As someone who has to deal with a lot of boilerplate construction permits all day, this is actually something that AI could be useful for.
I'm creating the permits, so I have total access to the spatial design databases for internal QA, but permit reviewers with the city/county usually only receive a PDF or some other document format.
These drawings frequently have thousands of dimensions each, and tallying them up by hand can be a bear. Training a hyper specific model to look at an image of a permit page and highlight areas where possible dimensional conflicts could be or find other commonly overlooked mistakes would be really helpful.
They absolutely just mean slapping a screenshot into GPT3o and calling it a day though...