Model-based systems engineering (MBSE) with SysML is established practice for complex systems, but in software-intensive cyber-physical systems its specifications lose effect at the handover to software development: model and code are artifacts of different kinds, and the model lags the implementation it is meant to govern. SysML v2 provides a textual notation, so the model can be versioned like source code, and large language models (LLMs) generate working code from structured text. This single-author case study combines the two in a closed loop. Requirements and architecture are held in a SysML v2 model; the behavior it does not express is supplied by an annex of numbered decisions. Model, annex, the procedure the agent follows and the criteria a run must meet form an execution contract from which an LLM agent regenerates code and tests; what each run and its use reveal is returned to the contract, so the specification is completed while the system is being built. A provenance rule is proposed that keeps what the specification did not state visible and countable. The loop was run on a distributed flight-simulator prototype through four contract versions, which show how model and annex can govern LLM code generation and testing. Integration into CI/CD pipelines and multi-agent orchestration are identified as future work. Author’s submitted version. Submitted to Systems Engineering (Wiley/INCOSE) on 25 October 2026; under review. The published version, if accepted, will differ.
No takes yet. Share an insight, caveat, or question.
Gianluca Marsiglia (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: