On the web, AI replies now appear as they are generated. If the connection drops during a reply, the saved answer can be recovered after reconnecting. Conversations saved on another device are also reflected, making it easier to pick up the same project.
We have also removed the wait before AI edits to manuscripts and reference materials, and reduced repeated notifications after routine saves.
If AI work pauses at an execution limit, the reply explains why it stopped and what remains. When it can continue, send “Continue” to check the saved results and resume the remaining work.