Skip to content

sergiovazquezdml/Base-de-datos-aluminio / .agents/skills/agentic-ai-generative-ux

Agentic AI & Generative UX

Verified

Design patterns for agentic AI systems, multi-agent orchestration, generative UI, RAG interfaces, AI safety guardrails, trust calibration, and the paradigm shift from tool-based to agent-based interfaces. Covers AI copilot frameworks, prompt engineering UX, and AI anti-patterns.

references/ai-native-interface-patterns.md

View on GitHub
# AI-Native Interface Patterns

## Overview

AI-native interfaces are those designed from the ground up around AI capabilities rather than retrofitting AI onto traditional UI patterns. This reference covers the foundational patterns — chat, copilot, inline suggestion, generative UI, and autonomous agent interfaces — and the emerging paradigm shifts that are reshaping how interfaces mediate between humans and AI systems.

---

## Core AI Interface Archetypes

### Chat / Conversational Interface

The most familiar AI interface pattern. A turn-based conversation where the user provides input (text, voice, images) and the AI responds. Effective for open-ended exploration, question-answering, and creative generation. Limitations: poor for structured data entry, spatial manipulation, and tasks requiring persistent visual context.

### Copilot / Inline Suggestion

The AI observes the user's work in real time and offers contextual suggestions inline — code completions, writing continuations, formula suggestions, design recommendations. The user accepts, modifies, or dismisses each suggestion. Effective when the AI has strong context from the surrounding work. Requires careful attention to non-intrusiveness: suggestions must be visually distinct from user content and dismissible without friction.

### Generative UI

The AI generates interface elements dynamically based on context — forms, charts, cards, interactive components — rather than returning plain text. The interface adapts its structure to the content. Requires a robust component library and careful sandboxing to prevent malformed or insecure UI generation.

### Autonomous Agent

The AI operates independently on multi-step tasks with periodic user oversight. The interface emphasizes plan visibility, progress tracking, approval gates, and rollback capability rather than turn-by-turn conversation. See `ai-safety-trust-guardrails.md` for the control/consent/accountability framework.

---

## The Post-UI Paradigm (2026)

### UI Is No Longer a Differentiator

NNGroup's 2026 research delivers a provocative finding: as AI handles an increasing share of user interaction, traditional UI is becoming a commodity. The argument is structural: when users primarily interact through natural language, voice, or autonomous agents, the visual interface layer matters less as a differentiator than the quality of the AI orchestration behind it. Two applications with identical AI capabilities but different visual designs will converge in user satisfaction scores, because the AI interaction — not the pixel-level UI — drives the experience.

This does not mean UI is irrelevant. It means the role of UI is shifting from being the product experience to being the orchestration layer between the user and the AI. The interface exists to show what the AI is doing, explain why, and give the user meaningful control.

### From Information Display to AI Orchestration

Traditional UI design optimized for information display: presenting data, enabling navigation, facilitating input. These remain necessary but are no longer sufficient. AI-native interfaces must excel at a fundamentally different set of tasks:

- **Showing what the AI is doing.** When an agent is executing a multi-step plan, the interface must make the plan visible, show current progress, and indicate what comes next. Opacity into AI behavior is the new form of "good information architecture."
- **Explaining why.** Every AI decision, recommendation, or action should be accompanied by accessible reasoning. This is not just an XAI requirement — it is the interface's primary value proposition. Users choose the product that helps them understand the AI, not just the one with the prettiest layout.
- **Giving control.** The ability to pause, redirect, constrain, and override AI behavior must be surfaced through the interface with the same prominence that navigation and search received in the pre-AI era. Control affordances are the new primary navigation.

### New Differentiators

If traditional UI is commoditized, what differentiates AI-native products? Four capabilities emerge as the new competitive surface:

**Trust calibration.** The interface's ability to help users develop an accurate mental model of when the AI is reliable and when it is not. Products that calibrate trust well retain users; products that either over-promise (leading to trust collapse after failure) or under-communicate (leading to under-utilization) lose them. See `ai-safety-trust-guardrails.md` for trust calibration patterns.

**Transparency.** Making AI reasoning, data sources, confidence levels, and limitations visible without overwhelming the user. The three-tier progressive explanation model (summary, detail, technical) becomes the standard design pattern. Transparency is not optional — it is the EU AI Act's legal requirement and users' psychological requirement.

**Error recovery.** When AI fails — and it will — the interface must recover gracefully: acknowledge the error, explain what happened, offer a corrected output, and provide increased oversight temporarily. Error recovery quality is the strongest predictor of long-term user retention in AI products.

**Handoff between AI and manual modes.** Users must be able to seamlessly switch between AI-assisted and fully manual modes for any task. The interface must make this transition frictionless: "Let me do this myself" should be a single action that preserves all context. The reverse transition ("AI, take over from here") should be equally smooth, with the AI picking up from the user's current state without requiring re-explanation.

### The "AI Slop" Quality Problem

As AI generates content at scale, a quality crisis emerges: mediocre AI-generated outputs flood every surface. NNGroup 2026 documents this as the "AI slop" problem — technically correct but substantively empty content that degrades user experience through sheer volume and blandness.

When AI can generate a million passable outputs, the differentiator is not generation capability but curation quality. Interfaces must implement quality gates that reject outputs below a defined standard rather than presenting everything the AI produces. Human oversight becomes the UX differentiator: products that invest in quality curation — whether through human review, AI-powered quality scoring, or hybrid approaches — will outperform products that simply surface raw AI output.

Design implications: build quality scoring into the generation pipeline, show quality indicators to users, and provide easy "regenerate" and "refine" actions when quality falls short. Never present AI output as final without either automated quality validation or user review.

> For quality gate implementation patterns, hallucination detection UX, and confidence-based content filtering, see `llm-hallucination-design-guardrails.md`.

---

## Cross-References

- For trust calibration, guardrail design, and confidence visualization patterns, see `ai-safety-trust-guardrails.md`.
- For conversational dialogue patterns and agent interaction flows, see `conversational-ai-dialogue-patterns.md`.
- For multi-agent orchestration and plan visibility patterns, see `agentic-multi-agent-patterns.md`.
- For generative UI and RAG integration patterns, see `generative-ui-rag-patterns.md`.

## Key Sources

- NNGroup: State of UX 2026 ("UI as Commodity," "AI Slop," Trust Calibration)
- Smashing Magazine: Agentic AI Design Patterns (February 2026)
- Google PAIR: People + AI Guidebook
- Microsoft HAX Toolkit: Guidelines for Human-AI Interaction
- EU AI Act: Regulation (EU) 2024/1689
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall