What moved overnight, on one screen.
Who replied, what changed, and who is worth texting first today. It is a query over the log rather than a summary written from memory, which is why it cannot report a send that did not happen.
What is in it
- Replies since you last looked, with the sentence that matters rather than the whole thread.
- What is waiting on you: drafts held for approval, and how long they have been held.
- What went out on its own, if you are running copilot, reported as sends and queued separately.
- Who is next, ranked. Not the alphabet, and not the oldest: the person with the most specific thing to open with.
Why it is a query and not a summary
Everything the briefing says is derived from the event log, the same append-only record the cost meter and the compliance trail are read out of.
That is not an implementation detail, it is the reliability argument. A briefing composed from a model’s recollection of the day can say “reached out to five” when three were queued. A briefing that is a query over the log can only say what is in the log, and what is in the log is what happened.
The one thing it deliberately is not
It is not a nudge that arrives whether or not anything happened. A daily digest of nothing trains you to stop opening it, and then the day something does move you miss it.
The briefing also does not close the conversation. The end of a turn is a door, not a full stop: whatever Lewis says last is a question naming something it can actually do (rank the list, reconstruct a thread) rather than tomorrow’s briefing time, which reads as the assistant clocking off.