Practical adoption of Round-Trip Engineering (RTE) is limited by synchronization complexity, manual intervention, process uncertainty, and artifact multimodality. We propose a novel agentic framework using Large Language Models (LLMs) to automate and enhance RTE. The framework employs specialized AI agents for tasks like transformation, reverse-engineering, and verification, all operating within a human-in-the-loop (HITL) process that enables iterative refinement. We demonstrate the framework’s feasibility with a proof-of-concept that transforms natural language requirements into UML Class Diagrams. A qualitative evaluation reveals that the iterative HITL refinement loop consistently yields the most significant quality improvements. Notably, this human-agent back-feed loop collaboration allows smaller LLMs to achieve high-quality results comparable to larger models after just a few cycles. This work provides a foundation for a new generation of intelligent, automated, and human-centric software artifact management tools.