返回 Skill 列表
extension
分类: 内容与媒体无需 API Key

nlx-design-stack

在设计AI代理或聊天机器人时使用,当原始模型输出感觉像一个黑盒子时,或者在构建需要结构和信任的自然语言接口时

person作者: jakexiaohubgithub

NLX (Natural Language Experience) Design Stack

Overview

A framework defining the new principles and UI constructs required for Natural Language Interfaces. Moves beyond simple chat to structured, designed conversations.

Core principle: NLX is the new UX. Conversations have grammars, structures, and invisible UI elements.

The Four Elements

┌─────────────────────────────────────────────────────────────────┐
│                      NLX DESIGN STACK                           │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│   ┌──────────────┐   ┌──────────────┐                           │
│   │   PROMPTS    │   │    PLANS     │                           │
│   │              │   │              │                           │
│   │  New UI      │   │  Structured  │                           │
│   │  Construct   │   │  Editable    │                           │
│   │  (Input)     │   │  Outputs     │                           │
│   └──────────────┘   └──────────────┘                           │
│                                                                  │
│   ┌──────────────┐   ┌──────────────┐                           │
│   │   PROCESS    │   │  FOLLOW-UPS  │                           │
│   │              │   │              │                           │
│   │  Show the    │   │  Guided      │                           │
│   │  Work /      │   │  Navigation  │                           │
│   │  Thinking    │   │  Suggestions │                           │
│   └──────────────┘   └──────────────┘                           │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Key Principles

| Element | Principle | |---------|-----------| | Prompts | Treat prompts as distinct UI elements (like dropdowns) | | Plans | Return structured, editable plans for complex goals | | Process | Visualize reasoning to build trust ("thinking aloud") | | Follow-ups | Proactively suggest next steps ("happy path" guidance) |

Common Mistakes

  • Assuming "the model is the product" (neglecting design)
  • Making the AI a black box without feedback
  • Failing to guide users after an output

Source: Aparna Chennapragada (Microsoft CPO) via Lenny's Podcast