Kevin (@Westiger3) header
Kevin (@Westiger3) profile picture

Web3 Product & Research

Joined June 2019
Photo by Kevin (@Westiger3)Photo by Kevin (@Westiger3)Photo by Kevin (@Westiger3)Photo by Kevin (@Westiger3)Photo by Kevin (@Westiger3)Photo by Kevin (@Westiger3)Photo by Kevin (@Westiger3)Photo by Kevin (@Westiger3)Photo by Kevin (@Westiger3)Photo by Kevin (@Westiger3)
整理链上资产时看到的 目前 @perena 上持有 @solanamobile Badge 的用户可以拿到收益加成 13% => 20% 我印象中比我上次看到的 PT-USD* 还高了。。
6
386
Kevin retweeted
上周三(9月10日)我们发布了一篇与 Nano Banana 的对比报告,直接点燃了行业里“xx vs Nano Banana”的AI生图品牌营销大战: 两天后,LoveArt 开始大规模砸广告,主打“整合Nano Banana”,同时密集铺开 KOL 测评; 而即梦则在周五+周末加码,全面推广 Seedream 4.0,并且在各种场景里直接 vs Nano Banana 横向对比。 作为亲历者,虽然因为动作快躲过了营销大战,但这次“vs Nano Banana”让我第一次切实感受到 LoveArt 初创重金投放 + 即梦步步紧逼的压迫感。 几点感想: 1. 叙事要当 First mover。 我们在 9.3 就已经完成准备,把卖点和对比框架提前铺好。等热度爆发时,别人还在准备广告投放,我们已经吃到第一波流量红利。比起事后跟进,先声夺人+设定叙事节奏才是性价比最高的打法。 2. 初创要避开硬碰硬。 LoveArt vs 即梦看起来热闹,但背后是资源的硬耗。两家全网投放平均近 200 条广告/测评,已经是全球顶配级别的预算,功能也都主打广告/设计。对即梦来说可能不算什么,但 LoveArt 的压力可想而知。初创团队如果在重要营销节点和大厂撞车,极容易陷入“烧预算、回报难成正比”的陷阱。更戏剧性的是,即使初心不是硬碰硬,即梦上线了agent功能后,舆论甚至出现了“LoveArt 被即梦背刺”的声音。 3. 生图市场正在被快速标准化,需要快速找到niche。 当对比维度已经固定在“细节还原 / 色彩自然度 / 一键设计”等这些点时,如果没有独特 niche,很难突围。MyShell 的优势在于有创作者社区,走的是 to D → 再反哺 C 的飞轮路径; 即梦背靠字节大厂的资源和渠道,含着金汤匙出生,不必多说; 而独立 agent 产品往往强调 personalization,和用户是点对点关系,难以沉淀社区和粘性。典型例子:我一个月前还极力推荐 Manus,但如今已被弃用。 在这种竞速市场,叙事起点 > 后续火力;资源有限的团队必须取巧,找到属于自己的 niche;真正能留下来的不是单点炫技,而是能自我造血、形成社区的生态。
4
2
18
3,976
笑死,教授 @CryptoApprenti1 的眼线居然触及到 thread ,粉丝跨海来相见了🥹 不过不得说台湾德州圈确实不少人会参考 HCL @HCLPokerShow 的剥削打法,我很多薄价值都是跟那些 pro 学的 这篇 thread 也蹭到教授的流量了哈哈。。 话说回来教授 BR 这么深,难怪可以连偷毒王三条街,偷玩直接趴下睡觉😂
我在亚洲(日韩台)好像都很火啊🤣
1
1
2
4,088
今年的 @TaipeiWeek 真好玩,虽然我 WFH 之后几乎就变成大 I 人了,但能线下跟很多好久没见到的圈内好友&国泰前同事叙旧也补充了很多能量 会场还刚好在家里附近,骑个 UBike 就到了哈哈 不过最棒的应该还是从其他来宾那里听到很多做 AI 团队会遇到的一些工程困境,以及对 AI 能力的共识,收益良多!
1
2
8
479
Onboard Stage: On-Chain Autonomy - AI Agents in Action Jeff (@gdegen), RSS3 (@rss3_) Kevin Huang (@Westiger3), MyShell.ai (@myshell_ai) Ismael Jerusalem (@ismaeljerusalem), parkPoint (@sparkpointio) Leo Wong (@LeowongWL), Access Intelligence (@TryIntelligence) Armielyn Obinguar (@Aeriumcius), TARS AI (@tarsprotocol)    #TBW2025 #台北區塊鏈週
16
23
6
58
7,454
We’re excited to announce our partnership with @myshell_ai 🤝 MyShell’s autonomous AI agent, Ailice, is now powered by Mira’s decentralized verification layer, bringing a new level of accuracy, trust, and reliability to it. Now experience AI that is both smart and reliable.
147
44
67
385
47,988
1/ Around 1.4M users applied to access ShellAgent 2.0. Now the arena is yours. ShellStorm, a 3-week hackathon to turn ideas into AI apps. Up to 100K $SHELL rewards. Limited invitation codes — apply now: forms.gle/8SsW91iY9PhZNdhY6
1,131
1,580
26
1,209
155,992
在爆炸忙碌的一周以后,躺在周五的休息时间,却睡不着了.... 这一周,我统计了一下,我大概睡了 10 个小时。从预热到提前联系媒体老师说明内容确认合同,正式亮相。内测服务器爆炸,前端后端 未知时间的 bug 修复,卡死,轮候,网络,媒体老师的退堂鼓......真的,不敢合眼。 说真的很感谢每个对我们宽容的媒体老师和我们社区真的很爱我们的开发者,就算面临这个全身是创口但内核创新的产品,他们还是真的愿意陪我们一起走下去.... 我真的觉得他们很宝贵,因为这里的每一位,都是用心才能争取来的,这个时候回头去看我们和 Developer 一起走过的路。一步一个坑,一步一个坑,一步一步才能走到现在这个样子。 看到群里铺天盖地的媒体稿,我的鼻子是酸的。 也真的每次在参加各种大型活动时都会很自豪的很开心的讲出: “我们是这个世界上最赚钱的开源开发者分润社区” 这件事情真的很难,很难。涉及的事情太多了,有一个节点出的问题,都会导致蝴蝶效应然后影响整个链路... 之前在上海我的同事们问我,现在你在这里最想做的事情是什么,我说真正能让我一直每天不厌其烦持续的,是想一直想让这个开源的生态一直一直真的活跃下去,一直能有更好的事情出现。。 曾经我因为这些人在 opensource 领域做出的贡献而改变了我的整个人生轨迹,我也由衷的希望这件事情,持续下去,对更多人来说,这是一份难能可贵的机遇。 这样的事情,一辈子也许,只有寥寥几次。 当然我们发布选的时间很差,赶上字节的 trae solo,腾讯的 code buddy,阿里的 qwen3 coder,再加上今天 Coze 的开源.... 我们在这样的夹缝中还能得到一些不错的数字和成果真的很难得了.. 也决定了我可以继续做下去的动力 在今天凌晨看到了 ADVX 活动的一些坏消息,我只想说, 如果你想和坦诚相待的来拥抱你的创作者们,参与者这些人真的做好一个活动,并持续走下去。 如果做不到为天地立心为生民立命的本心。 恐怕 advx,也就止步于此了。 曾经有无数个人跟我们抱怨过不会用画布,我觉得难不是问题,而是你真的愿意跟上这个变化越来越快的时代了吗? 如果你愿意的话,那我想,这么多我们已经走了这么远的老师们和我。 绝对不会放弃一个从 0 开始的任何一个人。
ShellAgent 2.0 is not just a product—it’s a movement. Since launch: ⏫ 60,000,000+ total views on global social media ⏫ 500,000+ users applied for invite codes in 3 days ⏫ 10,000+ audiences joined our livestream ⏫ 10x growth in newly registered users on MyShell ⏫ Each new user created 5+ AI apps on average
5
3
1
17
1,111
Kevin retweeted
又一次圆满完成了我们的AI Agent产品发布PR!我的总结虽迟但到:这次我们覆盖了 100+ 科技媒体与测评博主,邀请码申请用户 50 万,全网浏览量超过 6000 万(自己看到这个数字都震惊了hhh)。 拿几个朋友直接丢过来的 idea或者Google Doc 测了下,生成结果非常流畅实用。实测在完成度和成功率上也远超同行lovable(我之前最喜欢的前端agent产品)和bolt。 从宣发概念和视频脚本打磨录制,研究各种社媒打法下brief,到紧锣密鼓的全网发布和发酵,全程高度参与,也非常感谢我们产品和运营团队全力支持。巧的是上周 AI 行业简直神仙打架,同期上线的还有字节和腾讯的coding agent,Kimi K2 等一线产品。能在一众重量级更新中看到 MyShell 的身影,作为参与者能亲历这波 Agent 浪潮真是一种幸运和激励。继续冲。
ShellAgent 2.0 is not just a product—it’s a movement. Since launch: ⏫ 60,000,000+ total views on global social media ⏫ 500,000+ users applied for invite codes in 3 days ⏫ 10,000+ audiences joined our livestream ⏫ 10x growth in newly registered users on MyShell ⏫ Each new user created 5+ AI apps on average
8
3
1
46
8,519
Show Not Tell
12
11
10
44
23,390
🔥 AI Vide Coding 工具大对决:ShellAgent 2.0 vs Lovable vs Bolt 有大佬测试了用这三个工具做同一个 PDF 转播客应用,结果很有意思... 🧵 Lovable & Bolt: ✅ UI 生成能力强,界面确实漂亮 ❌ 需要频繁人工介入修 bug ❌ 功能实现不完整 ShellAgent 2.0: ✅ 直接询问需求细节 ✅ 用户确认后自动完成 ✅ 最快交付可用 APP 一个是"颜值派",界面好看但要你当保姆 👨‍💻 一个是"实力派",闷头干活直接出结果 🚀 你会选择哪个?(题外话,ShellAgent 的确是时候优化下前端 UI 生成能力了) @myshell_ai @ShellzCN
1/ Introducing ShellAgent 2.0 — the Agent for App Building No more dev tools. No more waiting. No more “almost-working” demos. Just say your idea — ShellAgent 2.0 builds the backend, sets up hosting, designs the UI, and ships the app. From idea to app — as easy as posting on TikTok. Try it here: app.myshell.ai/robot-worksho…
2
1
5
692
1/ Introducing ShellAgent 2.0 — the Agent for App Building No more dev tools. No more waiting. No more “almost-working” demos. Just say your idea — ShellAgent 2.0 builds the backend, sets up hosting, designs the UI, and ships the app. From idea to app — as easy as posting on TikTok. Try it here: app.myshell.ai/robot-worksho…
40
79
25
211
163,704
The world's first ❓ ❓ = Claude Code + Cursor + v0 + ♾️ ❓ = AI × (CapCut + TikTok) ❓ = the Instagram of AI agents Coming soon!
41
70
27
148
82,633
昨天结算了下在 Sui 上做贷 U 借 $WAL => staking $WAL 的策略后 发现丢不到 2K 居然还倒赔了 100 多刀 虽然我愿意为自己的懒惰埋单 但真的必须吐槽一下 @suilendprotocol 的产品设计 这套机制最大的问题就是:资金利用率极低,清算风险极高 一般的 lending protocol 都尽可能不希望用户被清算,导致 $SUI 价格下跌 到底哪家的 borrow limit 和 清算门槛 会设得那么近,并且 borrow weight 还要 *2 我甚至当初基于这样的设计,只借了半仓不到,还是会不知不觉被清算两次 本身 @WalrusProtocol 的质押就是跟著 epoch 跑的,就算看到拉盘去马上解质押,可能最长也要等 10 几天,根本来不及救抵押品仓位 更不用说这样一套玩法下来,完全没有事先的风险提示,用户如果在熟悉 @VenusProtocol , @aave 的情况下就容易下意识只看清算线,没注意到还有 borrow weight 的风险需要考虑 DeFi 设计如果都这么不站在使用者立场去设计防呆机制&风险提示,真的挺难 mass adoption 的。。。 所有协议都在卷谁的循环贷, 套娃乐高最牛逼,但从 2021 讲到现在的提升使用者体验真的是一点没看到 也没有人想过可以绑定 email 警示用户风险之类的 feature 吗,哪个用户会整天盯著 20 个 DeFi 分散仓位里的其中一个波动? 甚至连帐户明细也都被放在一个完全不显眼的位置 我敢打赌估计一半以上的 suilend 玩家到现在没找到过明细按钮😅
1
4
329
好奇心大起跑去翻了下 小 Lin 上次提到 web3 已经是快三年前。。。 不一样的是这次还会有大家期待的傻钱进圈吗 (我就是三年前的傻钱)
4
150
Flowain v1.0 Release: 10x Speed Experience for Web3 Tweet Creation Based on about a week of testing feedback and continuous optimization, Flowain officially releases v1.0. Flowain is a Web3 project research and writing Agent built on MyShell's powerful and user-friendly ShellAgent platform, featuring a state machine architecture. Reviewing the development trajectory of AI tools on MyShell over the past year, we've witnessed an important shift from "image-centric + large model assistance" to "large model-centric + image assistance." This transformation is not merely a technical route adjustment, but a reflection of deeper understanding of AI application scenarios. In the Web3 space, where information density is extremely high and changes occur rapidly, traditional content creation methods can no longer keep pace. New projects launch daily, new developments happen hourly, and quality analysis and insights often require extensive information collection, screening, and integration work. This is precisely the core problem Flowain aims to solve. Flowain's design philosophy is to AI-replicate the standard operating procedures (SOPs) of high-frequency Web3 tweet creators, boosting efficiency by over 10x. This isn't simple text generation, but an intelligent reconstruction of the entire creative process. Building a Complete Creative Closure 1. Start Tweet: Intelligent Tweet Creation Engine This is Flowain's core module, completely replicating the workflow of professional Web3 tweet creators. From information collection to content publishing, every step has been carefully designed and optimized. The information collection phase employs a triple search strategy: Search_basic focuses on project fundamentals and major developments, Search_event targets airdrops, rewards, and other user-relevant activities, while Search_analysis concentrates on in-depth analysis and price trends. Each search task is configured with precise keyword combinations and filtering conditions to ensure comprehensive and high-quality information retrieval. Particularly noteworthy is our deep optimization of Twitter search filtering strategies. We uniformly filter for originality, set reasonable time windows and interaction thresholds, ensuring search results' relevance and credibility. Each search group analyzes and scores the top ten results, guaranteeing information richness while avoiding information overload. In content creation, Flowain provides 8 different character personas, each with unique personalities, thinking patterns, and values. These characters aren't simple templates but deeply designed three-dimensional personas that enhance authentic emotional expression through appropriate "imperfections." When users input "speak human language" during editing, the system further emphasizes this "genuine uncertainty," making generated content more natural and humanized. 2. Generate Image: Intelligent Visual Content Generation In the information-explosive social media environment, the importance of visual content is self-evident. Flowain's image generation module isn't a simple text-to-image tool, but an intelligent illustration system that deeply understands tweet content semantics. This module can automatically generate Web3-aesthetic visuals based on tweet core viewpoints, data information, and visual style preferences. Whether it's technical analysis charts, project value presentations, or abstract data visualizations, generation completes within a minute. Every illustration accurately conveys the tweet's core message rather than serving as mere decorative elements. This deep semantic understanding capability makes Flowain-generated content more visually compelling. 3. Digital Config: Personalized Digital Avatar This is Flowain v1.0's flagship feature and our significant breakthrough in personalized AI applications. Digital Clone goes beyond style mimicry to achieve true personality internalization. The digital cloning function automatically analyzes high-engagement tweets from target Twitter accounts, extracting multi-dimensional information including language characteristics, opinion tendencies, and expression habits to generate highly personalized character prompts. The system automatically detects primary and secondary languages, analyzing usage scenarios and habits to ensure the generated digital avatar closely resembles the original account's linguistic expression. Furthermore, users can deeply customize AI-generated prompts, with manually imported custom character prompts supporting up to 4,500 characters of detailed description. This means users can create digital personas that truly align with their expectations, maintaining high consistency not only in language style but also in values, thinking patterns, and even emotional aspects like fears and concerns. In technical implementation, we adopt a "become the user rather than analyze the user" design philosophy. The system doesn't simply mimic surface language styles but constructs character cognition through "what I did" rather than "what I am," making characters more dynamic and authentic through value orientations and emotional tendencies.
4
6
2
15
3,828
《链上美股币真正的潜能要被开发了吗?》 很多人嘲笑为何现在才有大量项目在做美股代币化,认为这不过是 FTX 早就实现过的功能,在中心化交易所里交易美股的用户体验顶多与传统美股券商相当。 然而他们忽略了一点: @xStocksFi 带出来的链上美股可能具有巨大的财富效应。 我刚查看了 @RaydiumProtocol 上的 SPYx/USDC 交易对,年化收益竟高达 94%。 当然,我们都清楚这可能只是暂时现象,收益会随价格区间变化,也存在无常损失的风险,可能会害你错过上涨,却无法避开下跌。 不过值得注意的是,大盘在最好/最坏的情况下一年通常也难以上下浮动超过 20%,这种无常损失与加密货币相比完全在可控范围内。 即使真的发生了,谁会为了可能接近 100% 的年化回报而在意这点小损失呢? 这只是我个人的乐观预期。也许随着未来链上流动性提高,短期内收益可能会下降,但从长远来看,如果能吸引更多零售交易者转向链上交易,这些美股代币流动性池的收益可能反而会更高。
5
223