Lanlang

嘻嘻～

[Lanlang @laaaaaandy]

刚玩这个，不懂怎么弄，关住丝我就回的

[Lanlang @laaaaaandy]

刚拍的！！ 刚玩X

开脱！！！！! !

上供！！ ！

刚拍的新鲜出炉的照片or视频！

[Lanlang @laaaaaandy]

@zintjmbutsgmg4 来

[Lanlang @laaaaaandy]

～高考结束啦！嘿嘿～

刚玩～新人刚接触X ！

有没有人要锐评17岁的福呢？

[Lanlang @laaaaaandy]

@nwzlxdsbfx90wx 不是人机吧？

[Lanlang @laaaaaandy]

当然都有啦！丝我就行！关优先！

[Lanlang @laaaaaandy]

1/ 🧵 The magical @: “what do AI products assume users already know?”

AI tools are built by people who know them inside-out—but the users don’t. This gap often leads to hidden assumptions that confuse, frustrate, or even alienate users. #AI #UXDesign

2/ For example, small interface details—like using @ for mentions, or a particular icon for “settings”—may feel obvious to the designer. But for first-time users, it’s not intuitive. #UserExperience

3/ These “assumed knowledge” gaps are invisible until someone stumbles. Users may feel anxious, embarrassed, or simply give up. The result? Less trust in the product. #AIUX

4/ Good design isn’t just about cool features. It’s about aligning the product’s mental model with the user’s mental model. Anticipate what people don’t know. Teach without being patronizing. #DesignThinking

5/ When AI tools fail to account for this, even advanced tech can feel magical in the wrong way—more confusing than empowering. Awareness of these assumptions is key to accessibility and adoption. #HumanCenteredDesign

6/ TL;DR: Ask yourself—what does my AI product assume users know? The answer will make your UX clearer, friendlier, and more trustworthy.

[Lanlang @laaaaaandy]

🚀 I created 100 posters with Google’s Nano Banana🍌.

Sometimes the best way to explore a new tool is to push it to extremes. So I gave Google’s Nano Banana a challenge:

Here they are, arranged into three big canvases:

Which one is your favorite? 👀

Let me know in the comments if you want the prompts!

[Lanlang @laaaaaandy]

Building Agentic Systems Feels Effortless — Until It Doesn’t

1/ 🧩 AI agent diagrams look simple:
Understand → Plan → Execute → Evaluate → Repeat
Looks neat, right?
Reality: chaos. 😅

2/ 6 months building agentic systems taught me this:
Agents are easy to imagine… hard to make work in the real world.

Here’s why 👇

3/ Macro view: scheduling a meeting ✅ easy
High-value work ❌ messy:
- Draft proposals
- Run training sessions
- Manage client comms
- Coordinate complex projects

Agents stumble here.

4/ Real workflows are messy:

- Agents inherit half-finished tasks
- Humans jump in midway
- Outputs get interrupted → agent must re-plan

5/ Efficiency is the silent killer ⏱️

Every human feedback triggers:
Re-understand → Re-plan → Re-execute → Re-evaluate

Time piles up. Patience disappears.

6/ Design challenge = balancing 3 forces:

1️⃣ Quality (is it good enough?)
2️⃣ Completeness (task done?)
3️⃣ User time (tolerable?)

Tradeoffs everywhere. That’s why agents are deceptively hard. ⚡

7/ Micro view: each step = a minefield 💣

8/ Mini story 📝

User wanted one slide with EVERYTHING.

Agent laid it out → slide broke. Fonts tiny, spacing collapsed.

Choices:
- Keep details or simplify?
- Redo or hand back “good enough”?
- Stick to one slide or invent a new layout?

No universal answer. Tradeoffs everywhere.

9/ Other messy steps:

Goals shift mid-process 🎯

Planning = dynamic 🛠️
Execution = humans edit in parallel 👥
Evaluation = context matters 🧐
Feedback = vague (“Make it pop”) 💬
Final output = never final 🔄

10/ Takeaway:

Workflows = messy
Collaboration = constant
Feedback = vague
Success = contextual
Agents = easiest to prototype, hardest to deploy. 🚀

11/ Builder tip 🛠️

Ask:
- Can your agent handle incomplete/midstream tasks?
- Can it adapt when goals shift?
- Does it work iteratively with humans?
- Can it process vague feedback without collapsing?
- Feedback is the hardest step. 🧠

12/ Your turn:

Which step in your agent loop has been the biggest challenge? 👇

[Lanlang @laaaaaandy]

I just published Why Building Agentic Systems Feels Effortless — Until It Doesn’t https://t.co/mRLkmvDw9d