Artificial intelligence (AI) in mineral processing has developed almost entirely within a single methodological family: predictive models that convert measured inputs into numerical outputs. Since 2022, large language models (LLMs) and agentic AI systems have transformed adjacent fields, including chemistry, materials science, and geoscience, by enabling reasoning over unstructured evidence and autonomous multi-step action. Mineral processing has not yet adopted these capabilities. To the best of the author's knowledge, this paper is among the first to systematically position the AI trajectory in mineral processing against this broader trend, rather than treating LLMs as an incremental extension of existing predictive modeling workflows. The paper identifies five cross-cutting opportunity areas, each grounded in a documented precedent from an adjacent field: agentic multimodal process copilots, domain-specific foundation models, self-driving laboratories, retrieval-augmented knowledge synthesis, and language-enabled digital twins. The paper also identifies five structural barriers explaining the current adoption lag, including data fragmentation, an unresolved verification methodology, and a workforce gap. These findings are synthesized into a prioritized, phased research agenda. This agenda offers the mineral processing research community a concrete, actionable path into this underexploited capability space, rather than a general appeal to adopt AI.
