In Models and Harnesses, you saw that the model asks and the app does. This lesson gives you a picture to hold on to, and then uses it to explain one of the most confusing moments with an AI assistant: when it can’t open something you gave it.
Picture the model as an expert sitting in a sealed room. The expert can’t walk out to look at your files or open your email. Instead, someone outside the room, the app, slides material under the door: your request, some instructions, and any documents that might help. The expert reads everything and slides a note back. Sometimes the note is an answer, and sometimes it’s a request, like “Please open this web page,” which the app can carry out or refuse:
The analogy leaves out a few things. Models can write requests in a precise, structured format, and many models can look at images and other kinds of files, depending on what the app supports.
A model can’t reach your files, apps, or accounts on its own. The app around it provides the material and carries out the actions.
