A conceptual agentic AI architecture for MASLD-associated significant fibrosis in primary care

Metabolic dysfunction-associated steatotic liver disease (MASLD) is highly prevalent yet often underdiagnosed or undertreated in primary care due to asymptomatic early disease, uneven uptake of non‑invasive tests, limited elastography access, competing clinical priorities, and persistent challenges in sustaining lifestyle modification even after risk is recognized. This opinion introduces ATLAS‑Liver (Adaptive Triage and Learning Agent Suite for Liver disease) as a conceptual reference architecture, not a validated system, for how agentic artificial intelligence (AI) could support guideline‑aligned MASLD pathways by integrating risk estimation, explainability, calibration and fairness monitoring, curated guideline retrieval, and clinician‑retained decision authority within routine workflows. ATLAS‑Liver distinguishes between currently feasible components (e.g., probabilistic models using routine EHR data, local explanation layers with appropriate caveats, subgroup calibration checks, and version‑controlled guideline repositories) and aspirational elements such as dynamic retrieval‑augmented guidance, continuous drift surveillance, and automated agent‑level disagreement resolution. The framework is intended to complement established sequential pathways such as FIB‑4 followed by elastography rather than replace them, offering potential value through improved workflow integration, transparency, follow‑through coordination, and equity monitoring. We situate ATLAS‑Liver within emerging work on AI agents in chronic liver disease while emphasizing its primary‑care orientation and governance‑focused design. We outline key implementation considerations as well as patient‑facing needs such as explanation formats, communication preferences, and support for lifestyle adherence. We acknowledge substantial limitations including lack of empirical validation. ATLAS‑Liver is offered as a hypothesis‑generating framework to guide responsible exploration of agentic AI in primary care MASLD pathways.