Clarify Before You Draw: Proactive Agents for Robust Text-to-CAD Generation

A research team has introduced ProCAD, a proactive agentic framework designed to make text-to-CAD systems more robust by asking clarification questions before generating parametric code, according to a paper posted to the arXiv preprint server [1]. Large language models have enabled systems that synthesize parametric CAD programs, such as CadQuery scripts, directly from natural-language descriptions [1]. In practice, however, user prompts are often under-specified or internally inconsistent: critical dimensions may be missing and constraints can conflict [2]. Existing fine-tuned models tend to reactively follow user instructions and hallucinate dimensions when the text is ambiguous [2]. To address this, the researchers propose ProCAD, a framework that pairs a proactive clarifying agent with a CAD coding agent [1]. The clarifying agent audits the prompt and asks targeted clarification questions only when necessary to produce a self-consistent specification, which the coding agent then translates into an executable CadQuery program [2]. The team fine-tuned the coding agent on a curated high-quality text-to-CadQuery dataset and trained the clarifying agent via agentic supervised fine-tuning on clarification trajectories [2]. In experiments, ProCAD outperformed frontier closed-source models, including Claude Sonnet 4.5, reducing the mean Chamfer distance by 79.9% and lowering the invalidity ratio from 4.8% to 0.9% [1]. The paper, titled "Clarify Before You Draw: Proactive Agents for Robust Text-to-CAD Generation," was submitted by Bo Yuan and colleagues on 3 Feb 2026 and revised on 15 Jun 2026 [1]. The code and datasets have been made publicly available on GitHub [2]. The work arrives as arXiv, the open-access repository where the paper appears, continues to serve as a primary distribution channel for machine learning research [6]. arXiv hosts e-prints across mathematics, physics, computer science, and related fields, and has grown to a submission rate of about 24,000 articles per month as of November 2024 [6]. The repository is not peer-reviewed, but provides rapid dissemination of findings at no cost to readers and submitters [6]. The paper is accompanied by arXivLabs integrations, a framework launched in 2020 that allows community collaborators to develop experimental tools on the article page, such as bibliographic explorers and code finders [5].