Frontend builders create UI prototypes to judge options, which is a time-consuming strategy of repeated iteration and refinement. Generative AI code assistants allow speedy prototyping just by prompting via a chat interface quite than writing code. Nonetheless, whereas this interplay provides builders flexibility since they’ll write any immediate they want, it makes it difficult to manage what’s generated. First, pure language by itself might be ambiguous, making it tough for builders to exactly talk their intentions. Second, the mannequin could reply unpredictably, requiring the developer to re-prompt via trial-and-error to restore any undesired adjustments. To deal with these weaknesses, we introduce Squire, a system designed for guided prototype exploration and refinement. In Squire, the developer incrementally builds a UI element tree by pointing and clicking on completely different options recommended by the system. Further affordances let the developer refine the looks of the focused UI. All interactions are explicitly scoped, with ensures on what parts of the UI will and won’t be mutated. The system is supported by a novel intermediate illustration referred to as SquireIR with language help for managed exploration and refinement. Via a consumer research the place 11 frontend builders used Squire to implement cellular net prototypes, we discover that builders successfully discover and iterate on completely different UI options with excessive ranges of perceived management. Builders moreover scored Squire positively for usability and common satisfaction. Our findings counsel the robust potential for code era to be managed in speedy UI prototyping instruments by combining chat with explicitly scoped affordances.
