Skip to content
Velum

Personal agents

An AI executive assistant that works inside your inbox and calendar

Short answer

An AI executive assistant agent works inside your email and calendar to sort incoming messages, draft replies, schedule meetings, prepare briefs before each call and track follow-ups. It drafts and proposes; you or your human assistant approve anything sent or accepted on your behalf. Most executives get back several hours a week from inbox and scheduling work alone.

Key takeaways

  • It runs in Outlook or Gmail and your calendar, not in a separate app.
  • Meeting prep pulls context from email, CRM, documents and past notes.
  • It works alongside a human EA, taking the repetitive load off them.
  • Access is scoped, logged and revocable, and sends need approval.

What an AI executive assistant can do

An AI assistant for executives covers the work that fills a day without moving it forward. It reads the inbox and sorts messages into what needs you, what can be delegated and what can wait. It drafts replies in your tone for review. As an AI scheduling assistant for business, it finds times across calendars and time zones, proposes slots, holds focus time and reshuffles when a meeting moves.

It also keeps track of commitments. When you promise someone a document on Friday, the agent notes it and reminds you on Thursday. After meetings it drafts the recap and action list from the transcript. For an AI assistant for CEOs, the most valued output is usually a short morning brief: what changed overnight, who is waiting on you, and what today's meetings are about.

How AI meeting prep works

Before each meeting, an AI meeting prep tool assembles a one-page brief. It pulls the calendar invite, recent email threads with attendees, CRM notes and deal stage, shared documents and notes from your last meeting with them. It summarises what the meeting is for, open items from last time, and questions worth asking. For external meetings it can add a short public profile of the company and people.

The brief arrives in your inbox, Slack or Teams at a set time before the call, usually 15 to 60 minutes ahead. After the meeting, if you use a recorder such as Otter, Fireflies or Teams transcription, the agent turns the transcript into decisions and actions, drafts the follow-up email and adds tasks to your task manager, all waiting for your approval.

Can an AI assistant replace an executive assistant?

For most leaders, no, and it should not try. A good human EA manages relationships, reads situations, handles sensitive matters and makes judgement calls that an agent should not. What the agent replaces is the mechanical part of the job: sorting, scheduling, drafting, reminders and prep. Many EAs use the agent themselves and support more principals or take on higher-value work.

For leaders without an assistant, the agent covers much of what a part-time EA would do on inbox and calendar. On safety, the agent should use delegated access through Microsoft 365 or Google Workspace with scoped permissions, keep a log of every action, avoid storing email content longer than needed, and never send, accept or decline without approval unless you have set a narrow rule for it.

How it works

  1. 1

    Shadow a normal week

    We review how email, meetings and follow-ups actually flow, and what you or your EA spend the most time on.

  2. 2

    Connect inbox and calendar

    The agent gets delegated, scoped access to Outlook or Gmail and your calendar, plus read access to CRM and documents where useful.

  3. 3

    Set rules and voice

    We define triage categories, scheduling preferences and protected time, and train reply drafts on examples of your writing.

  4. 4

    Pilot briefs and drafts

    For two weeks the agent produces morning briefs, meeting prep and draft replies that you review and correct.

  5. 5

    Launch with send approval

    The agent runs daily, and every email sent, invite accepted or meeting moved on your behalf waits for your approval.

Before and after

TaskBy handWith agents
Time in inbox2 to 3 hours a dayTypically 45 to 90 minutes, mostly on messages that need you
Scheduling a meeting3 to 6 emails back and forthProposed slots in one message, confirmed by you
Meeting prepScrolling threads 5 minutes beforeOne-page brief delivered ahead of time
Follow-upsTracked in memory or sticky notesLogged and reminded before they are due

Typical ranges from comparable deployments. Your baseline is measured before anything is built.

Tools it works with

  • Microsoft 365
  • Outlook
  • Gmail
  • Google Calendar
  • Slack
  • Microsoft Teams
  • HubSpot
  • Salesforce
  • Notion
  • Claude

Questions people ask

01

What can an AI executive assistant do?

It triages email, drafts replies, schedules meetings, prepares meeting briefs, writes recaps and tracks follow-ups. It works inside your existing inbox and calendar. Anything sent or accepted on your behalf should need your approval.

02

Can an AI assistant replace an executive assistant?

It can replace the mechanical tasks, such as sorting, scheduling and drafting, but not the judgement and relationship work a good EA does. Many teams pair the two, with the EA supervising the agent. Leaders without an EA get much of the inbox and calendar help a part-time assistant would give.

03

Does an AI executive assistant work with Outlook and Gmail?

Yes. It connects through Microsoft Graph for Outlook and the Google Workspace APIs for Gmail and Calendar, using delegated permissions. Access can be limited to specific mailboxes and revoked at any time.

04

How does AI meeting prep work?

Before each meeting the agent gathers the invite, recent emails with attendees, CRM notes, shared documents and past meeting notes, then writes a short brief with purpose, open items and suggested questions. It arrives at a set time before the call.

05

Is it safe to give an AI assistant access to email?

It can be, with the right setup: scoped delegated access, action logs, limited data retention, and no autonomous sending. Check where data is processed and whether the model provider trains on it. Enterprise API terms from major providers typically exclude training on your data.

Start with one workflow.

Thirty minutes. One real process. A practical next step.