Everything is commanded from the chat.
The buttons that opened, closed, focused and woke browsers are gone. Ask instead - "open another browser", "close the second one" - because the agent already does all four, and a control beside the chat is a second way to move the same thing.
That makes the small panes views. Clicking one pins what you are looking at and tells the agent nothing; clicking it again gives the view back to the agent, which is what you want while it works. The pane the agent is driving carries a mark.
And the session column is closed until you ask for it, and remembers what you last did with it.