AI for CAD Tools
What Dassault’s built-in AI does in CATIA and 3DEXPERIENCE, what the FD03 release adds, and where an engineer still needs retrieval across their own data.
·
⏱
8 min read
Michelle Ben-David
Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.
BOTTOM LINE
Dassault’s built-in AI is a credible set of tools for people working inside CATIA on the 3DEXPERIENCE platform: Aura for guidance, Command Intelligence for next steps, and the FD03 generative features for parts and assemblies. Its boundaries are the platform, the vendor’s data model, and the fact that predictions come from geometry rather than from your own standards and history. Several of its claims, including the 30 percent figure and the physics-grounded checks, come from the vendor and are worth testing on your own parts. Use the native tools for modeling speed, and add a retrieval layer across your PDM, PLM, and ERP when the real bottleneck is finding what your company already knows.
Dassault Systèmes has spent the past year putting AI inside CATIA and the 3DEXPERIENCE platform, and the announcements read as a single story. In practice they are several different things: a chat-style guide, a command recommender, a set of named virtual companions, and a family of generative features that arrived with the R2026x FD03 release. Each one answers a different question, and each one has a boundary.
This post separates them. It uses Dassault’s own descriptions of what each feature does, notes where the vendor’s language is a claim rather than a measurement, and then looks at the part of an engineer’s week that none of the built-in features were designed to cover: finding what your own company already decided.
What Aura, Command Intelligence, and the Companions Actually Do
Start with the most visible piece. Aura is a chat companion inside cloud-based CATIA on the 3DEXPERIENCE platform. Develop3D describes it as in-context guidance from large language models, and notes it is the same Aura Dassault launched for SolidWorks earlier in 2026. Dassault’s press material frames Aura as a business expert that orchestrates knowledge and context across requirements, projects, and changes so teams stay aligned. It is available now.
Command Intelligence is narrower and more practical. It suggests the next command based on what you have selected and what you are likely to do. Dassault’s Sappin told Develop3D it speeds up design time by 30 percent. That is a vendor figure without a published method, so treat it as a direction rather than a benchmark for your own models.
The platform also introduces two more named companions. One is called Leo and is described as the engineer: it handles technical problems across mechanical, knowledge-based, systems, and requirements engineering, and works directly with CATIA and Simulia. The other is Marie, described as the scientist, with expertise in materials, chemistry, and formulations. Dassault lists both for 2026. A note on naming: Dassault’s Leo companion is a separate product from Leo AI, the company that publishes this blog.
What ties these together is where they run. Dassault says the companions are available on the 3DEXPERIENCE platform SaaS, and they ship with CATIA R2026x FD03. An engineer on a desktop installation of an older release is not the audience for any of this.
IN PRACTICE
It surfaces the relevant internal material, previous design decisions, past calculations, and backs everything with a cited source I can actually click on and verify.
- Yuval F., Clalit
What the Generative Experiences in R2026x FD03 Add
Dassault groups the more ambitious capabilities under the name Generative Experiences. The idea is that you provide text, documents, drawings, or scanned data, and the system produces a deliverable: a design update, a completed task, or an enriched virtual twin.
The FD03 capabilities Dassault lists are solid feature creation, part generation from multimodal inputs, assembly generation through predicted mechanical interfaces, generative engineering rules, generative systems requirements, and generative systems architecture suggestions.
The assembly item deserves a closer look because it touches the daily work of a mechanical engineer. Mechanical interface prediction infers how two parts relate, for example where a pin passes through a clevis and what the hinge axis should be, and then builds the connection. That is a real saving on repetitive mating work. It is also a prediction made from geometry, which matters in the next section.
Alongside the AI features, CATIA now runs in a browser with many native tools in the same interface. Develop3D positions that as a way to log in and view or edit files away from the office rather than as a headline change, but it reinforces the same point: Dassault’s direction is cloud first, and the AI features arrive on that path. Teams that plan a migration should treat the AI as one benefit among several, not the sole reason to move.
Dassault says outputs are grounded in an Industry World Model and checked against physics, and that customer data is never used to train its AI. Those are vendor statements. The announcement does not publish accuracy figures, usage limits, or a clear line between generally available and preview features, so an evaluation on your own parts is worth running before you plan around any of it.
Where the Native AI’s Reach Ends
Every built-in assistant shares three boundaries, and CATIA’s are no different.
It follows the platform. The features live on 3DEXPERIENCE with the current CATIA release. Our guide to AI tools for CATIA covers why CATIA V5 teams do not get these features, and many manufacturers still run V5 for large parts of their portfolio.
It follows the vendor’s data model. A companion that works with CATIA and Simulia is strong inside Dassault’s world. A bill of materials that lives in an ERP, a drawing archive on a network share, or a supplier spec in a PDF is a different system.
It predicts from geometry, not from your standards. A predicted interface is a reasonable guess. Whether it matches your company’s preferred bolt pattern, pilot fit, or dowel practice is a question about your own history, not about shape.
None of this is a criticism of the feature set. It is the normal shape of a vendor tool: deep inside one product line, thin outside it. The same pattern shows up with other vendors, as we described in our look at Creo’s built-in AI and in the wider comparison of built-in AI across CAD and PLM.
The Retrieval Gap: PDM, Standards, and Institutional Memory
Most of the time an engineer loses is not spent modeling. It is spent finding: the previous revision, the reason a tolerance was chosen, the standard part that already exists, the failure report from three years ago. Generating a new solid feature does nothing for that, and a general chat companion cannot cite a decision it never saw.
This is the gap where an AI layer on top of your existing systems earns its place. Leo offers integrations with leading PDM and PLM platforms, including SolidWorks PDM, Autodesk Vault, PTC Windchill, Siemens Teamcenter, Arena PLM, and others, and it connects to local and network directories and ERP data as well. It is trained on more than one million pages of standards, books, and articles, and every answer carries a citation you can click to check. It sits above the systems you already run rather than replacing them, which is why it works the same way whether the CAD underneath is CATIA, Creo, or something else. For how that layer connects to the rest of the stack, see our piece on PLM and ERP integration.
Consider the predicted interface from the previous section. A prediction can be right geometrically and still wrong for your plant. If the retrieval layer can show that the last six housings in this family used a four-bolt pattern with a pilot spigot, the engineer accepts or corrects the prediction with evidence in hand instead of by memory.
The same logic applies to standards. A tolerance callout, a weld symbol, or a fastener torque is only correct against a specific edition of a specific standard, and a retrieval layer that points to the source page lets an engineer confirm the number before it goes on a drawing.
Security matters for this kind of layer, and Leo is SOC-2 certified and GDPR compliant. No AI is trained on customer data, and your IP stays protected.
How to Decide What to Rely On, Case by Case
A practical way to split the work is to ask what kind of question you are answering.
If the question is how to do something in CATIA, such as which command comes next or how a feature behaves, the native assistants are the closest tool, provided your team is on the cloud platform and the current release.
If the question is whether a generated or predicted result is acceptable, test it on a handful of your own parts and compare against what your team would have built. Ask for the evidence behind the checks, not only the claim.
If the question is what your company already knows, such as past designs, standards, supplier history, or the reason behind a change, you need retrieval across your own data, and that is outside what any single CAD vendor’s assistant covers.
Most teams will end up using both layers. The built-in AI speeds up the modeling step. A cross-system layer shortens the search step. Our guide to what Aura does in SolidWorks makes the same distinction for another Dassault product, and it is a useful second read if your team runs both.
FAQ
Find what your team already knows
Search past designs, standards, and decisions in seconds.
Leo connects to your PDM, PLM, and ERP so engineers can answer questions with cited sources from their own data. See it on your parts.
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams
