Mengxue

@zhouluobo 谢谢萝卜老师 我要去探索一下。一直都是拿claude或者GPT当助理用 这样听上去好像Hermes更合适？

@MengxueBi Hermes是agent，可以接入很多大模型，然后围绕agent还能安装很多工具，方便我们完成日常工作。Claude和ChatGPT都是大模型或者网页版对话工具，虽然它们本身能力强，但是没有办法直接帮助我们整理资料、搜索内容等等。

X 蓝 V 用户可以试一下 Hermes + Grok X Search。 不用先折腾 X API，也能让 Hermes 直接搜索 X 上的实时讨论和各种帖子信息。 我现在的用法非常简单，每天自动找商机。 比如给 Hermes 一个 niche：AI 视频、独立开发、内容自动化、跨境卖家、设计师工作流。 然后设一个每天 9 点的自动化任务 今天这个领域谁在抱怨？ 哪些问题被反复提到？ 哪些需求背后可能有付费意愿？ 能不能做成一个小工具、服务包或内容产品？ 能不能直接生成一个原型？ 这种非常适合做需求雷达。 别一上来就让 Agent 替你创业，多少有点强人所难，哈哈哈。 先让它稳定帮你发现问题、归类问题、持续跟踪问题，循序渐进比较好。 同时我现在的视频生成模型和TTS都是走的 Grok，反正有会员额度，不用白不用啊。

@zhouluobo 写的好诚恳！冷启动能涨500粉已经很难了！大号很多都是自带流量的。搞自媒体真是个费脑子的事情🤣 我现在完全还是在把X当场thoughts dump 的地方

@steipete I'm not saying I'll use lavish for every single task, or loop is the answer to every vibe coding project. What's great about these 2 sharing is they broke my habit of thinking/using AI, and prompted me to explore from a new angle.

2. Reading @steipete 's post on "you shouldn't been prompting you should be designing loops that prompt your agents". The whole hype of /loop really threw me off at the beginning, as a designer, not a coder. How could looping help me in my design process? But Peter's post prompted me to explore harder.

The best 2 inspirations I drew from X this week: 1. @kunchenguid shared his workflow and skills created in this conversation (youtube.com/watch?v=88B6Di…) with @petergyang . I started using Lavish-axis and no-mistakes the next day, while building my own design-iteration-skill

虽然现在skill是热点 但我也不觉得啥啥都要一个skill。有时候一个prompt完全能搞定 model挺聪明的🤣

Once you started using Fable 5 only as the orchestrator and code generator, but leaving discovery and validation to other models, the result quality and ROI on your tokens are significantly improving.

Current idea is to introduce a scoring system at the Discover phase to get a baseline and recalculate the scores at the Validate phase. The loop can pass when the score meets a min or have a mim increase, with no other blockers.

Building a Design-iteration-loop to have agents discover-fix-validate design solutions on their own. The whole workflow is now running okay, but the biggest problem I'm facing now is how to define a quantifiable pass criteria. UI and usability are not easily quantifiable.

在X上经常会有一种错觉，那就是AI已经成为了世界主流。可能从聊天这一件事上来说的确如此，不过让builders无比热衷的agent到今天为止还只是一个少数人的狂欢。如何让更多的普通人、不懂技术的人用他们更熟悉的方式用上AI，难道不是一个很大的商业机会吗？

@SavannahFeder It is super cool! Curious tho, why did you want to create it instead of using Claude Design/Cowork/Codex or any other existing agentic tools? Also, is the image/video generator done by another model like Gemini or something?

holy shit. claude fable just let me build something in 30 hours that should've taken months. it's an agentic canvas, and it's fully replaced my usage of claude cowork and claude design. agent harnesses are worthless now. taste is the last human moat!

@mathildeanita @cursor_ai very zen. love it. Who provided all those beautiful carpets? The company or the employees?

Coolest office @cursor_ai

@TheChowdhary One challenge is to get all those context/data, and another challenge is why would your customer share all these things with you, thinking you might use this information to raise price on them?

I have yet to see somebody build this: AI agents for customer success. An agent that reads everything about your customer. Their workflows, their news, their hiring, their funding. It watches the shared slack channel, the email threads, the tickets. And you connect it to your Claude and just ask it questions. It analyzes, predicts and tells you who needs "more" attention And you can use it for GTM: expansion signals, and tracking champions leaving to join new companies - which is a warm intro to a brand new logo. Many have tried building something like this. I haven't found the right fit yet. You can build it on the Crustdata API + Watcher API today, so the data is not the problem. But it needs someone willing to think deeply about the overall problem and build the all-in-one system. Who is building something like this? Would love to check it out!

@chrstnerode @diabrowser This is beautiful. Nowadays you really don't often see AI generated things that you can call beautiful. But this is very well crafted. And hey I'm a Dia user!

we just wrapped our Spring quarter at @diabrowser. before i jump to the next big thing (!), a short behind-the-scenes on one of my favorite projects to date: artifacts! the design challenge: how do you make AI-generated docs that don’t *feel* AI-generated? (🧵 1/10)

@zhouluobo @op7418 赞。这两天正在磨一个skill，感觉为了将来可以省掉一些manual work现在正在投入更多的精力 不听的test、给反馈、优化、再test。隐隐怀疑我这是不是流程不对，不过看了你的说法感觉安心了

我个人理解的skill，首先应该是个人经验的具象体现，先满足个人的日常任务诉求，然后才是不断的优化，甚至与满足更多用户的使用要求。 比如我最近写的这个skill，最开始就是因为我想要制作很多FPV类型的视频，但是每次写提示词又有点费劲，所以就固化成了skill，再接着看到平台上很多朋友需要这类的提示词，所以就不断优化，慢慢达到可以给别人用的程度。 skill应该是不断优化而来的，不是一蹴而就的。 x.com/zhouluobo/stat…

@op7418 很多看法我都想站起来鼓掌👏 echo一下 用户不在意这个东西叫什么，更在意的是产出以及如何使用，所以我也认为“做skill”比“做产品”要有前景很多。我们谁都不非得要一个新app，新窗口，新媒介，所有的事情就是通过agent gui一个交互就可以达成。

@BrainsAndTennis going to send this article to my agent 🤣

@esandurrani You lost me at the word “course”

Introducing Honen The world's fastest way to train skills needed to succeed in the Al era Simply drop in your org's docs Get interactive courses that self-improve over time To prove out our vision, we partnered with NVIDIA to bring AI literacy to 250,000 learners By 2030, 78M jobs will need reskilling due to Al, and we're ensuring no person or company gets left behind Learn more honen.com