# Agentic UI

Agentic UI is user interface that an AI agent renders and acts through while it carries out a task, such as a card to confirm, a form to complete or a picker to choose from, shown inside the conversation. It lets the user see, adjust and approve what the agent is about to do.

## How does agentic UI work?

Agentic UI sits between a request in plain words and an action in the product. When text alone would be slow or risky, the agent shows a small piece of interface that fits the step, and the user's interaction with it feeds straight back into the task.

Take a scheduling tool. A user types "move my one-on-one with Dana to Thursday afternoon." The agent checks both calendars and finds three open slots. Instead of listing times in a paragraph and asking the user to type one back, it shows a picker with the three slots. The user taps 3:30. The agent shows a confirmation card (old time, new time, Dana notified), the user confirms, and the agent moves the meeting through the product's own calendar function.

Three things make that agentic rather than a regular dialog box:

1. **The agent chose it.** The interface wasn't on a fixed screen. The agent decided the step needed a choice and picked the component.
2. **It's filled with the user's live data.** The slots, names and current time came from this user's account at this moment.
3. **It leads to an action.** The input isn't stored for later. It completes or confirms a step the agent then carries out.

It overlaps with generative UI, where a model generates the interface itself. Most agentic UI today fills in components the product team has already defined.

## What are examples of agentic UI?

Most agentic UI comes from a small set of components the product team defines and the agent fills in:

- **Confirmation card.** Summarizes what's about to happen ("Archive 14 projects with no activity since June") with Confirm and Cancel.
- **Pre-filled form.** The fields the agent could work out are already filled; the user completes or corrects the rest.
- **Picker.** A short list of options, such as time slots, records that match a vague name, or plan tiers.
- **Result card or table.** Shows what the agent found or changed, with links into the full screens.
- **Upload block.** Asks for a file at the step where the agent needs one, like a CSV of users to invite.

The components look and behave like the rest of the product, so users don't have to learn a second interface.

## How is agentic UI different from a regular app screen?

A regular screen is designed in advance for a path the designer expected. Every user sees the same form in the same order. Agentic UI appears only when the task needs it, for the step the user is on, with their data in it.

That makes the design questions different. Instead of drawing every screen, teams decide which components the agent may use, what each one can trigger, which steps always need a confirmation card, and how results link back to the full product for anyone who wants the detail.

The full screens don't go away. Agentic UI handles the request in front of the user; the regular screens stay there for browsing, comparing and detailed work.

In Foldspace, the Product Agent [draws agentic UI (cards, forms, pickers) into the conversation](https://docs.foldspace.ai/guides/in-chat-ui/) for the step at hand, then runs the action on the product's backend APIs with the signed-in user's own access.

## Learn more

- [Product Agent](/platform/product-agent)