Coming back after a break? Take a five-minute warm-up on Section 1 before you start.
When an agent works on a task for you, it’s using a computer somewhere. Sometimes that’s your own computer, and sometimes it’s a computer that belongs to the company behind the agent. Where it works decides what it can reach and what happens when you close the window. It’s also the clearest way to tell apart the kinds of agents you’ll hear about.
On Your Own Computer
Some agents work right on your computer, with your own files. Coding agents such as Claude Code and Codex are the clearest example. You open one in a project folder, and it reads and changes the files there. They’re built for code, but people use them for documents and spreadsheets as well. Some desktop chat apps can also work this way when you give them a longer task and pick a folder on your computer, such as a folder of sales reports to pull into one summary. Either way, the agent works only in a folder you chose.
An agent working here can reach whatever you’ve given it on your computer. If you hand it a whole folder, it can open every file in it, including ones you weren’t thinking about when you asked, and it may change or delete them. Many apps ask before deleting, but don’t count on it. An agent working on your computer usually stops when you close the app or your computer goes to sleep.
The Model Still Runs on the Company’s Computers
Answer this before you read on. A coding agent on your laptop reads a client contract in your project folder. Does the contract’s text stay on your laptop?
No. The file stays in your folder, but the agent sends the contract’s text over the internet to the model, which runs on the company’s computers.
