Researchers propose AI4SE taxonomy to bridge model-driven engineering and generative AI

A comprehensive research effort has synthesized advances in artificial intelligence with model-driven software engineering (MDSE) practices, introducing a structured taxonomy to classify how AI can augment development workflows.

The Integration Challenge#

Model-driven software engineering has long offered proven benefits for managing complex systems, yet its real-world adoption faces significant hurdles. The core obstacle lies in substantial maintenance overhead for models combined with steep skill requirements that push many organisations away from the approach. Meanwhile, AI—particularly generative and agentic systems—has demonstrated remarkable potential for automating code-centric tasks including comprehension, generation, and defect identification. This gap between two powerful domains creates an opportunity for integration.

AI4SE Framework#

Researchers have developed the AI4SE taxonomy to map and connect diverse AI applications across software engineering disciplines. The framework aims to provide a shared vocabulary for understanding how machine learning systems can enhance traditional development practices. This classification structure serves as a foundation for discussing how AI can address the maintenance and skill challenges that have hindered MDSE adoption.

The Vision of "Big Models"#

Building on the taxonomy, the authors propose a concept called "big models" in software engineering—an approach designed to merge MDSE's structural strengths with the scalability that AI systems provide. This would leverage vast repositories of open-source code ("big code") that power modern AI training, combining the rigorous discipline of model-driven practices with data-driven automation.

Pair Modelling and Human-AI Partnership#

The research further outlines a pair modelling paradigm as a collaborative framework for the MDSE industry. This model envisions human developers and AI systems working in partnership to enhance software quality, drawing parallels to established practices like pair programming while extending the concept to architectural and design-level decisions.