and energy transfer. To organize this argument, the paper proposes a four-layer model: capture, interpretation, decision, and intervention. The model places AI between measurement and coaching, while biomechanical theory remains the foundation for deciding which variables matter. The review also suggests that personal sport records become meaningful only when linked to validated movement variables and coach-readable explanations. In this view, the next step for AI in sport is not merely larger datasets, but more interpretable, task-specific, and coordination-aware feedback.
