This paper addresses the use of AI and Virtual Reality (VR) to enhance professional training in manufacturing. Industry 5.0 prioritizes human-centricity, sustainability, and resilience, requiring technologies like AI and VR to support these goals. Despite their value, current solutions rarely combine AI and VR in ways that effectively foster Human AI Teaming (HAT). The authors propose a three-level framework (Interface, Interaction, and Teaming) for designing AI-driven Digital Assistants (DAs) in VR training environments. The Teaming level defines AI roles and behaviours. The Interaction level shapes adaptive training journeys and learning exercises. The Interface level facilitates human-DA communication. This framework was applied to use case in an Italian manufacturing company, involving user research to define tasks, capture tacit knowledge, and design DA roles and learning interactions. The framework aims to contribute to the design of effective Human-AI experiences. Initial application shows promise in generating tailored training scenarios. Ongoing validation will assess usability and human factors, with future research exploring long-term trust dynamics and broader applicability across different manufacturing tasks and domains.