AI CAD and Specialized Vertical AI Tools Keep Launching
Adam, a YC W25 company, launched an open-source AI CAD tool on HN. The discussion quickly got technical, asking about face and edge selection UI, MCP integration for driving generation through an existing LLM, and whether OnShape or Fusion plugins were the right distribution vector. The tone was genuinely curious rather than dismissive, which is notable given how many AI-plus-vertical-tool launches get skeptical receptions.
The pattern here is the continued proliferation of AI tools aimed at specific professional workflows that are genuinely underserved by general-purpose models. CAD is a good example: the geometry constraints, parametric relationships, and export formats are specialized enough that bolting GPT-4 onto a generic UI doesn't work. You need domain-specific training and a domain-specific interface.
This connects to the local Qwen discussion and the Midjourney Medical thread. The builders getting traction are the ones treating AI as a component in a domain-specific system, not as the product itself.
So what?
The window for building AI-native vertical tools in professional workflows is still open, but it's closing as incumbents like Autodesk and Dassault add AI features. The Adam launch suggests the competitive advantage comes from open-source distribution and deep integration with existing file formats and tools, not from having a better model. If you're building in a professional vertical, your moat is domain knowledge and workflow fit, not the AI itself.