Three-in-one:
1. BETTER CRASH RECOVERY
2. LAYER REPARENTING
3. TEACHING MOMENTS DURING FRENETIC ACTIONS
I'm not crazy about AI unless it's more like stored procedure or triggers in databases, or id it cannot just go off and be an agent that destroys me (no specific app referenced, I just mean any app any user might run or have idling on their system).
.......................
1. BETTER CRASH RECOVERY
But, I hope Tim and team can bypass what I suspect is a horrifically nasty patent in the design space. I know from using AutoCAD around 2011 (from 2007-2012), of AutoCAD crashed, and I hadn't saved or autosave didn't happen RIGHT BEFORE a crash, AC still could recover my work up to maybe 1 or 2 seconds.
With SharkCAD and ViaCAD, save intervals were not useful to me if I had a save that took 2-3 minutes to complete (a timeframe that eventually got in the way as my files got bigger, making me turn off saving altogether, especially nefore I got 16+ GB RAM). So, if I work for 3 or 4 minutes or 3 hours, and SharkCAD blows up (like it did 3 days ago after 2 or 3 Undos on some simple splines but in a rats-nesty geometry-loaded file), I was debilitating. I can spend an hour on small moves, but not always move a lot in a small timeline. So, I can easily forget what I did, intended to do, and have to frenetically retrace my steps.
An AI Shadow might do wonders for anyone. It shouldn't records view ports or non-geometry stuff. Just real-time log any geometry actions and layer, layer names edits, or creations.
It should be able to record thousands of concluded actions and be a small file of under a dew dozen or few hundred kb when saved invisibly, not live-stream as any image that might eat up GB of space if the user doesn't reboot for weeks, or if the user opens and closes countless files.
When the app inevitably crashes or power is lost, Shark should notice the crash event (independently of Windows 11, as it can't be trusted, as I've had crashes and Windows utterly didn't recreate my supposedly-saved session states even from before the crash, infuriatingly (randomly) bringing a blank desktop) and offer to reconcile differences by doing a save-as (to reassure users their last-restarted file won't be changed).
It would be like a high-speed replay integration no human can do.
If there's a patent on such an outcome, and if its obnoxiously overly-broadly written or hogs the entire CAD landacape, it needs to be unwound because even some word processor and other apps have such a recovery.
SharkCAD should not be left out of the fun. I suppose an alternative is an AI agent or process watches the user and makes logs for recovery purposes, but, without developer access, the app can't get into the kernel for steps-reintegration.
.......................
2. LAYER REPARENTING
Once the AI copilot can achieve that, it should be able to learn what the user does with layers, and then be able to offer to help the user re-parent layers across current layer/tree/branch. If there's a patent on that, it, too, needs to be unwound.
.......................
3. TEACHING MOMENTS DURING FRENETIC ACTIONS
If a user neurotically, frenetically, or in panic mode keeps trying to do something not possible, the app should internally notice the commands and tell the user alternatives exist, even if they won't effectuate a desired outcome. But, the user could at least elect to operate in drive-and-still-learn mode.
The AI should also point out that a solidify that fails is failing because of exactly at what location some inflection, inversion, self-intersection, or whatever is kinking, knotting, or buggering the math. I spent DAYS making hundreds of neurotic, crest falling hopeful and outright dumb moves of points knowing that in the past I could stitch surfaces and shell solids in under a second each with even worse starting geometry, yet fail on (visibly) better splines.
To be fair, I eventually got the solid and the shelling partly by using auxiliary reference lines to get spline control points ever so closer to cleaner, and, pleasingly, the solids ops were lickety-split fast.
