Sam Pasupalak, the founder of Maluuba (later acquired by Microsoft), regretted selling his company and established Skyfall AI in San Francisco in April 2025. Eighteen months later, the company launched its first product, SKY-001, positioned as an 'engineering world model,' primarily targeting physical world design and manufacturing scenarios in the CAD and BIM fields. Addressing the inefficiency and geometric misalignment issues of traditional large language models in handling spatial engineering relationships, SKY-001 adopts a two-stage architecture featuring a hierarchical componentizer and a world model builder. This architecture simulates the hierarchical decomposition and collaboration logic of engineering teams, first breaking down a complete design tree and then planning an ordered sequence of visual states for part manufacturing and drawing creation. According to self-test data released by Skyfall AI, SKY-001 demonstrates a speed improvement of tens to hundreds of times and a cost reduction of over a hundred times compared to competitors like GPT in CAD and BIM-related tasks, with superior performance in relevant metrics. This advantage stems from the high-granularity engineering dataset cleaned and reconstructed by the team, creating a significant data barrier. However, the product still has some issues, such as a gap between generated results and industrial-grade usability standards, a lack of third-party reproduction and validation for relevant performance data, and the absence of a closed-loop for parametric modeling. The team plans to extend related capabilities to more hardware fields, such as PCB design, aiming to create an AI chief engineer capable of supporting autonomous manufacturing.