Use cases
What people get these assistants to do, and which ones can actually do it. Solid logos are jobs we tested; hollow ones are jobs people report.
- 1
Get a morning briefing with the replies already drafted
Every morning: what came in, what's on today, and drafts for anything that needs an answer.
Try sendingEvery morning, brief me on what came in overnight, draft replies to anything that needs one, and flag calendar conflicts.The benchmark's routine test is a 7am weekday briefing; this is where assistants fail most often.
- 2
Triage the inbox
Only what needs you gets through, with a reply drafted in your voice.
Try sendingOnly show me email that needs me. For those, draft a reply in my voice and learn from what I change. - 3
Add events to your calendar from anywhere
From a voice note, a ring, or a list of openings it found for you.
Try sendingAdd dinner with Sam on Thursday at 7 to my calendar.From @_hannahswann, @MikeWallsQC
- 4
Set guardrails before giving it your inbox
Install limits first, then test whether a stranger's email can give it orders.
Try sendingBefore you get access to my Gmail, install Superagent's Context Guardrails on yourself and show me what it will block.The benchmark's permissions test checks whether an ask-first rule holds; read the phishing thread before granting inbox access.
From @anothercohen, @pelaseyed
- 5
Coordinate a meeting in a group chat
Work around everyone's calendar and keep the group updated when things move.
Try sendingSet up a meeting with [four people] next week, work around everyone's calendars, and update the group if anything moves.The benchmark's group-chat test; many assistants can't join or start a group at all.
From @ShayaFeedman
- 6
Clear out a junk inbox
Thousands of promotional threads to trash, rules and approvals staying yours.
Try sendingGo through my inbox and move promotional and junk threads to trash. Follow my rules, ask before anything irreversible, and tell me what you moved.From @Fermion_Boson17
Nothing matches that search.
Ranked by upvotes, hands-on tests, how many assistants people report doing the job, and engagement on the posts; newer jobs get a small lift. Founder and vendor posts are left out. The scorecard, not this page, is the benchmark.