The problem
A shared inbox in a small office gets a few hundred messages a week. The front desk spends the first hour of every day sorting before any real work starts, and the important email from a new client sits between a toner ad and a vendor statement.
What Cowork does
- An hourly task reads new messages in the shared Gmail or Outlook inbox.
- It applies labels from a short list the office agreed on.
- For routine requests (office hours, document received, appointment change, where to send payment) it writes a reply draft using the office's FAQ document.
- Anything that needs judgment gets a label and a one-line summary but no draft.
- It never replies to messages that ask for anything unusual, like payment changes or passwords.
Where the draft lands
Labels in the inbox and reply drafts in the same threads.
What stays with you
Sending replies, handling anything with advice or money, and checking anything that looks like fraud.
Labeled this hour: New client (1) · Existing client (6) · Vendor (3) · Billing (2) · Spam (9)
Draft (existing client, document received): Hi Pat, got the September bank statements, thank you. We'll have the reconciliation done by the 10th and will reach out if anything needs explaining. Jo
Watch-outs
- Inbound email can contain instructions meant to trick an AI (prompt injection). Keep sending manual and never let an email trigger a payment or data change.
- Law and insurance offices should check what client information is acceptable to process before turning this on.
Questions owners ask
Will it reply on its own?
No. It leaves drafts. Sending needs approval by default, and this setup keeps it that way.
Does it work with Outlook?
Yes. Microsoft's official connector covers Outlook, OneDrive, SharePoint and Teams, and needs a one-time admin consent.