Agentic AI in Latin America: characterization of the structural gap and proposal of a contextual adaptation framework

Agentic Artificial Intelligence (AI) systems are rapidly emerging as a dominant paradigm in applied AI. However, existing architectures are developed and evaluated predominantly in high-income, English-speaking, and well-connected contexts, embedding structural assumptions that do not apply to much of Latin America and the Caribbean (LAC). This article addresses the absence of a systematic characterization of the contrast between agentic AI architectures and the multidimensional constraints of the LAC operational context. To this end, a data-driven conceptual framework development methodology is employed, through which four structural dimensions are identified: (1) digital infrastructure heterogeneity, (2) linguistic marginalization of Spanish, Portuguese, and indigenous languages, (3) regulatory fragmentation, and (4) asymmetric access and adoption patterns. Building on this characterization, the LAC Contextual Adaptation Framework (LATAM-CAF) is proposed, a three-layer architecture comprising context profiling, an adaptive execution engine, and a contextual evaluation suite. The results suggest that deploying standard agent architectures in Latin America without contextual adaptation risks replicating and amplifying existing structural inequalities.