Most of what you say to an assistant is not a big research request. It is a quick question. We were still routing all of it through the same heavy tool-picking pipeline, which is a bit like sending a postcard by freight. Now plain conversation takes the fast lane, and everyday chat came back about five times quicker.

You also see the answer sooner. A quick draft shows up right away with a small pulsing dot, then the fuller reply slides into the same spot and replaces it, instead of stacking a second message underneath. And if you use a screen reader, new messages now announce themselves as they land.