AWS 发现工程师 onboarding 效率低、缺乏个性化。我作为研究负责人，带领 4 人团队完成从用户调研到产品原型的全流程，最终交付 Neoboard —— 一个由 AI 驱动的 onboarding 工作台。
Based on company confidentiality, sponsors requested that the project scope be expanded to the entire U.S. technology industry.
企业 onboarding 质量参差不齐，HR 资源不足、流程缺乏个性化，新入职 SWE 难以快速融入团队、掌握内部工具。
研究负责人，带领 4 人团队；主导调研设计、insight 提炼，并将研究结论转化为产品决策。
交付 Neoboard 原型，80% 用户满意功能，87.5% 认可易用性，83% 认为产品达到预期。
核心矛盾：onboarding 对新员工至关重要，但企业却把它交给最忙的人来做。
The critical needs of new hires are informally delegated to overwhelmed hiring managers — creating a system that routinely fails the people it's meant to support.
The challenges of implementing high-quality onboarding programs are significant, particularly due to the stress and burnout experienced by HR management leaders. Many organizations struggle with the resources and time required to develop effective onboarding. Despite these difficulties, the importance of well-structured onboarding cannot be understated — it impacts new hires, their teams, and the entire organization.
"During the rest of the new hire's first year, firms leave it to overwhelmed and under-prepared hiring managers to address informally the critical needs that all new hires face."
Stein and Christiansen, 2010
AWS recognized these patterns and posed the challenge: How might we better support SWEs onboard to a new team by leveraging generative AI as a core component?
我们设计了三层研究结构：用问卷建立量化基线，用访谈挖掘深层原因，用日记研究捕捉真实工作流。访谈横跨 Apple、Amazon、Google、Microsoft、HubSpot 的 8 位工程师。
We ran stakeholder surveys, 8 in-depth interviews with engineers across Apple, Amazon, Google, and Microsoft, and a one-week diary study with a new-grad SWE at Workday.
We mapped the onboarding journey across five stages to identify where friction peaks and where AI could intervene most effectively.
设计原则：不替代人，而是让人在正确的时机得到正确的帮助。
AI-Driven Onboarding System: Quick and Easy Adaptation for New Software Engineers.
Before users land on the dashboard, we offer a personalization opportunity to customize the features they would like to see in their dashboard.
研究发现每个人起点不同 → 第一步让用户定义自己需要什么，而不是被迫接受统一流程。
Users can quickly see the most recent tasks that need to be completed, the specific progress of a certain task, and quick reading pages that can help users access them.
新员工每天最大的问题是"我今天该做什么" → 用一个视图解决优先级焦虑。
After toggling the "Linker View" option, it guides users to read related documents in sequence and provides AI-generated summaries from related documentation. Offer other related documents as optional reading.
文档散乱是头号痛点 → AI 自动建立文档之间的关联，把碎片化信息变成可跟随的阅读路径。
Gen AI provides guides for fixed onboarding tasks based on feedback from past new hires. It also recommends improvement tasks during the early employment phase. Some tasks are AI-generated — the user's leader will flag anything needing attention.
早期新员工有热情但没有方向 → AI 生成可操作的任务引导，同时保留人工审核入口，避免 AI 幻觉带来误导。
All task processes are automatically recorded. Neo Report generates suggestions based on time spent, knowledge gaps, and common errors — and gets smarter as more tasks are completed.
新员工很难自我发现知识盲区 → 用完成数据反推薄弱环节，把进度可视化变成主动学习的工具。
The user can ask any doubts or queries to the AI chatbot. If the answer is not sufficient, the chatbot maps relevant experts or colleagues for follow-up. All questions are collected to continuously improve the system.
访谈中最常见的场景是"我不知道该问谁" → AI 先答，不够再匹配人，同时积累问题数据库持续优化。
We ran usability testing with 4 SWEs and 2 MS-CS students, plus a separate expert evaluation with 1 ML engineer and 2 MS-HCI researchers using cognitive walkthroughs and Nielsen's 10 heuristics.
专家评测之后，我们做了三个优先级最高的迭代：UI 元素尺寸、图标可识别性、功能间的联动逻辑。
Expert evaluation identified UI sizing inconsistencies, icon ambiguity, and weak cross-feature integration. These became our top three iteration priorities before the final prototype review.