What we shipped. What we held.

Delivery and judgment
Shipped and iterated
Held for research
Update visibility Stronger cues so agents notice when the summary has changed since they last looked.
Change-highlighting Marks only newly added content, visible long enough to catch in real work.
Attachment handling Predictable inclusion behavior so agents know what the summary read and what it did not.
Authorship and provenance The line between AI text and human edits stays visible when content moves into work notes.
Open prompting How agents want to ask follow-ups and drive actions. Needs dedicated interaction research.
Quick actions What belongs near the summary and how to customize by org. Not ready to design without more signal.
Source transparency Best UX for citations at scale. The pattern needs more definition before it can ship consistently.
Sentiment Held until workflow purpose and emotional impact questions are resolved. Not a sequencing issue — a trust issue.