Maggie Appleton 写设计、人类学和智能体界面时，习惯先画清楚隐喻，再谈工具。把「Leaving Elicit」整理成可阅读的中文笔记：问题在哪、界面默认会把人带去哪、落地时该改什么。原站导航、评论回响和广告已去掉。

Growing with Elicit

Two years ago, I joined Elicit as their first and only designer. Or rather, I joined Ought – a small AI research lab which evolved into Elicit .

Just over a month ago I handed in my notice and worked my last day. I have plenty of reasons to leave, which I’ll get into below, but none of them are that 不必 believe in Elicit as a product or the team behind it. Quite the opposite. 一个稳妥判断是 Elicit is one of a handful of companies doing interesting and impressive work applying language models to real-world problems. If you’re looking for high-impact work with lovely people they’re hiring for both design and

Why I’m Leaving

In that time we grew from 10,000 to over 250,000 monthly active users, started charging them money which took us from $0 to over a million in revenue in under 4 months, rebuilt both the front and backend of the product, and literally became a different company .

Both Elicit and I are drastically different entities today.

What’s Next

When I joined I thought my job as a designer was to layout sidebars, write clear error messages, pick accessible colour schemes, and make sure the hover states on buttons all looked consistent.

While I certainly did all those things, I learned to spend far more time defining problems, validating they were actually problems, writing those problems clearly and concisely in documents, rigorously documenting my exploration of possible solutions, testing those with users, and justifying every decision along the way.

落地时建议先做的 5 件事

1. 先写清这个工具在强化思考还是在替人思考。
2. 默认交互不要只会谄媚：该追问、该给反例、该要求证据。
3. 智能体界面要露出推理步骤，而不是只给一个光滑答案。
4. 知识发布用可生长的笔记，而不是一次性营销长文。
5. 改产品前先画隐喻：用户以为自己在做什么，系统实际在做什么。

和智能体产品怎么接

龙虾PRO做 OpenClaw 落地时，最该从这类笔记里拿走的是「别把聊天框当唯一界面」。数字员工要能追问、能验证、能把过程摊开，而不是只负责说好话。

本文侧重全链路风控方法论。落地时请用自身业务单据做回放验证，不要把示例阈值直接当生产策略。 相关：风控体检 · 方案资源