Abstract for: Code-First System Dynamics: Building Validatable Models with AI and Python

This hands-on workshop introduces participants to a "code-first" workflow for system dynamics, leveraging Large Language Models (LLMs) to generate executable simulation code directly from natural language. Traditional modeling often requires manual translation between conceptual diagrams and software equations. In this session, participants will bypass manual diagramming to collaborate with AI in constructing models that conform to the Validatable Model Standard (VMS). Working in a Python-based environment, attendees will build several canonical archetypes—including the bathtub, SIR, and Bass diffusion models—using AI as a reasoning and implementation partner. We will demonstrate how complex feedback structures can be programmatically recovered from code to automatically generate traditional stock-flow diagrams, preserving the "glass-box" transparency essential to the field. This workshop is designed for those interested in integrating SD with modern data science ecosystems, version control (Git), and automated validation. Participants should bring a laptop or smartphone with internet access; no prior Python expertise is required.