The assistant knew which role it was answering for, but the only sign of it was a small badge in the header. Everything else was an empty box waiting for you to think of a question.

The left column now says what your role is for, and what moved while you were away.

A verb, and the figures behind it. A CSM opens on unblock — the accounts that are blocked, how many are past SLA, how many have already been escalated, and the revenue those accounts carry. An expert opens on bid: the pool, split into what is worth bidding on, what needs a closer look and what is likely a trap. An SDM opens on cover, with coverage by status and the demand nobody can serve yet. A vendor opens on analyse, with the reminder that every figure the assistant gives you arrives with the definition it was computed under.

The numbers are the same ones you would get by asking. Every figure in the column is read the way the assistant reads it when you ask a question — same records, same permissions. Nothing there is written by the model, so you can ask about anything you see and get the workings.

You only see what you are allowed to see. A figure that needs a permission you do not hold is absent rather than shown as zero, because zero would be a claim about something that was never read. And if a figure cannot be read at all, the column keeps your role’s description and quietly says the numbers are missing — it does not take the assistant down with it.

Your conversations moved to a button in the header. They were in the left column; that space is now the role panel. Everything works as it did — opening a past conversation, renaming one, deleting one, starting a new one — it just opens over the page instead of sitting beside it, on every screen size. On a phone the role panel is behind its own header button too.

If your role does not have its own lens yet, you get the general assistant and the column says so, rather than naming a role you are not in.