This paper reviews the key advances in the application of artificial intelligence (AI) for stroke rehabilitation, including quantitative assessment, personalized treatment prescription, and closed-loop training, and proposes pathways for clinical translation. Based on recent high-quality engineering and clinical evidence, we centered on assessment/diagnosis, therapeutic intervention, and remote management, integrating computer vision, brain-computer interfaces, exoskeletons, virtual/augmented reality, and edge-to-cloud platforms. We emphasized the alignment of "algorithm-engineering-clinical" evidence. On the assessment side, AI leverages multimodal sensing and imaging analytics to automate functional scale scoring and enhance prognostic precision. On the treatment side, AI empowers brain-computer interfaces and exoskeletons through intent decoding and adaptive control to deliver highly individualized closed-loop training that maximizes neuroplasticity. On the remote management side, AI-driven wearables combined with remote platforms enable continuous home-based monitoring and quality control. Despite challenges－including variable evidence quality, limited algorithmic generalization, and potential safety risks－AI shows substantial promise for deeper integration of rehabilitation care and management and for achieving individualized precision therapy.