Abstract for: Code-First System Dynamics: Teaching Model Construction Through AI and Python

This virtual workshop introduced a "code-first" pedagogical approach to system dynamics, using Artificial Intelligence to lower the technical barriers to entry for simulation modeling. Traditionally, students must manually translate conceptual feedback into graphical diagrams before execution. In this session, we demonstrate a workflow where learners use Large Language Models (LLMs) to generate Python code directly from natural language problem descriptions, adhering to the Validatable Model Standard (VMS). Participants will engage in a collaborative "language sprint" to build canonical archetypes, such as the SIR and Bass Diffusion models, in a cloud-based Python environment. We will specifically focus on "structural recovery"—using automated tools to derive stock-flow diagrams from the student's code to ensure conceptual transparency is maintained. This workshop is ideal for educators looking to integrate system dynamics into data science or programming curricula. Participants should bring a laptop with a web browser; no prior coding experience is required.