Artificial intelligence–enabled pharmacogenomics promises to extend precision prescribing by integrating genomic variation with increasingly rich clinical information. Its validity, however, depends on more than predictive performance. Pharmacogenomic evidence has been generated from populations, variant catalogs, assays, and implementation environments that differ substantially in genomic representation, while the clinical meaning of genotype may change as co-medications, disease state, organ function, and treatment history alter drug exposure and phenotype. This article develops a normative and methodological account of precision prescribing under these conditions. It separates population representation from individual ancestry inference, ancestry from race, germline genotype from current metabolic phenotype, and prediction from clinical action. Unequal genomic representation is treated as an evidentiary problem that affects which variants are discovered, characterized, measured, and validated, whereas polypharmacy and changing clinical context are treated as interpretation problems that can modify the translation from genotype to expected drug response. The resulting argument is that representation-aware pharmacogenomic systems require explicit evidence provenance, patient-level genomic characterization, context-sensitive phenotype updating, uncertainty handling, and safeguards against extrapolation beyond adequately characterized populations. Equity therefore cannot be evaluated only after deployment or reduced to average predictive parity.  It must be considered across evidence generation, variant interpretation, clinical decision support, monitoring, and the distribution of uncertainty itself. . A representation-aware approach can strengthen AI-enabled pharmacogenomic prescribing only when ancestry, genomic variation, medication context, and clinical change remain analytically distinct and are recombined at the point of individual treatment decision.