# Agentic UX (AX) Design Patterns

> Agentic UX is lifecycle design for supervised delegation: users state goals, agents propose plans and act over time, and interfaces must align before action, show work in progress, make recovery explicit. Curated by Daniel Albinsson

This resource is for product designers and engineers implementing agent surfaces.
Each linked `.txt` file is self-contained implementation guidance (definition, anatomy,
limitations, do/avoid, build notes, production examples with critique). Interactive mockups
live at `/patterns/{id}`.

## Use with your coding agent

Before implementing agent UI, read the relevant pattern file. Prefer these specs over
inventing new interaction models.

```text
1. Index: https://agentic-ux.com/llms.txt
2. Pattern: https://agentic-ux.com/llms/patterns/{pattern-id}.txt
```

Local dev: reference `public/llms.txt` and `public/llms/patterns/{id}.txt` directly
(e.g. Cursor `@` file references). Ask the agent to cite limitations and anti-patterns.

**Lifecycle framework**

- Before the agent acts: scope, plan, permissions
- While the agent works: work in progress, rationale, uncertainty
- After the agent acts: audit, explicit recovery, escalation

**Topics:** Agentic UX, AI agent UI, human-in-the-loop, intent preview, autonomy controls,
progress transparency, explainability, undo, escalation, generative UI

## Before the agent acts

- [Intent-First Interface](https://agentic-ux.com/llms/patterns/intent-first.txt): Goal-first entry screen
- [Intent Preview](https://agentic-ux.com/llms/patterns/intent-preview.txt): Show an approvable plan before work starts
- [Clarifying Questions](https://agentic-ux.com/llms/patterns/clarifying-questions.txt): Scope-narrowing questions before the plan
- [Autonomy Dial](https://agentic-ux.com/llms/patterns/autonomy-dial.txt): Policy controls plus runtime Allow/Deny for new scopes

## While the agent works

- [Explainable Rationale](https://agentic-ux.com/llms/patterns/explainable-rationale.txt): Expandable why for key decisions
- [Progress Ledger](https://agentic-ux.com/llms/patterns/progress-ledger.txt): Live step list for long runs
- [Generative UI](https://agentic-ux.com/llms/patterns/generative-ui.txt): Task-specific generated controls
- [Sandbox Preview](https://agentic-ux.com/llms/patterns/sandbox-preview.txt): Safe preview before promote

## After the agent acts

- [Action Audit & Undo](https://agentic-ux.com/llms/patterns/action-audit.txt): History plus recovery paths
- [Escalation Pathway](https://agentic-ux.com/llms/patterns/escalation-pathway.txt): Human handoff with context
- [Return Moment (Smart Briefing)](https://agentic-ux.com/llms/patterns/return-moment.txt): Briefing when user returns


## Site

- [Framework article](https://agentic-ux.com/framework): Full supervised-delegation narrative
- [About](https://agentic-ux.com/about): Daniel Albinsson
- [Crosswalk](https://agentic-ux.com/references): Map to Shape of AI and IF catalogue
- [Guide](https://agentic-ux.com/guide): Practitioner playbook: agent-flow audit, lifecycle checklists, symptom-based starting points