# Agentic UX

Agentic UX is the practice of designing products in which an AI agent carries out tasks on the user's behalf. It covers how the agent understands what the user wants, when it asks for confirmation, what it is allowed to do, how it shows its work and how the user undoes it.

## What are the principles of agentic UX?

Agentic UX rests on five patterns that let a user hand a task to an agent and stay in control of it:

1. **Intent capture.** Turn a loose request into a precise one. When the request is ambiguous ("clean up the old deals"), the agent asks one focused question or shows its interpretation, rather than guessing.
2. **Confirmation.** Match the friction to the stakes. Reading data needs no confirmation. Changing one record might need a quick check. Deleting, sending or billing always shows exactly what will happen and waits for a yes.
3. **Permissions.** The agent can do only what the signed-in user can do by hand. A user who can't delete a workspace in the settings screen can't delete it by asking.
4. **Visibility.** Show the plan before acting, progress while acting, and a clear record afterward of what changed and where.
5. **Undo.** Every change the agent makes should be reversible the same way a manual change is, or flagged as permanent before it runs.

## How is agentic UX different from traditional UX?

Traditional UX optimizes a path. The designer knows the steps, lays out the screens and removes friction from each one. Success is a user finishing the flow quickly without errors.

Agentic UX designs for a delegate. The user describes an outcome and the agent picks the steps, so the designer can't draw every path in advance. The work moves to the boundaries: what the agent can do, how it explains itself, where it stops to ask, and how trust builds over time. Early on, users want to see each step. Once the agent has been right a few times, they want fewer interruptions for low-risk work.

It also changes what you measure. Alongside task completion, teams track how often requests resolve, which ones the agent couldn't handle, and how often users undo or correct its work.

## What does good agentic UX look like in practice?

Take a sales manager in a CRM who types "reassign all of Priya's open deals to Sam, she's going on leave."

- The agent restates the request with numbers: 23 open deals, total value, and two deals where Sam isn't on the account team.
- It asks one question: add Sam to those two accounts, or leave those deals with Priya?
- It shows a confirmation card listing what will change, including the note it will add to each deal.
- It reports progress, then a summary with a link to the filtered deal list and an Undo option.

Nothing here is clever. Each step answers a question the manager would otherwise worry about: did it understand, what will it touch, can I take it back.

Foldspace's Product Agent builds in the confirmation and permission steps: it acts only as far as each signed-in user's permissions allow, a team can [switch on a confirmation message for any action](https://docs.foldspace.ai/user-guides/authoring-actions/) so the user approves it before it runs, and the agent puts agentic UI such as cards and forms in the conversation when a step needs the user's input.

## Learn more

- [Build an Agentic Roadmap That Accelerates Product Adoption](/blog/build-an-agentic-roadmap-that-accelerates-product-adoption)