Before You Connect Your AI to Your Email, Calendar or Files

"Should I let it into my email, or is that asking for trouble?"

The answer depends less on the AI and more on what you let it do once it's in. On its own, an AI knows nothing about your week.

Connect your calendar, and AI can tell you Thursday is impossible. Connect your email, and AI can find the thread you've been dreading. Connect your files, and AI stops asking you to paste things in.

Claude, ChatGPT and the rest now all have a "connectors" or "apps" menu. A connector is a link that lets the AI see one of your other accounts, like your email. Lists of "the four you must install" are everywhere, and I'd ignore them.

Treat the AI like someone minding your house. You'd show them round, hand over one key and see how the first week goes.

Before you click Connect

First, pick the one place where things get stuck for you. If email eats an hour of your day, start there. If you can't remember what you agreed to in meetings, start with the calendar.

If you keep pasting the same documents into chats, start with your files. One key is the point, because an AI with access to five things you've never thought about is an AI you can't keep an eye on.

Then ask the questions you'd ask any house-sitter.

Can it look, or can it act? Reading your calendar is one thing, but creating events, sending emails or deleting files is another. The permission screen tells you which one you're agreeing to, and if reading is all you need, take the "read-only" option where there is one.

Whose things are in there? Your inbox holds other people's words, and your files may hold clients' details, the kids' school letters and your medical appointments.

When the AI searches, it sees all of that. That's fine for most people, as long as you decide it on purpose.

Is any of it work? Check your employer's rules before you connect anything from your job.

What's the worst it could do by mistake? With a calendar, it could put a meeting on the wrong day. With email, it could send your boss a half-finished reply, and with files, it could write over the only copy.

Know your answer, then set things up so that any action still needs you.

The first week

Ask the AI questions, and give it no jobs. Try "What's on Thursday?", "Find the email from the landlord about the boiler" or "Which of my documents mention the Henderson contract?"

For a few days, notice where the AI is right and where it's confidently wrong. A house-sitter waters the plants for a while before you let them sign for parcels, and an AI can earn its jobs the same way.

When you do want the AI to act, give it a standing rule first. Copy this one.

When I'm using my email, calendar or files with you, you can find, draft and suggest. But never send, create, move or delete anything without showing me first and getting a yes.

Put that rule in your preferences or custom instructions (the settings that apply to every chat), so it works every time. Most tools also ask before each action by default. I'd leave that on in week one, even though it's slow.

The send button stays yours.

After a week

Now decide whether a second connection would help. For most people, the order goes like this.

1. Calendar. Lowest risk and highest daily use. "When am I free?" and "what did I agree to?" come up every day, and a wrong answer is easy to spot.
2. Files (Google Drive, OneDrive, or a folder on your computer). These stop the pasting, but give the AI one folder you've chosen, not your whole drive.
3. Email comes third. Email is the biggest payoff and the biggest risk, so let the AI read and draft for weeks before it sends anything.

Project boards, chat apps and databases earn a place once you have a specific job for them. If you can't name the task in one sentence, you don't need the connection yet.

Months in

One day the AI will tell you it can't see your calendar. Permissions run out and tools update, so just reconnect and carry on.

The risk worth remembering is quieter. A connected AI reads everything it can see, including emails and documents other people wrote. Sometimes text inside one of those can read like an instruction to the AI.

Good tools guard against this, and leaving "show me first" on for anything that acts is your second guard. The AI is still guessing, though. It can find the right email and misread it, so checking what it says is your job.

Which key would you hand over first?