LEGOUI makes AI UI design transparent and controllable

A team of researchers from CSIRO's Data61, UNSW Sydney, and other institutions has developed LEGOUI, a novel framework designed to address the opacity and lack of control in generative AI tools for user interface (UI) design. Published as a preprint on arXiv (arXiv:2608.04293), LEGOUI introduces a staged, interpretable approach that breaks down the UI design process into sequential steps, each captured in a UI domain-specific language (UI-DSL) enriched with provenance data. This method exposes the model's intermediate reasoning, enabling designers to intervene and refine outputs iteratively rather than relying on trial-and-error prompt adjustments.

In technical evaluations involving 40 real-world design prompts, LEGOUI achieved over 95% accuracy in capturing explicit design requirements, with near-complete coverage and zero redundancy. User studies further validated the framework, with participants reporting significantly higher levels of transparency, controllability, and alignment with their design intent compared to traditional one-shot generative UI tools. The framework represents a shift toward more interpretable and user-guided AI-assisted design processes.