Two pieces of news landed on the same day, 15 September, and both touch software I work in: Inventor and CorelDRAW. Autodesk used its annual conference to show a new round of AI for Fusion, Inventor and Vault. Corel released the September update of CorelDRAW Graphics Suite 2026, built around generating vector artwork from a prompt. One is about models and assemblies, the other about the graphics that end up cut, printed and routed.
Autodesk: the assistant gets hands inside Inventor
The part that matters most to me is Inventor. The Autodesk Assistant can now work through Inventor's own API and iLogic: analyse a model, change parameters and components, create model states and automate repetitive tasks. Until now, that kind of automation meant writing iLogic rules or a script by hand. If the assistant does it reliably, a simple configuration job (the same frame in six widths, the same bracket for three panel thicknesses) stops being a small programming task.
Vault gets natural-language search that suggests existing parts to reuse instead of drawing them again. Inventor also becomes the first desktop program connected to Fusion's cloud data.
In Fusion itself, Autodesk announced AutoTimeline, which turns imported geometry into an editable parametric part with a feature history, and AutoAssemble, which puts such parts together into an assembly. For anyone who receives dumb STEP files from suppliers and has to modify them, AutoTimeline is the most interesting item on the list. Next to it came tools for reusable manufacturing automations (machine setup, toolpaths, production preparation) and an agent-driven layer for product data. Autodesk presents most of this as coming, not as something you can switch on today.
CorelDRAW: text-to-vector and a faster trace
Corel's update adds text-to-vector: you describe a graphic, or feed in existing artwork, and get editable vector shapes rather than a bitmap. PowerTRACE, the bitmap-to-vector tool that every sign shop knows, is now up to four times faster, with better object structure, better colour handling and AI background removal. The browser version gains uploading your own fonts and direct DWG and DXF export.
What I will check before trusting it
For signage, a vector is only useful if the machine can read it. The test I would run on any generated or traced artwork is the same one I apply to a logo sent by a client: closed shapes, no overlapping duplicates, a sensible number of nodes, no hairline gaps between letters that the router will treat as separate parts. Faster tracing is welcome; a trace that still needs an hour of cleaning is not a saving.
On the CAD side the question is control. An assistant that changes parameters in a production model has to leave a trail I can check, because a wrong width in a model does not stay on screen, it goes to the laser. I would try it first on copies of finished jobs, where I already know the right answer.
If you need production-ready files from artwork or models, see graphics and DTP and CAD automation.
