Behind the Screen

Who does what when an AI Worker works

Where this section fits

This is the third section of Claude and ChatGPT 2.0. The New Way of Working introduced AI Workers, the borrowed desk and Brightline's Friday payment run. Tour of the Apps showed each feature of the two apps.

This section looks behind the screen. It explains who does what when an AI Worker works, why the borrowed desk is cleared, and why permissions, not words, keep a job safe. The fourth section, The Architecture, gives the full evidence and detail.

Start here: the big idea

When you hand Claude or ChatGPT a whole job, like Brightline's payment run, more than one thing is working for you. A small team works behind the chat window. Each member has one job. This picture describes long, hands-on work. Regular chat may work differently, as Claude and ChatGPT: the same idea explains.

• The model is a brilliant consultant with no memory. You reach the consultant by phone. They give excellent answers, but forget everything after each call, so someone must brief them every time.
• The harness is the project manager. The project manager calls the consultant, keeps the notes, and does the tasks the consultant asks for.
• The sandbox is the borrowed desk. You met it in Where the work happens: a separate computer, usually on the company's servers, set up just for your work. The project manager does the hands-on work there, and the desk is cleared when you stop for a while. Some AI work runs on your own computer instead.
• The backend is the filing cabinet. It keeps a copy of your conversation, so you can come back tomorrow, on another device, and continue.
• Connectors are keys held at reception. They let the team use your other apps, like email or a calendar. The team asks reception for a key each time, so your password is never handed over.³

One sentence holds the whole idea:

The model is a smart helper that forgets everything between calls. The harness turns it into a worker that remembers and finishes the job.

The rest of this section explains each team member, and what this means for the way you manage an AI Worker. Start with What happens when you send a message.