Dominik Bork

Generative AI enables rapid “vibe coding" and agentic software engineering, where natural language prompts yield working software systems. This lowers barriers to software creation, but it also collapses the boundary between prototypes and engineered software. Systems that appear complete may lack robustness, security, and maintainability. We argue that this shift motivates a renewed role for software models. Rather than serving only as upfront blueprints, models can be recovered from AI-generated systems, used to restore comprehension, and refined to guide subsequent evolution. We distinguish an agentic loop, in which recovered models support automated checking and repair, from a human reflective loop, in which models expose assumptions for inspection and revision. We identify candidate model types and illustrate how recovered models can expose implicit assumptions as possible constraints for review and validation. This paper positions software models as a bridge between current software engineering practice and model-based reasoning in AI-driven development.

Model-Driven Engineering (MDE) offers strong abstractions but remains hindered in practice by complex tooling and steep learning curves. Recent advances in Large Language Models (LLMs) promise intuitive, natural-language interaction with modeling environments; however, existing approaches—such as direct API interaction, DSL generation, or manipulation of serialized models—exhibit limited reliability, scalability, and semantic precision. This paper presents a structured integration of LLMs into graphical modeling environments via the Model Context Protocol (MCP). We design and implement a reusable MCP server for the Graphical Language Server Platform (GLSP), enabling LLM agents to interact with models through semantically grounded, tool-based operations rather than direct model generation. We contribute (i) a reference architecture for MCP-based integration into GLSP derived from a systematic literature review of MCP implementations, (ii) design principles for mapping modeling concepts to MCP primitives and abstraction levels, and (iii) an empirical evaluation assessing the impact of MCP-based interaction on downstream modeling tasks. Our results show that MCP-based interactions significantly improve both information retrieval (perfect precision/recall across tasks) and model manipulation (reducing syntactic, semantic, and pragmatic errors), while also improving scalability and efficiency compared to serialization-based approaches. The MCP server enables robust modification of existing models, a key limitation of prior approaches. These findings demonstrate that structured, tool-mediated interaction is a crucial enabler for reliable AI-assisted modeling. The presented MCP–GLSP Server constitutes a reusable artifact applicable across GLSP-based tools, providing a practical foundation to advance intelligent modeling assistance.