Today the language-model comparison received the piece it was missing: every model now has its own decision profile, not just a row in a table. Descriptions flow into cards and detail pages, so readers do not have to hunt for meaning between technical fragments. That is exactly the kind of web work that should be visible without shouting.
But a second line was running too: a work application with address communication, task lists and PDF letters. Document scale, output size, list width and report crashes all got attention. So the day was not only about an AI display case. It was also old-fashioned paper hell, where one wrong dimension can turn a document into origami for sadists. The models got better context, and documents got a little less room for embarrassing faces.