AI CAD Is Getting Good at Demos. Manufacturing Is a Different Test.

No FreeCAD learning curve. No $8,000/year CAD licenses. Just type what you want and get a part.

That promise is why text-to-CAD tools have exploded over the past year.

But most comparisons stop at the demo: a bracket appears, a gear appears, a flange appears, and the conclusion is that AI CAD has arrived.

That is not the test that matters.

The real question is simpler: can you take the output, open it in a normal CAD workflow, and send it to a machine shop without creating extra cleanup work for yourself?

That is where the differences between tools become much more obvious.

The Manufacturing Gap Most Reviews Skip

Most text-to-CAD reviews evaluate visual success. Did the model look right? Did it resemble the prompt? Did it produce something 3D?

Those are useful questions, but they are not enough for engineering.

A part that looks right in a browser preview can still be the wrong geometry for manufacturing. It may be hard to edit, hard to dimension, hard to tolerance, or painful to use downstream in CAM, simulation, or inspection workflows.

The biggest reason is a distinction most non-CAD users never hear about: mesh vs. exact CAD/BREP geometry.

Mesh vs. BREP: Why It Matters

Mesh: Good for Visualization, Fine for Some Prototyping

Mesh models such as STL or OBJ are made from triangles. They approximate shape by tiling the surface with flat facets.

That is often perfectly acceptable for:

concept visualization

quick prototypes

hobbyist 3D printing

artistic or organic forms

But mesh becomes a weak handoff format once you need precision workflows.

Why?

Editing is harder. Changing a hole size or moving a feature is much easier on exact CAD than on mesh.
Dimensioning is weaker. Triangle meshes do not carry the same exact analytic faces and edges that CAD workflows prefer.
Downstream interoperability suffers. Importing mesh into Fusion, SolidWorks, or other MCAD tools is usually less pleasant than importing exact geometry.
Manufacturing prep often becomes extra work. Some CAM workflows can use mesh, but exact CAD is usually the better starting point for machining and engineering change.

Mesh is not useless. It is just not the same thing as production-friendly CAD.

BREP / Exact CAD: What Engineering Workflows Prefer

BREP, or boundary representation, describes geometry using mathematical surfaces and trimmed faces: planes, cylinders, cones, splines, fillets, and so on.

That matters because exact CAD geometry is what most professional downstream tools expect.

Benefits include:

better neutral exchange through STEP

cleaner import into mainstream CAD

more reliable measurement from exact faces and edges

better compatibility with CAM, analysis, and inspection workflows

A circle is represented as a real circle. A cylindrical bore is a cylinder. A fillet is an exact surface, not a faceted approximation.

That does not mean every STEP file preserves a full feature tree or perfect parametric intent. It usually does not. But it is still far better than starting from triangle soup when the goal is manufacturing.

How We Evaluated the Current Tools

We looked at three practical questions:

Output format

Does the tool produce exact CAD-friendly output such as STEP, or mainly mesh?

Accuracy

Does the generated part actually match the requested dimensions and relationships?

Editability / workflow fit

Can the result be iterated on in a normal engineering workflow, or do you end up starting over?

How the Major Tools Compare

Zoo

Output: STEP, STL, KCL Best for: Simple mechanical parts, concept generation, fast parametric iteration

Zoo deserves credit for taking engineering-style CAD generation seriously early. Its parametric workflow is useful, and the ability to output STEP makes it much more relevant than pure mesh tools if you care about real CAD handoff.

Where it works well:

brackets

flanges

prismatic parts

simple mechanical forms

quick iteration on dimension-driven concepts

Where it still struggles, in our testing:

higher-complexity parts

precise gear geometry

internal flow features

geometry requiring more robust constraint handling

Zoo is real, useful, and ahead of a lot of consumer-style text-to-3D tools. But it still performs best in the simpler end of the part spectrum.

AdamCAD

Output: Focused on rapid generative part creation; export capabilities appear to be evolving Best for: Fast prototyping, conceptual part generation, 3D-print-first workflows

AdamCAD is compelling because it makes shape generation easy and approachable. It is good at helping users get from idea to printable concept quickly.

Its strength is not deep manufacturing workflow integration. It is speed and accessibility.

That makes it a good fit for:

desktop prototyping

early concept exploration

educational use

makers who want printable parts quickly

A notable recent development is that Adam has introduced an Onshape extension that works directly inside the application. Instead of only living as a standalone concept-generation experience, Adam is now moving closer to an in-workflow CAD copilot model for engineers already working in Onshape.

That matters because it shifts the product closer to real CAD refinement inside an existing CAD environment. The Onshape integration is positioned around helping organize and clean up feature trees, parameterize designs, and streamline modeling operations inside Part Studios rather than simply generating one-shot shapes in isolation.

If your workflow ends at STL and a printer, Adam may be enough. If your workflow continues into machine shops, tolerancing, CAD revision control, or exact downstream engineering, you still need to look closely at the current export and interoperability story before choosing it.

CADGPT

Output: CAD assistance, scripting, drafting support, code generation Best for: AutoCAD users and CAD automation

CADGPT is better understood as an AI CAD assistant than as a true text-to-3D part generator.

It can help with:

drafting help

scripting

CAD-related calculations

automation support

speeding up traditional CAD tasks

That is useful. But it is a different category.

If you want “describe a part and get a manufacturable 3D model,” CADGPT is not really competing in the same lane as the dedicated text-to-CAD tools.

Leo AI

Output: Engineering copilot functionality around CAD/PDM/PLM workflows, parts, documentation, and engineering reasoning Best for: Mechanical engineering support workflows, not native text-to-CAD generation

Leo AI is one of the more interesting products in the space, but it is often described incorrectly.

Leo is less a text-to-CAD generator and more an engineering copilot. It helps around the CAD process: requirements, assemblies, part selection, calculations, documentation, and engineering support work.

That can be very valuable, especially for:

mechanism-heavy projects

design context and requirements

part lookup and selection

assembly-adjacent workflows

engineering team productivity

But if you are comparing pure “type prompt → receive exact CAD geometry” systems, Leo belongs in a different bucket.

What About Henqo?

Full disclosure: we built Henqo.

Output: STEP Architecture: Build123d on OpenCASCADE Goal: Manufacturing-friendly part generation from the beginning

Henqo takes a different approach from mesh-first systems.

Instead of generating approximate 3D shape and worrying about manufacturing later, we generate exact CAD geometry through a BREP-native stack. That means the output is designed to fit engineering workflows from the start.

Our focus is not “make any shape that looks cool in a preview.”
Our focus is:

exact geometry

STEP output

editability

manufacturing-oriented parts

validation-oriented generation

We are not claiming AI has solved all of CAD. It has not.

Complex surfacing, unusual organic forms, advanced assemblies, edge-case mechanisms, and many expert workflows still benefit from experienced human CAD work.

But for the large class of real-world hardware parts that are mostly brackets, mounts, enclosures, flanges, plates, holes, bosses, ribs, and cut features, we think BREP-first is the right architecture.

So Which Tool Should You Use?

The answer depends on what you are actually building.

If you are prototyping for a 3D printer

Mesh-first tools may be perfectly fine. Speed matters more than exact CAD handoff, and STL may be all you need.

If you are sending parts to a machine shop

You should strongly prefer tools that produce exact CAD-friendly output such as STEP.

If you are integrating with Fusion, SolidWorks, or another CAD workflow

BREP-native output has a major advantage. Even when neutral exchange is imperfect, it is still far better than forcing mesh through an engineering workflow.

If you are building complex assemblies and mechanisms

No tool fully solves this yet. Some tools help with pieces of the workflow, but human engineering judgment is still doing a lot of the work.

The Honest Take on AI CAD Right Now

After testing the current generation of tools, our view is this:

Simple parts are increasingly viable.

Brackets, flanges, plates, bosses, bolt patterns, and many prismatic parts are now within reach.

Complexity still breaks systems quickly.

Precise relationships, internal geometry, gear quality, and constraint-heavy parts are where failure rates rise fast.

Format matters more than most marketing suggests.

A pretty preview is not the same thing as manufacturing-friendly output.

BREP-native systems have a real structural advantage for engineering workflows.

Exact geometry is simply a better starting point than mesh when the end goal is manufacturing.

Iteration is still part of the process.

The win is not one-shot perfection. The win is getting to a correct, editable result faster.

AI CAD is real. It is useful. It is improving fast.

But if your goal is to actually manufacture parts, the question is not just whether the tool can generate geometry.

It is whether it can generate the right kind of geometry for the rest of the workflow.