Effective digital twin solutions require a suitable digital representation of the environment within which physical systems operate. This representation serves as the foundational layer, enabling the digital twin to mirror the real world with a fidelity that ensures the precise alignment of virtual and real-world elements essential for accurate simulations, analyses, and decision-making.  From manufacturing to urban planning or defence, digital twin applications require a representation of the physical environment relevant to specific domain needs and challenges.  This digital representation must address the particular context of the digital twin – a production line, city, or mission area – and support the digital twin’s specific purpose – from operational and predictive analysis to testing and evaluation.  

The design and construction of digital twin environments require significant expertise, effort, and data.  This paper will consider how AI (Artificial Intelligence) agents can assist environment design and development processes within the context of digital twin solutions composed of platforms and systems requiring representations of terrain, features, entities, and behaviours that form a virtual physical and operational environment.

In just the last year, the utilisation of Large Language Models (LLMs) as the “thinking brain” underpinning AI agents has made extraordinary progress.  Remarkable advancements in LLM reasoning, understanding, creativity, task-specific knowledge, interaction, and multi-modal capabilities have led to the development of versatile LLM agent solutions that match and even exceed human performance, transforming many fields, from code development to content creation.  This paper builds upon earlier LLM agent architecture research to explore the practical application of the latest state-of-the-art LLM capabilities to the digital twin design lifecycle.  In particular, it will consider how agents can support, assist, and improve the human-in-the-loop design process by enhancing human expertise and performance with augmented, domain-specific LLM knowledge and how AI agents may apply this to orchestrate the development of digital twin environments.