Clarify definitions and capability boundaries of LLM-Agents, MLLM-Agents, and Agentic AI in smart manufacturing

Determine formal definitions, capability boundaries, and practical applications of Large Language Model-based agents (LLM-Agents), Multimodal Large Language Model-based agents (MLLM-Agents), and Agentic AI within smart manufacturing, specifying criteria that delineate each paradigm’s roles, limitations, and deployment contexts to resolve current conceptual ambiguity.

Background

The paper surveys GenAI-enabled AI agents—specifically LLM-Agents, MLLM-Agents—and the emerging paradigm of Agentic AI, highlighting their potential to transform manufacturing through semantic retrieval, multimodal reasoning, adaptive learning, and autonomous decision-making.

Despite these advances, the authors note persistent ambiguity around what precisely constitutes each paradigm in the manufacturing context, what capabilities and limits they possess, and how they should be practically applied. This unresolved conceptual clarity motivates the review and the call for more rigorous definitions and boundaries.