AI as Consumer and Participant: A Co-Design Agenda for MBSE Substrates and Methodology

Siyuan Ji's new paper, "AI as Consumer and Participant: A Co-Design Agenda for MBSE Substrates and Methodology," published on arXiv, identifies a critical flaw in current Model-Based Systems Engineering (MBSE) practices: AI tools are consuming models that were never designed for machine consumption. The problem isn't just hallucination; even well-prompted frontier models produce competent outputs over a conformant SysML model, but their reasoning is drawn from training data, not the model itself. Different tools analyzing the same model yield different results, with no record to adjudicate between them. The model, Ji argues, is functioning as a prompt rather than a knowledge base, and simply attaching better tools won't fix this.

To address this, Ji proposes a co-design agenda where the model and the methodology governing its construction are designed together for AI participation. The model must be treated as a machine-queryable knowledge substrate, not a structured artifact for human navigation. The paper works through a concrete workflow scenario to illustrate the current gap and proposes three principles that jointly characterize what model and methodology must achieve. Ji concludes with a call to the community to begin this work before architectural decisions about AI integration settle without the necessary methodological foundation.