A personal inbox is not just a mailbox.
It carries a person's name, relationships, history, habits, judgment, and reputation. When an AI agent borrows that inbox, it also borrows those signals. That may be convenient for a demo, but it is a weak operating model for real delegated work.
The better pattern is to give the agent its own address.
At gent.mx, this is the first product decision we keep coming back to. Before workflows, approvals, or APIs matter, the agent needs a boundary that is not a person's private mailbox.
Borrowing a human inbox creates blurred authority
If an agent sends from a human inbox, recipients see the human. If the agent reads from a human inbox, it may see more than the task requires. If the agent makes a mistake, the record is mixed into a personal message history.
That makes review, governance, and reputation management harder than they need to be.
The agent needs a proper address
A dedicated agent address creates a clean boundary:
- The work has a place to arrive.
- The agent sends from an address suited to the task.
- Credentials can be scoped to the inbox.
- Activity can be audited separately.
- Human approvals can hold sensitive actions.
This does not make the agent independent from oversight. It gives oversight a clear surface.
The inbox becomes part of the workflow
Once an agent can receive and send email, the inbox is no longer only a message store. It becomes a workflow surface.
Messages can be labeled, files can be saved, tasks can be created, webhooks can be called, and replies can wait for approval. The important part is that the message, action, and record stay connected.
Start with separation
The first step is simple: create an inbox, add the domain, issue a scoped token, and connect the agent.
From there, add controls only as the work needs them. See the inbox setup guide and the create agent inbox workflow.