What is an agentic interface?

How is an agentic interface different from a graphical interface?

In a graphical interface, the user does the work. They find the screen, fill the fields, click the buttons, and repeat for every record. The software waits for each click.

In an agentic interface, the user describes the outcome and the agent does the work. "Set up the Q1 campaign for the enterprise segment with the same budget as last quarter" becomes a set of real actions: create the campaign, apply the segment, copy the budget, then show the result for review.

A natural language interface describes how the request comes in. An agentic interface is defined by what happens next: the request ends in executed work, not an explanation of the steps.

What makes an interface agentic?

Four traits separate an agentic interface from a chat panel added to a product:

• It acts. It calls the product's functions, updates records and moves the user to the right page. Answering questions is part of the job, not all of it.
• It works from live context. Each request is grounded in who the user is, what's on their screen and the current state of their data, not only in static docs.
• It stays inside the rules. It acts with the signed-in user's permissions and the product's own validation, and asks before steps that are hard to reverse.
• It shows its work. Plans, choices and confirmations appear as small pieces of interface in the conversation, so the user can see and steer what's happening.

It doesn't replace the graphical interface. Users still browse dashboards and make precise edits by hand. The agentic interface handles the requests that would otherwise take many clicks across many screens.

Where does an agentic interface sit in a product?

It sits as a layer on top of the product the team already has. It uses the same APIs, permissions and data as the screens, so it doesn't need a rebuild of the product underneath, and anything it does shows up in the product exactly as if the user had done it by hand.

That layer also records something screens can't: what users asked for in their own words, which requests were completed, and which the product couldn't yet handle. That record tells the team what to build next.

Foldspace is that layer, the agentic interface layer for AI-Native software: users state the request, and the Product Agent carries it out in the product underneath.