Autodesk just gave AI agents direct access to Fusion.
Within less than two weeks,
Autodesk released two very different AI capabilities for Fusion. And together, they show where CAD is heading.
The first release is potentially the more fundamental one:
Autodesk has launched the Fusion Compute MCP in public beta. Instead of simply helping an engineer operate Fusion more efficiently, Autodesk is opening Fusion directly to AI agents.
The cloud-hosted MCP exposes thousands of Fusion capabilities through its API. Agents can create and modify parametric models, work with assemblies, generate CAM setups and toolpaths, export files and interact with Fusion projects.
And Autodesk is already showing workflows that go far beyond simple prompt-to-CAD demos. In one example, an agent creates an injection-molded enclosure, builds the core and cavity for the mold and then generates CAM operations for machining the mold plates.
CAD is increasingly becoming an execution environment for AI agents.
But Autodesk is attacking the problem from the other direction as well.
The second release targets a workflow almost every mechanical engineer knows too well: manufacturing drawings.
The 3D model may be finished, but the documentation work has only started. Drawing standards need to be configured. Units and sheet formats selected. Orthographic, projected and isometric views created and positioned. Dimensions added. Everything checked for consistency.
Autodesk Assistant now allows designers to describe much of this setup in natural language.
It can support:
• Drawing standards
• Units & formatting
• Sheet selection and layout
• Automatic creation of drawing views
• Standards-based dimensioning
So Autodesk is approaching Engineering AI from two sides at once. On the agent side, MCP gives AI systems machine-readable access to the underlying engineering capabilities.
On the user side, Autodesk Assistant removes repetitive interaction with the CAD interface.
AI in engineering will not be defined by adding a chatbot to CAD. The larger shift starts when engineers can express intent while agents operate the engineering toolchain underneath. Today, this might mean generating a drawing or modifying a CAD feature.
Tomorrow, the workflow could become much more continuous:
Engineering intent into parametric CAD, verified, prepared for manufacturing and finally technically documented.
I expect technical documentation to become one of the fastest areas of CAD automation, before we see truly end-to-end workflows in which engineering intent is translated into code, code into 3D models and those models into validated manufacturing outputs.
Rick Bouter |
Dominik Schlicht |
Nitin Ugale |
Ruben Hetfleisch |
Martin Eigner