Esophageal cancer is a major global health burden and cause of cancer death. Accurate assessment of tumor regression grade (TRG) after neoadjuvant therapy is essential to assess response to treatment and plan postoperative care. Traditional assessment of TRG depends on the subjective opinions of pathologists, which causes high variability and low reproducibility. The study developed an automatic system for TRG assessment using artificial intelligence and weakly supervised multi-level learning on digital pathology images. They analyzed 157 patients with esophageal cancer and 1,298 whole-slide images stained with hematoxylin and eosin. The system achieved a TRG classification accuracy of 82.7% and strong agreement with pathologists' assessment. AI TRG scores better predicted progression-free survival and overall survival than traditional methods. This approach reduces the burden on pathologists and improves prognosis in oncology.