Archosh discusses details on the new AI foundation models from Autodesk (Neural CAD engines) with knowledge from interactions at AU25.
Parameter CAD engines have fundamentally remained the same for almost 40 years. The technology itself is accessible through menus, GUI-based tools as well as human mouse movements and data input. As a CAD user, we understand that fundamentally.
However, there is a significant problem with very mature parametric CAD engines, and that means that they were not designed for the AI era. They only understand the computer human input from the time before the AA and cannot understand new types of inputs associated with AI and soon acting AI. In addition, the parametric CAD limits the user to a preset detailed degree (from conceptually extremely detailed and dimensional) and cannot easily switch between these levels.
The first model of the AEC-related AI Foundation is now in the forma building design. With the neuronal CAD engine, users can quickly change design concept models to detailed building outs and structural systems. ((Click on a larger view)))
Imagine that a new tool has made quick and iterative workflows possible with which users were able to quickly explore and test ideas on a scale. It could then scale these iterative tests of conceptual, rough and approximate shapes into very detailed, dimensional versions in the same tool. And imagine that users could also enter a generative CAD system with voice text commands, hand sketches and pictures?
Neural CAD engines
All of this is possible, but only with neural CAD engines. Autodesk announced the existence of two neural CAD engines with a geometry focus on the D&M workflow with Autodesk Fusion and the others with a building logic, structural and spatial focus on the AEC workflow with Autodesk Forma building design.
Now Autodesk asks us to imagine the following scenario:
Imagine a scenario in which your computer includes the spoken language, sketches, three -dimensional design data and industry -specific workflows. Draw this scenario now with decades of the project knowledge of your team.
Autodesk demonstrated the early work of such a scenario. With voice for input, Autodesk was able to record new Ai Foundation models (neural CAD engines) and free sketch input for the design of a chair and also had instructions in the form of a picture of a chair that the user wanted to see in his neural CAD CAD machine-generated (generative AI)-based design.
Beyond LLMS
This moves far beyond the fair, large voice models (LLMS) with existing solutions. Autodesk approach for the next level of AI innovation is a complete restoration of traditional CAD engines behind tools such as Forma, Revit, Fusion, Civil3d, AutoCAD and more.
These new nerve CAD engines can create CAD geometry from an approach to system level on several inputs. At the same time, the new neuronal CAD engines from Autodesk, in contrast to the 3D graphics that can generate today's AI systems such as chatt, generate CAD geometry, which can also be processed with conventional parametric CAD inputs.
With regard to the architecture, Autodesks New Neural CAD AI Foundation models (AKA: Neural CAD engines) currently consist of one of the future of forma. It already forms the basis for the forma building design, which is why the system can allow the instructions for text box to create generative alternative building layouts.
This view of the geometric AI foundation model (Neural CAD engine) can be spontaneously designed from a text request. It is a completely new approach to machine learning to generate CAD objects in contrast to classic parametric CAD engines that have existed for over 40 years. ((Click on a larger view)))
The Forma building design can quickly create alternative constructions based on the change in the structural network, the selection of alternative structural systems and much more. You can ask the design of Forma building to create a individually loaded corridor design and a double loaded corridor design, and then the software can have it compared next to each other.
But in conversation with the people of the geometrically oriented nerve-caused motor via architecture and in the description of the process, in which architects today use physical trace paper in a recursive iterative process, during the design both in the plan as well as in the increase and in the 3D level, the meaning of the sketch input (something that has been proven) with the geometry that the geometry seems to own architecture in geometry. In other words, I was able to describe the neural CAD people and convince something that the AI foundation model for fusion has skills that are useful for architects.
These two AI foundation models will probably share common aspects over time. The merger -oriented model can create shapes that the fusion can now produce. The form-oriented design-oriented model only seemed to demonstrate box-shaped architecture. Admittedly, the vast majority of the buildings are naturally rectangular and does not have curved bodies or facades, but the joint work of “Maquee design companies”.
Argumentation with CAD geometry
Autodesk AI researchers have worked to inform their AI foundation models with CAD geometry, system design and the real world. Of course, this complements your ability to argue with LLMs.
The future of these AI foundation models lies in your availability for customers that you can personalize by adjusting them to the proprietary data and processes of your organization. This is the great view of the future of Autodesk Neuronal CAD engine models and offers a tempting view of the future of design.
Archosh analysis and comment
While Autodesk has shown his hand with its generative AI technologies, the software manager in AEC is not entirely in his own space with this ability. At AU25 there were Autodesk partners of third-party providers who had similar or related generative AI technologies, including those who were able to carry out sophisticated automation of multi-stage processes within tools such as Revit. Outside of the Autodesk world, there are competitors who also offer similar generative AI functions. (See: Archosh, “Archosh 11th 'Best of Show' Honors for Digital Technologies at AIA25 Boston”, July 2, 2025)
What is perhaps the most unique in Autodesk's neural CAD engines is the demonstration of multimodal input methods that range from language to hand sketches to pictures to slider user interface widgets and the various possible combinations. This in combination with the company's intention to enable organizations to adapt their KI fundamental models with their own proprietary data and to work detailed data over the entire Forma cloud platform, distinguishes Autodesk in its own ecosystem. It is the vision of the entire life cycle of a building and its data, in which AI technologies can actively participate in various workflow processes.