You are here. See how this question connects to other ideas.
Select a node to open its page · Expand to read within the map
One question
What does the agent see behind a button?
A visual control can correspond to named structure, state and actions, rather than only a region of pixels.
Imagine a task shown as a card. A person recognizes its title and status. An agent can work with the corresponding address and supported operations, without first inferring those known facts from a screenshot.
Other systems already expose structure through the DOM, accessibility trees and APIs. The AFS-UI question is how that structure participates in the same resource and operation model as the rest of the application.
The correspondence is semantic. It does not require a separate path for every shadow or decorative pixel. Visual inspection remains useful for layout, charts and appearance.
Teaching model · example paths, local state
Try the idea
Change the task status and switch its presentation. The object address stays the same.
What a person sees
In progress
What an agent can address · Example workspace
- /work/tasks/keynote/title
- Prepare the keynote
- /work/tasks/keynote/status
in-progress
Same object. Different representations.
This model illustrates semantic correspondence. It does not inspect this browser’s live AFS session.
Check your understanding
Would a pixel-level path for every shadow help an agent operate a task?
Usually not. Identify the object, meaningful state and permitted actions first.
Continue along a learning path
- Read an interface from both sidesStep 2 of 5