How Do I Connect Email to an AI Agent Without Giving It Too Much Access?

In the Women Applying AI session on October 7, 2026, a participant said they were hesitant to give the AI agent access to Gmail and asked if they could just watch. I answered in one word:

From the recording, 18:49. Absolutely.

What are the four ways in?

• Connect your real inbox.
• Give the AI agent an inbox only it reads, and forward to that inbox what you want it to read.
• Paste a list of real open items into the chat.
• Connect nothing and follow along.

What does each one let the AI agent see?

A real inbox means the AI agent reads the real thing. The session page describes it as the fastest path, with Gmail and Google Calendar signed in on the account you use daily. I put the choice to the group this way:

From the recording, 27:00. If you feel comfortable setting up an email account and connecting your agent to it today, you can do that.

An inbox only the AI agent reads means you forward it only the information you want it to read.

From the recording, 19:51. What a lot of folks are doing is setting up an agent account and only forwarding information that they want the agent to read.

I called it an extra step, and I said what it gives you.

From the recording, 19:58. It's a great way to be comfortable with the level of access that you're giving to a system.

For a closer look at practicing with a dummy account, see How Do You Practice With an AI Agent Without Risking Real Data?.

A pasted list means the AI agent sees only what you paste. On the session page this is the lane for a work email, no Google account, or anyone who would rather not connect today, and the page offers a sample inbox to paste instead. It is on the session page.

Just watching means nothing is connected. My line for it:

From the recording, 27:06. If you're not yet comfortable, just continue to follow along.

A participant also asked whether the AI agent reads past email or only mail that arrives after setup. My answer:

From the recording, 19:20. It depends on the level of permissions we give it.

How do you limit what it can do once Gmail is connected?

The build in the session ran in Gumloop. On my relationship to Gumloop, I said this:

From the recording, 33:22. I don't work for Gumloop. I found this tool over the summer through Lenny's podcast, and I started using it and have since made myself an ambassador for them. I've created a relationship with Gumloop. I am not an employee.

This is the second control, and I showed where it lives. The steps, as shown October 7, 2026:

1. 01

   Select Gmail under connectors.

   A new screen opens.

2. 02

   Open tool management and select read only tools.

   I said there are six options underneath, where you restrict or give permission on what is read only (44:05).

3. 03

   Look at write/delete tools beneath it.

   This is where you give permission or restrict access to actions such as sending emails, updating, creating drafts and forwarding emails.

From the recording, 43:44. This allows me to have extra clarity and control over what my agent does.

From the recording, 44:19. And in that section, I can then give permission or restrict access to sending emails, updating, creating drafts, forwarding emails, et cetera.

To remove access later, the same connector row has a circle with a check mark and three dots beside it.

From the recording, 45:04. If I click on the three dots, it allows me to switch an account, but it also gives me access to easily remove access.

For the Gmail toggles in more detail, see How Do You Control What an AI Agent Can Do in Your Gmail?. If your hesitation is about training on your data rather than access, read Should You Opt Out of AI Training on Your Data?.

What goes wrong

Two things I named.

A password change breaks the connection:

From the recording, 45:48. So whenever you change your password in a system connected to Gumloop, you have to come back to Gumloop and reauthorize permissions.

And a connectors screen does not close on its own:

From the recording, 31:20. You actually have to click the save button to exit out of the screen

What the recording does not show: the six read only options by name, or what the pasted list looks like in the chat.

Where this comes from

This article grows from the session I taught live with Women Applying AI on October 7, 2026. The quoted passages are my own words, taken verbatim from the recording. The steps, tables and checklists around them were drafted by my team's AI agents from those words and my published pages, and checked against the recording before publishing. Screen steps are dated as shown October 7, 2026, and screens change. Platform facts that have a public source carry the dated line below.

Build Your First AI Agent in One Hour: the full session with the clips, the prompts and the transcript.

Checked on 2026-11-05 against Gumloop's agents documentation, which groups a connector's tools into read-only tools and write/delete tools.

Related reading: How Do You Practice With an AI Agent Without Risking Real Data?, How Do You Control What an AI Agent Can Do in Your Gmail?, Should You Opt Out of AI Training on Your Data?.