Position: Generative AI in Precision Nutrition Should Prioritize Standards at System Boundaries, Not Convergence on a Single Reference Architecture

Generative AI (GenAI) is extending precision nutrition (PN) beyond task-specific AI through knowledge synthesis, multi-modal interaction, tool use, and multi-step workflows, including (multi-)agentic systems. However, adoption in practice remains limited relative to the rapid progress in research. We argue that convergence on a single reference architecture is unlikely to drive the adoption of GenAI in PN. Instead, we identify an under-addressed constraint: fragmentation at system boundaries connecting the pathway from measurement to understanding, action, and evaluation. GenAI, particularly (multi-)agentic systems, may compound this by retrieving and generating information, invoking tools, and chaining actions across heterogeneous components. We therefore propose architecture-independent boundary standards that preserve meaning, provenance, uncertainty, temporal context, knowledge applicability, authority, and evidence as information and decisions move across this pathway. We operationalize these through versioned conformance profiles, cross-system testing, and a risk-proportionate pathway from conformance and workflow validation to longitudinal evaluation and lifecycle monitoring.