The 4 Ways AI Is Transforming CAD – and the Cloud-Native Advantage

AI in CAD is developing in four primary areas: AI advisors that answer questions; discrete AI tools that automate specific tasks; AI agents that perform work within CAD systems; and text-to-code-to-CAD systems that use AI to create new, reusable CAD capabilities.

These approaches are fundamentally different. Some help engineers use CAD. Some automate individual tasks. Some operate CAD on the engineer's behalf. And the newest approaches can actually help engineers customize the CAD system itself.

There is another important part to this story. Across these four areas, cloud-native CAD architecture provides some significant advantages for AI — from keeping AI knowledge synchronized with the latest version of the software to enabling parallel agent workflows, traceability, permissions, and company-wide sharing of new CAD capabilities.

Here's how the landscape breaks down.

1. AI Advisors: The Ultimate Engineering Reference

An AI Advisor for CAD combines natural-language interaction with trusted knowledge about the CAD system.

Think of it as the ultimate reference manual with a conversational interface.

Instead of searching documentation, digging through forums, watching videos, or trying to remember where an answer was found, an engineer simply asks a question.

How do I do this? Why does this work this way? What's the best approach?

The AI interprets the question, searches the knowledge available to it, and provides an answer.

At Onshape, the implementation of this concept is Onshape AI Advisor, which draws from sources including Onshape Help, Learning Center content, technical blogs, and other Onshape resources.

The Cloud-Native Advantage

This is one of the first places where cloud-native architecture offers a clear advantage.

Traditionally installed CAD systems can exist across many releases, service packs, and configurations. An AI Advisor must account for those differences before it can reliably explain what capabilities are available or how something works.

Onshape doesn't have that problem.

There is a single production version of Onshape, and everyone uses it. AI Advisor can be updated alongside the product, keeping its knowledge synchronized with the software customers actually use.

There isn't a question of whether an answer applies to an older release or a different service pack.

Everyone is on the latest version.

2. Discrete AI Tools: AI Applied to Specific Engineering Tasks

The second category is discrete AI tools.

These are purpose-built capabilities that apply AI or machine learning to a specific engineering task.

One example in Onshape is AI Quick Render, currently in Early Visibility.

Quick Render uses a diffusion model to take CAD geometry and rapidly create sophisticated rendered imagery. The result doesn't have to be photorealistic. Users can describe the desired appearance, ranging from traditional product visualization to more stylized or illustrated results.

Another example already in Onshape is Replicate Annotations.

When creating drawings for similar parts, Onshape can examine annotations associated with one drawing view, look at the geometry in another view and intelligently determine where those annotations belong.

These are very different applications of AI, and that's exactly what makes this category interesting.

There are hundreds of specific engineering tasks where AI can save time without fundamentally changing how engineers work.

The Cloud-Native Advantage

Cloud-native architecture isn't required to deliver discrete AI tools. An installed CAD application can certainly connect to AI services running in the cloud.

But a cloud-native platform can make these capabilities easier to deploy and evolve because the CAD application, engineering data, and AI services are already connected through a continuously updated platform.

There is also another advantage that becomes increasingly important as AI takes on larger tasks:

The work doesn't always have to happen sequentially.

AI can perform work while the engineer continues working.

That becomes much more important with agents.

3. AI Agents: From Answering Questions to Doing Work

Advisors provide information. AI agents can actually perform work.

An AI CAD agent understands the capabilities available within the CAD system and can use them to accomplish a task.

Instead of manually executing a sequence of commands, an engineer can describe the desired outcome in natural language.

An agent might interrogate metadata, modify parameters, analyze a model, export data, check a drawing, or execute a sequence of CAD operations.

But the bigger opportunity comes when moving beyond a single agent.

Onshape has demonstrated a vision involving multiple specialized agents coordinated through an orchestration layer.

There might be a part modeling agent, an assembly agent, a drawing agent, a drawing-checking agent, a metadata agent or an import/export agent.

A user provides a task. The orchestration layer determines which capabilities are required and routes the work to the appropriate agents.

And unlike a person working through a CAD workflow sequentially, multiple agents don't necessarily need to work that way.

One agent could analyze metadata while another checks drawings and another works with geometry.

Multiple engineering tasks can potentially happen at the same time.

Why Cloud-Native Matters for AI Agents

This may be where cloud-native architecture becomes most important.

AI agents need more than access to CAD commands. They need access to engineering data, the ability to work concurrently, a record of what they changed, and clearly defined permissions governing what they can see and do.

Those requirements look remarkably similar to the requirements of human collaboration in a modern engineering system.

Onshape already tracks the history of changes made to a design. That same concept can apply to agents.

It should be possible to see what an agent changed, distinguish those actions from work performed by people, and restore the design to a point before an agent made its changes.

Permissions will also matter.

An agent might be allowed to query a project but not modify it. It might be able to measure geometry, but not export anything. Another agent could be given permission to modify a particular project, but nothing else.

The long-term vision is that agents can operate much like people inside the engineering system, with appropriate access, permissions, and traceability.

Cloud-native CAD and PDM provides a particularly interesting foundation for that model: centralized data, concurrent access, complete history, restore and granular permissions already exist for human users. Those same concepts can be extended to AI agents.

4. Text-to-Code-to-CAD: Using AI to Customize CAD

This fourth category starts with an important distinction.

What About Text-to-CAD?

Text-to-CAD uses natural-language prompts to generate CAD geometry.

The concept is compelling: describe the desired part, and an LLM operates the CAD system to create it.

There have already been some impressive demonstrations. But there is a fundamental challenge underneath them.

LLMs are remarkably good at language and increasingly capable of using software tools. What they don't inherently understand nearly as well is CAD.

And CAD isn't simply about producing geometry that looks right.

A production CAD model also needs to capture the relationships between that geometry: topology, constraints, references, dimensions, feature dependencies, and, ultimately, design intent.

Design intent is the logic embedded in a CAD model that determines not only what the geometry looks like today, but how it should behave when the design changes.

That's a much harder problem.

An LLM can potentially execute enough CAD commands to create something that visually resembles the requested part. But geometry that looks correct isn't necessarily a good parametric CAD model.

Change a dimension. Modify an upstream feature. Increase the number of holes. Change the overall size of the part.

If the relationships weren't created correctly, the model can quickly fall apart.

The result may look remarkably good while still being brittle as an engineering model — difficult to modify, difficult to reuse, and missing the design intent that makes parametric CAD valuable in the first place.

Shape isn't the same thing as a robust CAD model. And a CAD model isn't truly valuable unless it captures design intent.

Text-to-Code-to-CAD Takes a Different Approach

Instead of asking AI to build increasingly complex geometry one CAD command at a time, another approach is to ask AI to write the engineering logic that creates the geometry.

That's: Text-to-Code-to-CAD.

With Onshape and FeatureScript MCP, an engineer can describe an engineering capability in natural language. The AI writes FeatureScript, and that code runs natively inside Onshape to create parametric CAD geometry.

The distinction is important.

Text-to-CAD asks AI to create a model. Text-to-Code-to-CAD can ask AI to create the parametric logic that generates an entire family of models.

The AI isn't just creating geometry. It can create a reusable CAD feature that captures design intent.

For example, a company might regularly design products requiring a particular thread, bearing seat, hydraulic cavity, enclosure, heat exchanger, sprocket or specialized geometry unique to its products.

The engineering requirements can be described to the AI, which can create FeatureScript that captures those rules.

The result is a native, parametric feature that engineers can use just like the standard features already in Onshape.

It can expose the right parameters. It can adapt as the design changes. It can be edited and improved. And, importantly, it can be shared across the company.

The Bigger Opportunity: Customizable CAD

This is where generative AI could have a much bigger impact on CAD than simply generating geometry.

For decades, CAD vendors have built general-purpose systems designed to work across thousands of industries.

As a result, CAD systems give engineers a common set of building blocks: extrudes, revolves, fillets, shells, holes, patterns and so on.

Engineers use those tools to build products, but much of a company's actual engineering knowledge still lives outside the CAD system — in standards, documentation, spreadsheets, tribal knowledge and the heads of experienced engineers.

What happens when creating sophisticated CAD customization becomes dramatically easier?

Companies can begin turning more of that knowledge into native, reusable engineering capabilities.

Instead of asking every engineer to repeatedly recreate the same design logic using generic CAD commands, that logic can be captured once and made available as a purpose-built feature.

This changes the economics of CAD customization.

Customization has traditionally required specialized programming expertise or significant automation projects. AI can dramatically lower that barrier.

An engineer can describe a requirement in engineering terms. AI can help translate those requirements into code. The resulting capability can then become part of the production CAD environment used by the entire engineering team.

That's a fundamentally different application of AI.

Generating geometry saves work once. Capturing engineering logic can save work every time that logic is reused.

And because these capabilities live inside a cloud-native CAD system, they can be made immediately available across an organization rather than installed and maintained individually on engineers' workstations.

Four Categories, One Direction

There is a natural progression across these four ways AI is transforming CAD.

1. AI Advisors help engineers find answers.
2. Discrete AI tools automate specific engineering tasks.
3. AI agents perform increasingly complex CAD work and engineering workflows.
4. Text-to-Code-to-CAD helps engineers create new, reusable CAD capabilities.

Cloud-native architecture doesn't create these AI capabilities by itself. But it does provide some important advantages underneath them: one continuously updated version of the software, centralized engineering data, access to rapidly evolving AI services, parallel work, complete history and restore, granular permissions, and the ability to immediately share new capabilities across a company.

The future of AI in CAD isn't one capability. It's a spectrum.

AI will answer questions about CAD. It will automate individual CAD tasks. Agents will increasingly perform engineering workflows. And AI will help engineers create entirely new CAD capabilities tailored to their companies.

The important distinction is that engineering needs more than geometry that looks right.

It needs editable models, design intent, traceability, and control.

The future of engineering isn't simply prompt-to-product. It's AI accelerating engineers within the engineering system of record — and increasingly helping them tailor that system to the products and processes that make their companies unique.