AI Explained · Practical guide

What can an AI agent access and do?

Understand the difference between giving an AI agent access to information and allowing it to take actions.

Before you ask an AI agent to handle a task, it helps to know what it can actually reach. Can it read a file you attach? Can it open your calendar? Can it send a message? The answers depend on the product and the access you have granted.

Reading a message is not the same as sending one

An assistant might help you draft an email without being able to send it. Another may have access to your email service. If you want to review it first, say: “Prepare the email, but let me check it before anything is sent.”

Then look at the recipient as well as the message. The same idea applies to a purchase, a calendar invitation or a file change: know what the next action will do before you approve it.

Share what the task needs

If you want help with a spreadsheet, could you start with a copy or a sample? Remove names or other details that are not needed. Before connecting an account, read what access the service requests and where your information will go.

Be clear about the result too. For example: “Compare these three offers and prepare a table. Don’t place an order.” Afterwards, check the table against the offers themselves.

If you no longer need an account connection, review it in the relevant service’s settings. The controls differ between products.

For more context, read what an AI agent is or try the AI task chooser.