Designing delight for a complex AI tool为复杂 AI 工具探索 Delight
I facilitated a six-week design sprint that helped a cross-functional team move from an open question—what could “delight” mean in enterprise AI?—to four product-grounded directions and reusable design insights. 我主持了一次六周 Design Sprint，带领跨职能团队把一个开放问题——企业 AI 的 “delight” 可以是什么——收敛成四个贴近产品的方向与一组可复用设计洞察。
AzureML’s FY22 goal was to improve productivity and user experience for complex pipeline authoring and debugging. The team already knew how to add features; the harder question was how visual and interaction design could create differentiation without disrupting expert workflows. AzureML 在 FY22 的目标之一，是提升复杂 pipeline authoring 与 debugging 的效率和体验。团队知道如何增加功能；更难的问题是，视觉与交互如何在不打扰专家工作流的前提下创造差异化。
Several participants had no AzureML background. Too much onboarding would constrain imagination; too little would leave the group lost. I used subtraction rather than a full product tour: 部分参与者没有 AzureML 背景。讲得太多会限制想象，讲得太少又会让团队失去方向。我没有做完整产品培训，而是用“减法”：
- Start with a familiar analogy. I compared AzureML canvas authoring with Figma: open a canvas, place objects, connect or configure them, iterate.从熟悉场景开始。我把 AzureML canvas authoring 类比为 Figma：打开画布、放置对象、连接或配置、持续迭代。
- Keep only the happy path. Participants learned the asset library, canvas, settings panel, and submit flow—not every command.只保留 happy path。参与者只需理解 asset library、canvas、settings panel 与 submit，不必认识每个命令。
- Make information volume visible. Simplified screens retained realistic content density so later concepts still responded to the real product.保留真实信息量。简化界面仍维持真实内容密度，让后续概念不会脱离产品。
The proposals explored how color, hierarchy, node shape, grouping, minimaps, responsive detail, and connection treatment could make a complex graph easier to understand. The goal was not to pick the prettiest screen; it was to discover which ideas made the system feel clearer, more trustworthy, and more flexible. 提案探索了颜色、层级、node 形态、分组、minimap、响应式信息量与连接关系等方向。目标不是选出“最好看”的画面，而是找到哪些设计能让系统更清晰、更可信、更灵活。
I structured reviews by perspective instead of asking everyone the same “Do you like it?” question: 我没有让所有人回答同一个“喜欢吗？”，而是按角色设计反馈问题：
- Designers: Does the proposal express the intended feeling and visual principles?设计师：方案是否表达了预期感受与视觉原则？
- PMs: Does it support product goals and real user value?PM：是否符合产品目标并产生真实用户价值？
- Engineers and expert users: What is technically risky, behaviorally confusing, or worth carrying forward?工程师与专家用户：哪些点有实现风险、行为不清晰，哪些值得继续？
The sprint converged on four polished visual directions, but its more durable output was the insight library behind them: dozens of opportunities across color, typography, texture, shape, layout, interaction, and overall feeling. That gave the product team ideas they could reuse feature by feature instead of adopting one wholesale redesign. Sprint 最终收敛成四个视觉方向，但更持久的产出是它们背后的洞察库：覆盖颜色、文字、材质、形态、布局、交互与整体感受的数十个机会点。产品团队无需一次性采用整套 redesign，也可以按 feature 逐步复用。
