Use case
How executive assistants use Claude across their tools
An EA's day is a dozen systems stitched together by hand. Custom connectors let Claude read and write across email, calendar, docs, and expense tools in one conversation, while you keep the final say on anything that leaves the building.
Executive assistants use Claude through custom connectors to their email, calendar, docs, and expense systems. In a conversation the assistant starts, Claude triages the inbox, drafts replies, proposes meeting times, assembles briefing docs from multiple systems, and preps expense reports. Drafting is free-flowing, while sending, declining, and submitting stay gated behind a confirmation.
Key takeaways
- Claude triages the inbox and drafts replies in the exec's voice, but nothing sends without your confirmation.
- Scheduling becomes a conversation: Claude checks calendars, proposes slots, and drafts the invite for you to approve.
- Briefing docs pull from email threads, the calendar, and past docs in one pass instead of four open tabs.
- Claude acts with your own permissions, only in conversations you start, with no background autonomy.
What does an executive assistant actually do with Claude once it is connected?
An EA uses Claude to work the seams between systems: triaging the inbox against the calendar, drafting replies that reference the docs a meeting depends on, and prepping briefings that pull from all three at once. The work happens in one conversation instead of a tab-switching relay, and you approve every outbound action.
The unlock is not any single tool. It is that email, calendar, docs, and expenses become one workspace. Ask Claude to find every unanswered email from this week's meeting attendees, and it can cross-reference the inbox against the calendar because both connectors are in the same conversation.
Everything runs on the human-in-the-loop model. Claude reads and drafts freely, but actions with consequences outside your account, like sending an email or declining a meeting on the exec's behalf, wait for your explicit approval.
How does Claude help with inbox triage and drafting?
Claude reads the inbox through the connector, sorts what needs the exec, what you can handle, and what can wait, then drafts replies for the messages you pick. Drafts land as drafts. The `send_email` action is gated, so every reply is reviewed by a person before it goes out under anyone's name.
A realistic morning pass looks like this:
- Ask Claude to summarize everything that arrived overnight and flag anything that mentions today's meetings.
- Ask it to draft replies to the three messages that just need a scheduling answer, pulling free slots from the calendar as it writes.
- Ask it to draft a polite decline for the speaking request, matching the tone of past declines it can read in the sent folder.
- Review each draft, edit what needs editing, and approve the sends one by one.
Because Claude can read the sent folder, drafts come out in the voice the exec actually uses, not a generic template. The more history the connector can see, the closer the first draft lands.
Can Claude manage the calendar and scheduling?
Yes. Claude reads availability through the calendar connector, proposes times that respect the rules you describe, drafts the invite or the reply to the scheduling thread, and creates the event once you approve it. Rescheduling works the same way: it finds the conflict, proposes the fix, and waits for your go-ahead.
Scheduling is negotiation plus data entry, and Claude handles both halves. You can describe standing rules in plain language: no external meetings before ten, keep Fridays light, always leave a buffer after board calls. Claude applies them when it proposes slots with find_availability and writes the result with create_event.
For a reschedule cascade, ask Claude to find every meeting that conflicts with the new travel dates, propose replacement slots for each, and draft the notes to attendees. You review the whole plan before a single invite moves.
How does Claude prep briefing docs from multiple systems?
Claude assembles briefings by reading across connectors in one pass: the calendar for who and when, the inbox for the latest threads with each attendee, and the docs system for prior notes and shared files. It drafts the briefing as a new doc with `create_doc`, and you shape it before the exec ever sees it.
This is the workflow where the multi-system access pays off most. Ask Claude to prep tomorrow: for each external meeting, pull the recent email history with the attendees, the notes from the last meeting with them, and any open items, then write a one-page brief per meeting.
The output is a draft in your docs system, in your template if you point Claude at one. Nothing about the source systems changes. Reading for a briefing is a read, and the only write is the new doc you asked for.
Keep a briefing template doc
Store one example briefing in your docs system and tell Claude to match it. It will follow the structure, headings, and level of detail every time instead of reinventing the format each morning.
Can Claude handle expense and travel admin?
Claude can read the expense system, match receipts from the inbox to card transactions, draft the expense report, and prepare travel itineraries from confirmation emails. Submitting a report or booking anything stays gated. Claude prepares the paperwork, and a person approves the step that spends money or commits the exec's time.
Ask Claude to collect every receipt email from last month's conference trip, match each to the expense system's pending transactions, flag what is missing, and stage the report with create_expense_report. You review the lines and submit.
For travel, Claude can pull confirmation emails into a single itinerary doc and add hold events to the calendar so the trip blocks time correctly. The bookings themselves stay with you or your travel tool.
Is it safe to give Claude this much access to an exec's accounts?
The connector acts with the permissions of the account you sign in with, never more. Claude only works inside conversations you start, so there is no background process reading mail on its own. Reads and drafts are free, while sends, deletes, declines, and submissions are gated behind a confirmation a person approves.
You decide which account each connector uses. A delegated-access mailbox behaves exactly like it does for you today: the connector sees what that access sees, nothing broader. And because every action is initiated in a conversation, there is a clear record of what was asked and what was approved.