Design patterns for AI-native interfaces, conversational UX, spatial computing (AR/VR/MR), voice interfaces, multimodal interaction design, and emerging interaction paradigms for 2025-2030. Use when the user mentions: AI interface, voice UX, spatial design, AR/VR, conversational UI, chatbot design, mixed reality, multimodal, AI UX, spatial computing.
Installation
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Design patterns for AI-native interfaces, conversational UX, spatial computing (AR/VR/MR), voice interfaces, multimodal interaction design, and emerging interaction paradigms for 2025-2030. Use when the user mentions: AI interface, voice UX, spatial design, AR/VR, conversational UI, chatbot design, mixed reality, multimodal, AI UX, spatial computing.
AI interfaces fundamentally differ from traditional deterministic UIs. Traditional interfaces execute exactly what users request; AI interfaces interpret, generate, and sometimes surprise. This creates new UX challenges around trust, control, transparency, and error recovery that require fresh design patterns.
Core AI Interface Principles
Predictability in unpredictability — AI output varies, but the interface behavior around it must be consistent
Graceful degradation — when AI fails or produces poor results, the experience must remain usable
Progressive trust — earn user trust incrementally; do not assume it
Human in the loop — always provide mechanisms for human oversight, correction, and override
Transparency over magic — explain what the AI is doing without exposing technical complexity
AI Chat and Conversational Interface Patterns
Message Design:
Differentiate human and AI messages through visual treatment (alignment, color, avatar)
AI messages should stream/type progressively — it communicates processing and allows early reading
Show "thinking" or "processing" indicators during AI computation
Support rich content in AI responses: formatted text, code blocks, tables, images, interactive elements
Input Patterns:
Multi-line text input with Shift+Enter for newlines, Enter to send
Suggested prompts and conversation starters for discoverability
File and image attachment for multimodal input
Voice input as an alternative to typing
Conversation history with search and reference capability
Context and Memory:
Show what context the AI has access to (conversation history, documents, tools)
Allow users to explicitly provide or revoke context
Indicate when AI is using information from previous conversations
Support conversation branching for exploring alternative responses
Output format preferences (bullet points, paragraphs, tables)
Domain/role context setting (act as expert in X)
Safety and content filter controls with clear explanations
Spatial Computing UX (AR/VR/MR)
Spatial Design Principles
Respect physical space — digital elements must coexist naturally with the physical environment
Spatial hierarchy — distance from user communicates importance (near = primary, far = peripheral)
Comfort first — never cause disorientation, nausea, or visual strain
Body-aware interaction — design for natural human posture, reach, and movement
Environment-adaptive — interfaces must respond to lighting, surfaces, and obstacles
AR Interface Patterns
Spatial Anchoring:
Anchor UI elements to real-world surfaces (tables, walls) when contextually appropriate
Keep frequently-accessed controls anchored to the user's field of view (head-locked)
Use world-locked placement for persistent information (room labels, equipment status)
Ensure anchored content does not obstruct critical real-world visual information
Visual Design for AR:
Use semi-transparent panels with blur backgrounds — maintain awareness of real environment
High contrast text and icons — AR competes with unpredictable real-world lighting
Avoid pure white backgrounds — they wash out in bright environments
Dark UI with light text works better in varied lighting conditions
Minimum text size: 1.5-2x standard screen sizes due to rendering at distance
AR Interaction:
Gaze + pinch (Apple Vision Pro pattern) for selection
Hand tracking for direct manipulation
Voice commands for hands-free operations
Spatial audio cues for off-screen elements
Ray casting for distant object selection
VR Interface Patterns
Comfort and Safety:
Maintain stable horizon line — never tilt the world
Provide a fixed reference frame (virtual floor, cockpit, room) to prevent motion sickness
Field of view tunneling during movement reduces motion sickness
Maximum movement speed should be comfortable; offer teleportation as alternative
Rest-state UI: elements at 1-2 meter distance, slightly below eye level (ergonomic neutral)
VR Layout:
Primary content: centered, 1-2m from user, slight downward angle
Secondary panels: curved arrangement at 120-160 degree arc around user
Avoid placing content behind the user — require explicit turning
Maximum comfortable horizontal eye sweep: plus/minus 30 degrees from center
Maximum comfortable vertical eye range: +20 degrees to -40 degrees from eye level
VR Text and Readability:
Minimum angular text size: 1.5 degrees of visual angle (roughly 3x screen sizes)
Use bold, high-contrast fonts — thin strokes render poorly in VR headsets
Limit text to short labels and essential information — spatial UIs favor visual communication
Panel curvature: match content panels to a curved surface at consistent distance
Voice Interface Design
Voice UX Principles
Be conversational — match natural speech patterns, not command-line syntax
Keep it brief — voice responses must be shorter than visual text equivalents
Confirm ambiguity — repeat back unclear inputs for confirmation
Provide escape hatches — "cancel", "go back", "start over" must always work
Progressive disclosure via voice — offer summaries with option to hear details
Voice Interaction Patterns
Invocation and Wake:
Clear wake word or button activation
Visual + audio indicator that the system is listening
Timeout with gentle prompt: "Are you still there?"
Dialogue Design:
Present maximum 3 options in voice menus — more causes memory overload
Use confirming responses: "Okay, setting a timer for 10 minutes"
Offer correction mechanisms: "Did you mean...?"
Support barge-in (interrupting the system while it is speaking)
Error Recovery:
"I didn't catch that. Could you say it again?" (friendly, specific)
Escalation path: after 2-3 failures, offer alternative input method or human help
Never respond with silence — even "I'm not sure how to help with that" is better than nothing
Multimodal Voice + Screen:
Voice for input, screen for output works best for complex information
Show visual confirmation of voice commands
Allow switching between voice and touch seamlessly
Voice should complement touch, not replace it entirely
Multimodal Interaction Design
The Multimodal Stack (2025-2030)
Touch + Voice: Common on phones and smart displays
Gaze + Hand + Voice: Spatial computing (Vision Pro, Quest)
Keyboard + Mouse + AI: Desktop productivity
Gesture + Voice + Touch: Automotive and kiosk interfaces
Brain-Computer Interface (emerging): Assistive tech, early consumer exploration
Multimodal Design Rules
Each modality should be self-sufficient for critical tasks — never require a specific modality
Complement modalities: voice for command, visual for confirmation, haptic for feedback
Resolve conflicts between simultaneous inputs gracefully (voice says "blue" while touching "red")
Adapt interface emphasis based on active modality (larger touch targets when touch is primary)
Design Patterns for Emerging Tech
Ambient Computing
Interfaces that appear when needed and disappear when not
Context-aware: adapt based on time, location, activity, and user state
Proactive suggestions without being intrusive — the design challenge of the decade
Respect boundaries: ambient does not mean always-on surveillance
Generative UI
Interfaces that adapt layout and content based on AI understanding of user intent
Dynamic form generation based on the user's specific task
Personalized dashboards that surface relevant information automatically
Challenge: maintain consistency and learnability while personalizing
Cross-Referencing
For interaction and motion patterns, reference interaction-motion-design
For accessibility in emerging platforms, reference accessibility-inclusive-design
For ethical AI design, reference ux-ethics-content-strategy
For design system adaptation to spatial platforms, reference design-systems-architecture
v3.0 Cross-References
The v3.0 upgrade adds new skills and references that deepen and extend AI, voice, and spatial interaction coverage.
Conversational AI Dialogue Patterns
See agentic-ai-generative-ux/references/conversational-ai-dialogue-patterns.md for comprehensive conversational AI interaction design patterns including dialogue state management, streaming UX for LLM responses (token-by-token rendering, progressive disclosure of long outputs), AI persona design principles (tone calibration, personality consistency, brand voice alignment), and memory transparency patterns (showing users what context the AI retains, selective memory controls, conversation history management). This reference extends the dialogue design section above with production-level patterns specific to LLM-powered conversational interfaces.
Ambient & Calm Computing
See the ambient-calm-zero-ui skill for dedicated coverage of ambient computing, calm technology principles (Weiser & Brown), screenless interaction paradigms, and zero-UI design patterns. This skill complements the "Ambient Computing" section above with structured frameworks for proactive, non-intrusive interfaces — including notification philosophy for ambient systems, peripheral attention design, and context-aware information surfacing without screen dependency.
Cognitive Psychology for Voice and Spatial Interfaces
See the cognitive-psychology-ux skill for foundational cognitive science that directly impacts voice and spatial interface design — particularly Miller's Law constraints on voice menu options (the 3-option maximum for voice menus derives from working memory limits), attention models relevant to spatial UI placement (Treisman's feature integration theory for AR overlay design), cognitive load theory for multimodal input (split-attention effect when combining voice, gaze, and gesture), and Hick's Law implications for voice command discoverability.
Key Sources
Apple Vision Pro Human Interface Guidelines (2024)
