Four hundred messages arrived and you were not there for any of them. An agent reconstructs the fortnight: what was decided, what is still open, who you owe, and what has been waiting on you since before you left.
This is not the same problem as a daily inbox. A daily inbox is a sorting job where the newest message is usually the relevant one. A backlog is a reconstruction job, and nothing about it is chronological.
You left on the 3rd and came back on the 17th. The email that matters most may be from the 4th — a decision taken in your absence that everyone has since acted on — and reading the newest first puts it four hundred messages down. Sorting a backlog by date optimises for the wrong thing, because in a backlog the value is inverse to the recency.
You left with something outstanding. It is now buried underneath a fortnight of other people's conversations, and unlike a new arrival it has no notification, no unread marker, and no way of surfacing. This is the single most common way a holiday creates real work: the thing that was due while you were away, found on the Friday you get back.
Most of a fortnight is the same handful of conversations moving, plus notifications that never needed a decision. The reason reading it feels impossible is that the volume is not the workload — the workload is perhaps fifteen decisions spread across three hundred messages, and there is no way to find them without reading things you can skip.
Copy it once, paste it into Zaira, and it reads the window you were away, groups it by project rather than by sender, pulls out what was decided and what never got answered, and puts the things that were already waiting before you left at the top.
You get one page instead of four hundred messages, and the first thing on it is not the most recent email but the oldest commitment that is now overdue. The run before you leave takes ten minutes; the run after takes one.
The output is four headings and roughly a page of text. The work is finding the fifteen decisions inside three hundred messages, which is a volume problem rather than a reading problem, and that is the part an agent is actually good at.
You left with something outstanding and it is now invisible under a fortnight of mail. The prompt searches backwards past your own window for threads that were already open, and reports the past-due commitments first rather than leaving them in a cluster somewhere.
Reading chronologically tells you what arrived. Reading by project tells you whether anything is stalled — which is the question you actually have when you open a laptop after a fortnight.
The questions people ask before reconstructing a mailbox they were not watching.
Yes, and the difference is in what the agent is asked to do. A morning briefing is prospective: today's meetings, mail that arrived overnight, tasks. This is retrospective — it reconstructs a window you were not present for, groups it by project, and specifically goes looking backwards for commitments that were already open when you left. A morning brief cannot tell you what you owe from before Tuesday, because Tuesday is not in its window.
As far as you tell it, and the window is yours to set in the prompt. A week away is straightforward. Three weeks away and a busy inbox, you may need to run it twice with different windows rather than once with a huge one — a very wide sweep produces clusters too coarse to act on, and splitting it lets you read the urgent part properly and skim the rest.
No. The prompt stops at the report and explicitly refuses to draft, which is right for a first run: the list itself is the judgement, and you should be the one deciding which of those four hundred messages deserves a reply from you rather than receiving forty drafts written while you were on a plane. If you want drafting after that, that is the separate automated-follow-up workflow.
They are in scope, and they are frequently the valuable ones — a thread you were cc'd on is often one where somebody else made a decision your project now depends on. The prompt reads the thread as stored rather than only the messages addressed to you, and reports who actually made the call, which is frequently not you and not the person you would have asked.
Yes, and it works better that way. Run it the evening before you return and you read it with the full context of the trip still available, which makes the open questions much easier to judge. The one thing to add in that case is a line telling it you are not available to reply, so it reports commitments as outstanding rather than assuming you will pick them up when you land.
The tool calls are cheap; the cost is in the reading, and it scales with message count rather than with your mailbox size. A fortnight on a normal inbox is well within what an agent handles in one pass. If you have come back to something exceptional — a month, or a mailbox with several thousand messages — narrow it to your priority senders and open threads first and let the rest wait, because a rough reconstruction of the important half is more useful than a complete one you cannot finish reading.