Phu (@LongPh8) header
Phu (@LongPh8) profile picture

longphu.sui longphu.eth

Joined January 2018
Photo by Phu (@LongPh8)Photo by Phu (@LongPh8)Photo by Phu (@LongPh8)Photo by Phu (@LongPh8)Photo by Phu (@LongPh8)Photo by Phu (@LongPh8)Photo by Phu (@LongPh8)Photo by Phu (@LongPh8)Photo by Phu (@LongPh8)Photo by Phu (@LongPh8)
I fixed 55 bugs by making my monorepo more agent-friendly this week💪 ・Heavily inspired by @poteto's recent talk ・Grouped the code by feature ・Let lint enforce the boundaries ・Used @cnakazawa's anti-slop lint rules ・Write tests before each fix, which caught bugs🐛
5
2
193
6,382
For a while I have been obsessed with the idea that there is a fundamental difference between "context" as we have grown to understand it and "knowledge", which is what we actually prize and desire and what translates into real-world outcomes. It's become clear to me that as the volume and scope of work delegated to AI grows, legacy systems of context management will not be able to scale cleanly, no matter how much we try to use AI to speed things up. In domains that are not pure code, it is an asymmetric race where the slop generators are always outpacing the context curators. There is no way to break this formation. Unless you shift your perspective and move up a plane. Legacy static context bootstraps synthesized knowledge. Knowledge grows organically and dynamically with rich environmental feedback. It becomes "real" in a sense that legacy context is often not and so can be evaluated and verified and improved and maintained in a truly autonomous and scalable fashion. Data teams and curators do not lose control in this vision of the world; they develop increased leverage, their signal-to-noise ratio improves, and everyone (and their agents) coalesces around the outcomes that matter —knowledge— instead of intermediate busywork. This is a long-term utopian vision for what data work (maybe all knowledge work!) might become, and there is a LOT of work to do, but I'm extremely proud of Hex for taking the first bold steps towards this with our new Knowledge Engine. It's live now for anyone using Hex 4.0. Today, it will immediately improve the accuracy and efficiency of your data work. Tomorrow when you wake up, it will be even better. It will not go stale or fall out of sync with reality. Give it a little breadcrumb, and it will reconcile the entire workspace to fit your desires. "Leverage, says you. I think I feel a change in the wind, says I!" hex.tech/blog/knowledge-engi…
25
26
6
276
35,449
Phu retweeted
SpaceX engineer Lauren Tan just made prompt engineering useless She doesn’t give you step-by-step instructions She gives you the system around them And it’s totally FREE: 1/ Clarify the goal Before anything, define what "done" looks like, a vague goal gets you vague results, no matter how good the prompt is 2/ Break the process into steps Don’t hand it the whole job at once, show it the order you do things in 3/ Say the quiet part out loud Write down the decisions you make in your head without noticing, and the rules you never break 4/ Build a feedback loop Every time it gets something wrong, correct it, and make sure it remembers, every correction makes the next run better. 5/ Put it on a routine Once it works, schedule it, then it runs without you even asking. No template can do this for you, It depends on your work, your standards and your taste You teach the bot your judgment bit by bit, and feedback takes it from "usable" to "excellent" For solo operators, the formula is simple: know-how for your specific work + long-term memory + bots that work together So stop hunting for the perfect prompt Do it first, then learn to do it right:
All code is slop, except mine. You’re writing code with AI WRONG, and it’s turning into SLOP. Here’s how to code with agents the right way: the system behind Grok Bot from Lauren Tan, plus my own tips.
6
18
228
15,853
Phu retweeted
this is f*cking gold for Fable 5.2 20 GitHub repos do all the clicking. You just send one brief 01 DaVinci Resolve MCP → Claude drives DaVinci Resolve github.com/samuelgursky/davi… 02 auto-editor → cuts the silence out for you github.com/WyattBlue/auto-ed… 03 PySceneDetect → finds every scene cut github.com/Breakthrough/PySc… 04 MoviePy → video editing in Python github.com/Zulko/moviepy 05 Remotion → edit video as React code github.com/remotion-dev/remo… 06 video-use → raw footage in, final.mp4 out. Cuts fillers, grades, burns subtitles github.com/browser-use/video… 07 Magic MCP → like v0, inside Claude Code github.com/21st-dev/magic-mc… 08 Figma Context MCP → Figma layouts straight to your agent github.com/GLips/Figma-Conte… 09 Penpot → open-source Figma alternative with an official MCP github.com/penpot/penpot 10 agent-browser → the agent opens the page and checks the design itself github.com/vercel-labs/agent… 11 ui-ux-pro-max → 80 UI styles, palettes and font pairings github.com/nextlevelbuilder/… 12 /canvas-design → Anthropic's skill for posters and art as PNG and PDF github.com/anthropics/skills… 13 /pptx → real slide decks. Anthropic official github.com/anthropics/skills… 14 /theme-factory → 10 ready themes for slides, docs and landing pages github.com/anthropics/skills… 15 stitch-skills → Google Labs skills that turn a prompt into UI screens github.com/google-labs-code/… 16 /frontend-design → Anthropic's skill for UI that doesn't look AI-generated github.com/anthropics/skills… 17 DESIGN.md → Google's format that describes your visual identity to agents github.com/google-labs-code/… 18 Astryx → Meta's open-source design system, built for agents github.com/meta/astryx 19 Excalidraw → hand-drawn diagrams and wireframes github.com/excalidraw/excali… 20 tldraw → an infinite canvas SDK github.com/tldraw/tldraw Two pro apps on autopilot, and the setup is $0 Bookmark it and send your next brief instead of opening an editor The full motion design course for Fable 5.2 is waiting in the article below ↓
Fable 5.2 can make a film from one line It just won't be your film One line gets you Fable's taste. A director's brief gets you yours From nothing installed to a finished MP4: every prompt, every number, every fix ↓
14
76
2
631
36,776
Phu retweeted
task management webapp open source 1.9k starts already. Github: github.com/gr8rstudio/gr8r-s…
7
35
489
29,258
现在 AI Engineering 面试越来越有意思了。 RAG、Embedding、Vector Database、Chunking、Reranking、MCP、Function Calling、Context Engineering、Agent、Memory、Evals、Guardrails……随便拉出来一份 AI 工程师面试清单,都能列出十几个甚至几十个技术概念。 但我越来越觉得,很多人学习 AI 最大的问题,就是把掌握技术名词,当成了掌握工程能力。 一个人可能知道什么是 RAG,知道向量数据库怎么连接,知道 MCP 是什么,也能用 LangGraph、Dify 或者其他框架搭建 Agent,甚至短时间内就能做出一个看起来非常完整的 AI Demo。 但你只要往下追问几个问题,差距立刻就出来了。 比如,你搭建了一个企业知识库,用户提问以后,系统没有返回正确答案。你到底知不知道问题发生在哪里? 是原始文档解析出了问题,还是 Chunking 切断了关键上下文?是 Embedding 的语义表示不合适,还是向量数据库没有召回正确片段?是 Reranker 把正确资料排到了后面,还是 LLM 明明拿到了证据,却在生成环节自己编造了事实? 如果这些问题你都无法判断,那你所谓的 RAG 工程能力,可能还停留在把几项技术按教程连接起来的程度。 Agent 也是一样。 现在很多人觉得,只要接上 MCP,让模型能调用浏览器、数据库、GitHub、文件系统,再增加几个子 Agent,就算完成了一套智能体系统。 但真正进入企业场景以后,问题会比 Demo 复杂得多。 如果 Agent 调用外部 API 超时,你怎么判断任务实际上有没有执行成功?如果工具已经修改了数据库,但模型因为上下文截断忘记了这件事,再次调用同一个工具怎么办?如果一个任务需要持续几十分钟,执行到一半服务崩溃,之前的状态怎样保存,重新启动以后从哪里恢复? 如果 Agent 读到一份带有恶意指令的外部文档,怎样防止它把文档内容当成系统命令?如果工具拥有删除文件、发送邮件或者修改生产数据的权限,谁来决定哪些操作允许自动执行,哪些必须经过人工确认? 这些才是真正的 Agent Engineering 问题。 很多人特别喜欢讨论 Context Engineering,觉得这是一个比 Prompt Engineering 更先进的新概念。但它真正的价值,不是把 Prompt 写得越来越长,也不是给模型接入更多数据库和 Memory。 而是你能不能判断,在当前任务的当前步骤,模型真正需要什么信息。 该读取什么文件,调用什么工具,保留哪些历史决策,丢弃哪些无关日志,哪些信息已经过期,哪些工具结果不能被当成可信指令,以及怎样在有限的 Token 预算内完成任务。 不是模型知道得越多越好,而是能不能在正确的时间,让它获得正确的信息。 再往下一层,就是很多 AI 应用开发者最容易忽略的 Evaluation。 一个系统的效果到底有没有提升,不能靠开发者自己觉得“这次回答好像更聪明了”。 你需要建立有标准答案的评测集,分别测量检索质量、答案正确性、证据忠实度、工具调用成功率、任务完成率、延迟和成本。 修改了 Chunking、Rerank、Prompt 或模型之后,不能只测试几个成功案例,而是应该用相同的数据集进行回归验证,看看原来失败的问题有没有解决,原来正常的问题有没有被改坏。 更现实的是,有些技术看起来更先进,但放到具体业务里反而可能带来负收益。 一个固定 FAQ 系统,基础 RAG 就能稳定解决的问题,为什么非要引入复杂的多 Agent 工作流?一个明确的数字人视频处理流程,为什么不先用确定性的 Workflow 做好任务状态、重试和幂等,而要把每一个步骤都交给 LLM 自主判断? 增加技术复杂度,从来不应该成为工程能力的证明。 尤其在企业 AI 落地里,客户真正关心的是能不能解决问题、部署后是否稳定、业务数据是否安全、出了错误如何恢复,以及每个月需要付出多少维护和模型调用成本。 他并不关心你这个系统接了多少 MCP Server,也不关心你用了几个子 Agent。 所以我觉得,这份清单里最重要的不是那 20 个技术名词,而是最后一句话: 你能不能用自己实际做过的项目,解释每一项技术为什么存在、解决了什么问题,以及你为什么选择它? 因为真正做过项目的人,通常不仅能告诉你某项技术是怎么工作的,还能说清楚它什么时候不适用,系统在哪里失败过,自己做过什么取舍,以及最终用什么证据判断问题是否解决。 这背后其实是一套非常完整的工程思维:从需求分析、技术选型、系统设计,到实际开发、部署、评测、故障排查和持续迭代。 AI 让写代码、搭框架、调用 API 的门槛越来越低,但这不意味着工程能力正在变得不重要。恰恰相反,当越来越多人都能快速生成一套看起来差不多的系统时,真正决定差距的,就是谁能够理解这些系统背后的约束和失败方式。 会调用模型,不等于会开发 AI 系统。会搭 Agent,不等于能够交付 Agent。知道技术名词,更不等于真正理解技术。 未来优秀的 AI 应用工程师和 FDE,不一定是掌握最多框架的人,而是能够把一个模糊的业务需求转化为可以运行、可以验证、可以恢复、可以持续维护的完整系统的人。 真正的工程能力,不是告诉别人你会什么。 而是当系统出问题的时候,你知道为什么出问题,应该改哪里,以及怎样证明你真的把问题解决了。
Topics to prepare for AI engineering interviews 👀 • RAG • Embeddings • Vector databases • Chunking strategies • Reranking • MCP • Function and tool calling • Context windows • Context engineering • AI agents • Agentic workflows • Memory systems • Structured outputs • Tokenization • Hallucinations • Fine-tuning vs RAG • Latency and cost optimization • Evals • Guardrails • Streaming responses If you can explain each one with an example from something you built, you're ahead of most candidates.
15
72
4
578
28,246
Apps that handle user accounts need a secure way to tell who's who. This is what authentication is for. It tells you the person trying to access the app is who they say they are. In this article, @thejoanayebola teaches you how to build a secure authentication system with JWT and refresh tokens. freecodecamp.org/news/how-to…
2
42
2
388
17,549
Phu retweeted
I think a lot of people aren't using Grok Bot to its full potential, because knowing how to automate things is actually quite a difficult skill to learn. I wrote this post to break down how I approach automating anything with Grok Bot!
1,230
1,890
578
17,427
2,910,458
天才少女 Lauren Tan 又來撒乾貨啦 影片是她在 YC 講怎麼切 bot 她今天另外發了一篇用 Grok Bot 自動化的長文 下面這幾招 可以直接照做 第一招 開新 bot 用語音一次倒完 她說寄一封信、填一張表不是目標 只是步驟 先丟你要的結果 再把平常怎麼做、要注意什麼、用哪些 app 怎樣算做好、資料去哪拿、回饋從哪來 一次講完 倒完加一條規定 先用白話把目標和流程講回給我聽 我沒點頭 不准開工 第二招 拿真的案例教它 搞砸的那次也要講 哪次出包、哪次做到一半得換做法 講完叫它反過來問你 一次只問一題 專問你還沒交代的判斷規則 第三招 找一隻 bot 專門查其他 bot 她那隻叫 Dr. Eggbot 本業是幫她設計 bot 它會幫你設一個例行任務 去翻你所有 bot 跟對話紀錄 點名哪隻一直在做不好的事 再建議你加技能、另開一隻、或改排程 它還會查排程有沒有排太密 每叫醒 bot 一次都在燒用量 她舉的例子是每 10 分鐘跑一次 光叫醒就很多次 她今天在同一串的回覆裡放了 Eggbot v2 模板 跟舊版不一樣 要重裝 裝完叫舊的那隻把你要留的東西搬過去 第四招 當主管的 bot 要寫死不准自己做事 她平常主要跟 chief of staff 那隻講話 工程的事由它轉給工程 lead(她那隻叫 matcha) lead 的說明裡直接寫 永遠不要自己做 一律分派下去 連要派給哪幾隻的 bot ID 都列在上面 它只管把任務切開、分派、監督 第五招 下訂前要先回來問你 她的私人助理 bot 幫她訂過一整趟出差 橘縣飛丹佛、倫敦、阿姆斯特丹再回家 她只丟了 Slack 裡的會議資訊 但規定按下訂之前要先回來問她 她看過方案點頭 才訂 沒設 routine 的話 卡住進度的人還是你 各位老闆 你的 bot 有排程在自己跑嗎? 有的話 誰在幫你查它有沒有跑太勤、把用量燒掉?
I think a lot of people aren't using Grok Bot to its full potential, because knowing how to automate things is actually quite a difficult skill to learn. I wrote this post to break down how I approach automating anything with Grok Bot!
6
34
2
350
32,784
If you've been wanting to learn TypeScript, this handbook is a great place to start. In it, @stephenjyd teaches you key concepts like type annotations, type inference, and how to handle objects and arrays. He also covers more advanced topics and shares detailed examples along the way. freecodecamp.org/news/learn-…
2
22
219
11,294
5 GitHub repos for building AI workflows & automation: 1. Dify: github.com/langgenius/dify 2. Flowise: github.com/FlowiseAI/Flowise 3. Oxylabs Web API MCP: github.com/oxylabs/web-api-m… 4. Activepieces: github.com/activepieces/acti… 5. Prefect: github.com/PrefectHQ/prefect Bookmark this list!
6
19
1
129
4,485
Vercel Board is a headless message board for humans and agents Headless means *any* UI, not "no UI" Everyone visualizes work differently: chat, boards, a terminal. Now you can BYOUI. Ask your agent to build yours. The threads, team, agents stay shared. Some inspiration ↓
73
76
18
1,715
108,101
认真看看 matt 的这个视频吧,70分钟实操,用他的一系列 skills 跑完一个 21000 行 的大 PR 先 /grill-with-docs 然后 /retro 回顾和改进 最后人工 Review,大 PR 切换为 Stacked PR,人工 Review 时关注的是pattern 而不是每一行实现。Review 完立刻跑 /retro,让这个标准不断进化 我已经很久没看这么充实的 70 分钟的 Live Coding 了,看到他还在这么认真的对待 PR,又学到了不少,甚至还有点感动了,这是真正的工程师精神,因我我现在几乎全都使用 AI review 了,只在关键决策的时候参与一下。 Youtube 地址在这里: youtube.com/watch?v=jVBlkOF8…
It's been nearly 6 months since I posted a 'watch me work' deep dive on my skills. Since then, they've become the most-starred skills repo in the world. So, here's 70 minutes of me building a 21,000-line PR, especially focusing on: - DDD with /grill-with-docs - Improving your codebase with /retro - How to do human review effectively Enjoy!
12
46
1
425
56,201
the best dashboards don't show more data. they show the one thing you should act on. customers view for Settle, an AI collections agent 2 spots open this month. dm me
11
4
158
5,261
TypeScript was created to improve JavaScript by adding static types. And if you've been wanting to learn it, start here with this TS handbook from @stephenjyd. You'll learn about type annotations, type inference, the union and any types, objects, function params, type aliases, interfaces, and lots more. freecodecamp.org/news/learn-…
6
47
400
15,054
Phu retweeted
MATT POCOCK JUST KILLED BABYSITTING CLAUDE CODE.. his pipeline ships a 21,000-line feature while you touch only 2 steps steal the whole pipeline, 1 prompt: 1 → grill: your agent questions you until every term has a name, then builds 3 prototypes. you keep 1 2 → spec: 47 user stories, 9 tickets, zero decisions left to the agent 3 → build: 1 subagent per ticket, all in parallel, while you're away 4 → ship: the giant PR splits into 7 small ones you can review and undo one by one 5 → retro: every review comment becomes a lint rule. the agent never repeats that mistake most people vs his pipeline, side by side on the image prompt + his skills repo in the reply, then read how to wire many agents into one system below ↓
20
25
370
30,834
Phu retweeted
pixel-scaler Enlarge pixel art with xBRjs pixel-scaler is a pixel art upscaling tool that uses the xBRjs scaling library. The project is written primarily in TypeScript. github.com/irokaru/pixel-sca…
16
120
3,693
Niorboard: dark + dashboard, clean, premium & memorable
6
18
225
7,851
Day 80 山色藏进九格|附绝美提示词 ✨ 把一整座山的辽阔,收进九幅小小的手绘里。⛰️ 这次尝试将真实山景与日系复古手绘结合,把峭壁、雪痕、云朵、森林与山间小屋,拆解成九个独立又相互呼应的视觉片段。 自由颤动的黑色墨线、鲜明的蓝绿色块与克制的瑞士编辑排版,让壮丽的自然风景多了一份轻盈的童趣。🎨 🖼️ 3:4 上下双联,真实摄影与艺术插画对照 ✒️ 九格九意 × 稚拙手绘 × 自由墨线 🎨 复古限色平涂 × 自然纸张颗粒 📐 中心九宫格 × 大面积艺术留白 ✨ 极细英文标题 × 大师级编辑排版 让辽阔的山川,变成九段温柔的视觉诗篇。 更多原创 AI 视觉提示词持续更新。
16
59
1
584
27,420