BONSAI: A Mixed-Initiative Workspace for Human-AI Co-Development of Visual Analytics Applications

Developing Visual Analytics (VA) applications is notoriously difficult, requiring tight integration of complex machine learning models with interactive interfaces. Developers face a painful trade-off: building rigid, tightly-coupled monoliths that are hard to maintain, or using simplistic frameworks that limit expressiveness. Single-shot AI code generation offers speed but produces unstructured, unauditable code. To solve this, researchers Thilo Spinner, Matthias Miller, Fabian Sperrle-Roth, and Mennatallah El-Assady created BONSAI, a mixed-initiative workspace that lets human and AI developers collaborate on VA applications within a structured framework.

BONSAI's key innovation is its modular four-layer architecture (hardware, services, orchestration, application) and a four-phase development process (plan, design, monitor, review). This ensures that all contributions—whether from humans or AI agents—are structurally bounded and fully tracked, providing complete provenance. The system was evaluated through case studies showing efficient creation of novel tools and rapid reconstruction of complex VA applications directly from research paper descriptions. This approach balances AI's generative speed with the structural rigor needed for complex VA development, offering a scalable solution for teams building data-intensive applications.