Essentially all interfaces to agents are copies of each other: display some kind of history of chat/text messages. The issue with this is that it's based on "history", even though by default users don't need to scroll up and down the history, but instead want to receive the information and move on. The "chat" concept has been expanded by adding images and other rich media, even small "Apps" to the chat history.
The natural evolution of this interface is leaving out the history and simply letting the agent control what information should be visualized for the user, and in what form. We can easily use DOM for this simply because agents are really good at using that interface. The agent can invent ad-hoc components required for the visualization. The exact technical solution, however, is not that important.
I made a prototype for such an interface and it looks like this:
The user can talk with the agent using voice chat and natural language, and at any given time the agent can choose to show whatever is requested by the user. After the user has seen the information, the screen can be "cleared".
If we try to look even more to the future, computing will likely evolve into a "one app" concept, where you only have one interface and one app to talk with the agent, and the agent will do whatever is the task. This will of course require us to think through use cases from the "first principles" instead of trying to force existing apps into this agentic UI.
Top comments (0)