The tag import angle is the one I keep coming back to as well. Turning a controls engineer's spreadsheet into Ignition tag JSON is tedious and error-prone, and the SLC-to-CLX tagpath remapping is exactly the kind of repetitive transformation where checking generated output against the source is straightforward. Like you, I would still sanity-check a generated query rather than trust tuning to it. The trend screen question is interesting too, but I would want the generated screen to stay editable in the Designer rather than being locked into a black box.
Personally, I have developed a whole new way of designing Ignition assets in a deterministic and repeatable process without the use of the Designer and without AI. I still use AI for designing unique components. Will post examples soon.
The point about SCADA migrations resonates. Legacy systems older than the internet often lack the consistent documentation AI leans on, so the tool has little to work from. Rebuilding everything isn't realistic for customers who've kept those systems running for decades. Same with P&ID screens: if auto-generated versions look worse than what exists, that reads as a downgrade, not progress. Greenfield sites with repeated equipment are a different story, since scale and similarity suit automation. The work in migrations still seems to be in the details.