GrassLobster is an experimental personal project by Miro Bannwart that connects the parametric environment of Rhino and Grasshopper with an external AI agent of the user's choice. The prototype enables designers to describe a building form, timber structure, or element pattern in natural language. The AI agent then guides the user through decisions regarding dimensions, rules, and materials until the task is clear enough to build a connected Grasshopper model.

Once generated, the parametric workflow remains fully adjustable; users can modify spans, spacing, or element counts while the geometry updates automatically. The system organizes models into "Geometry Stations" on the canvas, each representing a meaningful task like inputs, primary geometry, or structural logic. Station definitions and executable code are mirrored in readable project files, allowing the agent to locate logic and make targeted changes.

Project context is supplied via Agent Instructions, domain knowledge, and geometric references such as screenshots or example models. GrassLobster is designed around interchangeable agents rather than a built-in model, allowing the workflow to benefit as external agent capabilities improve. Future directions include agentic optimization, where agents could vary inputs to meet measurable goals like reducing material volume while maintaining floor area.