Claims about artificial intelligence in pharmacy frequently combine conceptual proposals, retrospective models, and operational systems, making it difficult to determine what work has actually been assigned to AI and where pharmacist judgment remains necessary. To map AI-supported tasks across dispensing, prescription and clinical review, medicines information, patient-facing services, and administrative pharmacy work while separating technical performance from workflow usefulness, safety, and autonomy. Peer-reviewed Q1 journal articles published from 2017 through 2025 were eligible when they evaluated an AI or hybrid AI component performing or supporting an identifiable pharmacy task and provided extractable evidence on setting, data, reference standard, human involvement, validation, comparator, or outcome. Searches covered MEDLINE/PubMed, Embase, Scopus, Web of Science, CINAHL, International Pharmaceutical Abstracts, IEEE Xplore, and supplementary citation and publisher searching. Evidence was charted by task, AI method, setting, data source, reference standard, pharmacist role, validation level, comparator, outcome, and autonomy, then synthesized using a review-derived task–maturity–oversight framework. Searches identified 1,266 records. After removal of 472 duplicates, 794 records were screened; 153 reports were sought, 149 full texts were assessed, and 22 studies were included. Evidence concentrated on prescription prioritization, intervention prediction, anomaly detection, pill recognition, targeted patient services, and medicines-information answering. Pharmacists generally retained contextual review, authorization, communication, or final action. AI has primarily taken on bounded tasks that rank, flag, recognize, retrieve, or draft. The evidence does not establish autonomous completion of comprehensive pharmacy work or permit technical accuracy to be treated as proof of safety, benefit, or deployment readiness.