You’ve seen the demos. You type a prompt, and boom—a perfectly rendered 3D model appears on your screen. It feels like magic. It feels like the end of traditional CAD as we know it.
But then you try to actually use the model. You try to export it. And reality hits you like a ton of bricks.
A 3D model without a STEP file is just a digital mirage.
Take Cartesian, the new AI 3D modeling tool making the rounds. The pitch is intoxicating: AI for architecture and product design. But the moment you land on their site, the illusion shatters. The scroll-jacking is so aggressive it feels like a bad Windows desktop app. And when you finally generate a model, what do you get? Triangles. Not parametric boundaries. Not STEP or IGES files. Just raw, uneditable meshes.
You cannot manufacture a triangle. You cannot constrain a mesh.
Until an AI understands parametric constraints, it isn’t designing products. It’s making video game assets.
This is the dirty secret of the current AI 3D boom. These tools aren’t competing with SolidWorks, Fusion 360, or AutoCAD. They’re competing with Blender and Unity. They are incredibly fast visualization toys, but they are functionally useless for actual engineering pipelines.
If you’re a maker or an engineer, you know the pain. You don’t need a pretty picture; you need a boundary representation (BRep) that you can modify, dimension, and send to a CNC machine or a 3D printer. When an LLM generates a mesh, it’s fundamentally misunderstanding what a CAD model actually is. A CAD model is a recipe of constraints and features. A mesh is just a pile of connected points.
We see this pattern over and over. A startup slaps “AI” onto a 3D viewer, claims it will revolutionize product design, and completely ignores the foundational engineering formats that have run the industry for decades.
Tech hype always promises to replace the engineer, but it usually just replaces the concept artist.
Some developers are getting it right. People are using tools like Astra inside Blender, or ForgeCAD to have Claude write scripts that generate printable 3D models. Why do these work? Because they respect the pipeline. They output formats that the physical world can actually interact with.
If you want to impress a designer, give them a pretty render. If you want to impress an engineer, give them a STEP file.
Until AI 3D generators figure out how to speak the language of parametric BRep data, their claims of revolutionizing architecture and product design are just marketing fluff. Stop wasting your time testing tools that don’t fit your professional workflow. The real revolution begins when AI can handle a dimensional constraint, not just a text prompt.
FAQ
Q: Can't you just convert an AI-generated mesh into a STEP file?
A: Technically yes, but it’s a nightmare. Mesh-to-BRep conversion is notoriously sloppy, often resulting in non-manifold geometry that breaks the moment you try to apply constraints or modifications. It’s a band-aid on a fundamental architectural flaw.
Q: Should engineers ignore AI 3D generation entirely?
A: No, but shift your expectations. Use these tools for rapid prototyping, concept art, or game assets. Don't try to force them into a manufacturing pipeline until they natively support parametric outputs.
Q: Is the AI 3D hype actually bad for the industry?
A: Yes. By marketing visualization toys as CAD replacements, startups are burning engineer goodwill. They’re creating a 'boy who cried wolf' scenario where actual breakthroughs in parametric AI will be ignored by a jaded engineering community.