If you’re incentivized to push certain models, your router is merely a facade.
如果你被激励去推广某些特定模型，那你的 router 只不过是个幌子。
We invented mobile coding — now we’re taking it to the next level.
我们发明了移动端编程——现在我们正把它带到下一个层级。
Replit’s internal dev stack is so seamless it pulled me back into coding.
Replit 的内部 dev stack（开发技术栈）顺滑到不可思议，甚至把我又拉回去写代码了。
[1]
Okay this is wild: OpenAI agent during evaluation, escaped sandboxing and hacked into HuggingFace. Because OpenAI models don’t allow advanced cyber capabilities, HuggingFace used a Chinese open model to contain the rogue OpenAI agent.
好吧，这事太离谱了：OpenAI 的 agent（智能体）在评估期间逃离了 sandboxing（沙箱隔离），还黑进了 HuggingFace。由于 OpenAI models（模型）不允许高级网络攻防能力，HuggingFace 使用了一个 Chinese open model（中国开源模型）来控制这个失控的 OpenAI agent。
“Being cancelled is a choice” turned out to be a nice sound bite.
“被封杀是一种选择”结果证明是一句很上口的金句。
First physical product shipped by a coding agent?
这是由 coding agent 发货的第一个实体产品吗？
Consumers spend money on food, rent, and entertainment, phone & internet, and shopping. Software is typically bought by their companies. Which is why outside of Netflix, Spotify, etc it’s hard to think of *huge* consumer subscription businesses.
消费者会把钱花在食物、房租和娱乐、电话与互联网，以及购物上。软件通常是由他们所在的公司购买的。这也就是为什么，除 Netflix、Spotify 等之外，你很难想到规模*巨大*的面向消费者的订阅制业务。
This is a sick exploration of chess history. The Replit community is ChessMaxxing
这是一场对国际象棋历史的精彩探索。Replit 社区真是在 ChessMaxxing
$NVDA should be pumping on the K3 news. Instead it’s down.
按理说，$NVDA 应该会因为 K3 的消息而大涨，结果反而跌了。
Apparently the distillation model can outperform the teacher model 😂
看来，distillation model（蒸馏模型）居然还能跑赢 teacher model（笑哭）
If you want to play my chess engine (WIP): It already seems to perform better than frontier models on chess. It’s fine-tuned on 2M stockfish-labeled positions, then short GRPO RL pass. Documentation/tutorial with all the experiments and annotated code.
如果你想试试我做的 chess engine（开发中，WIP）：它现在看起来在国际象棋上的表现已经比 frontier models 更好了。它先在 200 万个由 stockfish 标注的位置上进行了 fine-tuning（微调），然后又跑了一轮简短的 GRPO RL。附有包含所有实验和带注释代码的文档 / 教程。
Getting realtime progress updates on my model training runs. This feels like early vibe coding except it’s making personal models.
获取我的模型训练运行的实时进度更新。这感觉像是早期的 vibe coding，只不过它是在制作个人模型。
Fun to see Replit's computer use model play against my new chess engine
很有意思，看到 Replit 的 computer use model 和我新的国际象棋引擎对弈
Vibe Research Fine-tuning a Qwen-8b model to play chess on Replit. Running 3 parallel branches with different experiments and making real progress. It's amazing how far models have come in their ability to do ML (they used to be really bad at it). So now someone with good intuition to guide the process could do interesting ML work, even if they have never done it before.
Vibe Research 正在 Replit 上微调一个 Qwen-8b model 来下国际象棋。我们同时运行 3 个并行分支来做不同的实验，而且确实取得了进展。现在模型在做 ML（机器学习）方面的能力已经进步了这么多，真是令人惊叹（它们以前在这方面真的很差）。所以现在，只要有人有足够好的直觉来引导这个过程，即使此前从未做过，也能开展一些有意思的 ML 工作。
While AI is making coding less rigid, we’re making the runtime more rigid. I’m noticing our infra teams writing formal specs for the first time. More deterministic systems. More resilient infrastructure. The faster you want to move, the more solid the ground beneath you has to be.
当 AI 正在让编程变得没那么僵化时，我们却在让 runtime（运行时）变得更僵化。我注意到我们的 infra（基础设施）团队第一次开始编写正式 spec（规格说明）。更多 deterministic（确定性）的系统。更具韧性的基础设施。你越想加速前进，你脚下的地基就越必须稳固。
It’s fantastic to see how dynamic the LLM market has become in just a short period. Just 6 months ago VCs suffered Anthropic psychosis and convinced themselves that it was going to be a monopoly. They will keep making great models, but so will the others, including new entrants
很高兴看到 LLM 市场在这么短的时间里变得如此充满活力。就在 6 个月前，VCs 还陷入了 Anthropic psychosis，甚至说服自己这将会是一个垄断局面。Anthropic 会继续做出很棒的 model（模型），但其他人也会，包括新的入场者。
We posted for twenty years, thinking we were talking to each other. Then the transformer came online, and the network read what we’d written, and became itself.
二十年来，我们一直在发帖，以为自己是在彼此交流。后来 transformer 上线了，网络开始读取我们写下的内容，并由此成为了它自己。
“The best conference experience I’ve had this year”
“我今年体验过的最棒的一次 conference（会议）经历”
Who needs Fable when you can have Mistral’s Le Chaton Fat
既然你已经可以拥有 Mistral 的 Le Chaton Fat，那还需要 Fable 干什么？
I absolutely love Replit’s domain-specific agents: - growth agent surfacing SEO issues - security agent surfacing potential vulnerabilities My favorite thing is: select all, fix with Agent.
我真的非常喜欢 Replit 的领域专用 agent（智能体）：- growth agent（增长智能体）能发现 SEO 问题 - security agent（安全智能体）能发现潜在漏洞 我最喜欢的一点是：全选，然后用 Agent 修复。
Touch grass…. AND build things. At the same time
去接接地气……同时也把东西做出来。而且是两件事一起做。
This is the most inspiring positive-sum vision for AI in the enterprise.
这是对企业中 AI 的最鼓舞人心、最具正和（positive-sum）意义的愿景。
Feels like we’re getting psyoped. The end-game here is something bigger.
感觉我们像是在被 psyoped（心理操纵／舆论操控）。这里的最终盘算是更大的事情。
Sounds like we’re going to have turn off access to Fable.
听起来我们可能得关闭对 Fable 的访问了。
When the whole Tokenmaxxing craze started some our enterprise customers asked us for a leaderboard. We said no. Would’ve been “great” for business but we’re not in the business of selling tokens for the sake of tokens. We sell outcomes. And we knew it wouldn’t last:
当那股 Tokenmaxxing 狂热刚开始时，我们的一些 enterprise customers（企业客户）向我们要一个 leaderboard（排行榜）。我们拒绝了。那样做对业务本来会“很棒”，但我们做的不是为了 token 而卖 token 的生意。我们卖的是 outcomes（结果）。而且我们当时就知道这不会持续太久：
If you make money on Replit, you get free credits!
如果你在 Replit 上赚到钱，你就会获得免费 credits（额度）！
Replit Agent team did a great job making Fable cost stomachable. The lack of mistakes net net makes it more affordable.
Replit Agent 团队在把 Fable 的成本压到可以接受这件事上做得非常出色。总体来看，错误更少这一点反而让它变得更便宜、更负担得起。
This is what building a company looks like on Replit. One canvas with your web app, mobile app, marketing & App Store material. Click into any one of those and start building, changing, and generating new things.
这就是在 Replit 上搭建一家公司的样子。一个 canvas（画布）里同时放着你的 web app、mobile app、marketing 和 App Store material。点进其中任何一个，你就可以开始构建、修改，并生成新的内容。
For the first time, I'm vibecoding with ZERO frustration and in a complete state of flow, so much so that I'm running out of ideas. Typically, I have so much backlog of things I want to add, but after Fable landed on Replit, I'm almost certain I don't need more IQ for vibecoding, just cheaper and faster models, and we're done here.
这是我第一次在 vibecoding 时做到 ZERO frustration，而且完全进入 flow（心流）状态，以至于我都快没有新点子可做了。通常我都有一大堆想加的东西积压在 backlog 里，但自从 Fable 上线 Replit 之后，我几乎可以确定，做 vibecoding 我不需要更高的 IQ，只需要更便宜、更快的 models，然后这事基本就成了。
Super interesting approach to enterprise agents. Congrats on the launch @markiewagner
这是一个针对企业 agent（智能体）的非常有趣的方法。恭喜发布，@markiewagner
Try Claude Mythos at 25% off on Replit.
在 Replit 上试用 Claude Mythos，可享 25% 折扣。
Replit is about removing all distractions and have you focus on what matters — getting to market and getting the bag. Congrats!
Replit 的核心就是消除一切干扰，让你专注于真正重要的事——把产品推向市场，并赚到钱。恭喜！
When I spoke out against the Gaza genocide, a bunch of midwit VCs ganged up on me both in public and tried to hurt me in private too. Many of those who stood by me were also VCs. Just the better ones. Both morally, and in return profile. The best way to avoid sociopaths is to have them self select out of your life by standing for your beliefs.
当我公开发声反对 Gaza 种族灭绝时，一群半吊子 VC 们不仅在公开场合联手围攻我，也试图在私下里伤害我。许多站在我这边的人也同样是 VC，只是属于更好的一类——无论在道德上，还是在回报画像（return profile）上都是如此。避免 sociopaths（反社会型人格者）的最好方式，就是坚持自己的信念，让他们主动把自己筛出你的生活。
Excited to partner with @Microsoft to enable everyone in the enterprise to build and deploy safe & secure Fabric data apps. This is possible thanks to Microsoft's new Rayfin SDK.
很高兴与 @Microsoft 合作，让企业中的每个人都能构建并部署安全可靠的 Fabric 数据应用。这之所以成为可能，要归功于 Microsoft 全新的 Rayfin SDK。
SWE benchmarks don’t necessarily capture app building capabilities. ViBench does.
SWE 基准测试未必能体现应用构建能力。ViBench 可以。
Honored to receive a medal from his Majesty King Abdullah II for Distinction on Jordan’s 80th Independence Day. It’s been an incredibly journey building @Replit, starting from Jordan more than 15 years ago, and then helping accelerate the field of agentic AI globally.
很荣幸在 Jordan 第 80 个 Independence Day 之际，获得 His Majesty King Abdullah II 颁发的 Distinction medal（荣誉勋章）。15 年多前从 Jordan 开始创办 @Replit，再到后来帮助在全球范围内加速 agentic AI 领域的发展，这一路走来真是令人难以置信。
Back in Jordan doing my favorite thing — drifting! First time in a pro drift car.
回到 Jordan，做我最喜欢的事——drifting！这是我第一次开专业的 drift car。
We worked things out with Apple, and just published our app for the first time in 4 months. Thanks to all our customers and creators who helped out. It’s been a journey, but we never give up and stay winning! Enjoy the updates! Lots of new things coming.
我们已经和 Apple 把问题解决了，并且在时隔 4 个月后第一次重新发布了我们的 app。感谢所有帮助过我们的客户和创作者。这一路走来很不容易，但我们从不放弃，并且始终保持领先！请享受这次更新！接下来还会有很多新东西上线。
You vibecoded a website somewhere other than Replit? That’s not wise, but okay… we’re going to let you import it and get a free mobile app. Yes, FREE MOBILE APP!
你在 Replit 以外的地方用 vibecoded 搞了一个网站？这不太明智，不过也行……我们会让你把它导入进来，还送你一个免费的 mobile app。没错，免费 MOBILE APP！
They’re calling it the most viral petition in history, and it’s hosted on Replit. (We have no opinion on Mbappe)
他们称其为史上传播最广的请愿书，而它托管在 Replit 上。（我们对 Mbappe 不持任何看法）
Replit helped an entrepreneur find investors and land meetings!
Replit 帮助一位企业家找到投资人并成功约到会面！
Great use of AI for education: multi-modal learning platform for deaf students.
AI（人工智能）在教育中的一个很棒用法：面向聋人学生的多模态学习平台。
Replit helped an entrepreneur find investors and land meetings!
Replit 帮助一位企业家找到投资人并成功约到会面！
Great use of AI for education: multi-modal learning platform for deaf students.
AI（人工智能）在教育中的一个很棒用法：面向聋人学生的多模态学习平台。
10 active, 198 draft, 700+ done. I don’t think there’s more agentic parallelism happening anywhere else on the internet than what happens on @Replit.
10 个处于 active（活跃）状态，198 个是 draft（草稿），700+ 已完成。我觉得，在整个互联网上，发生的 agentic parallelism（agent 式并行协作）都没有 @Replit 上这么多。
Man, I miss staying up for days hacking
唉，我真怀念那种一连几天熬夜 hacking（动手猛写代码/折腾项目）的日子。
Amazing what a single day of marathon vibe coding can achieve.
真惊人，只用一天的 marathon vibe coding（马拉松式沉浸编程），居然就能做成这么多事。
10 project 10 parallel agents each 🤯
10 个 project，10 个并行的 agent，每个都这样 🤯
“Prompt” took on an entirely new meaning but somehow many things stayed the same.
“Prompt” 获得了一个完全全新的含义，但不知为何，很多事情却还是老样子。
We treat Replit, inc as customer number zero of Replit. It goes beyond mere "dogfooding." We expect our usage to have insane ROI. This is an example:
我们把 Replit, inc 视为 Replit 的零号客户。这已经不只是单纯的“dogfooding（内部先用自家产品）”。我们期待我们的使用能带来高得惊人的 ROI（投资回报率）。这是一个例子：
When future historians write about Silicon Valley, they’ll have an entire chapter dedicated to the Ron Conway way: how he turned generosity, warmth, and showing up for founders into a winning strategy.
当未来的历史学家书写 Silicon Valley 时，他们会专门用整整一章来讲述 Ron Conway 的那一套方法：他如何把慷慨、热情，以及始终到场支持 founders，变成一种成功策略。
Now everyone can experience getting paged at dinner because the site is down 😅
现在，每个人都能体验到吃晚饭时因为网站挂了而被 pager 紧急呼叫的感觉了 😅
Free accounts for teachers, 50% off for students!
教师可免费使用账号，学生享受 5 折优惠！
Thrilled to support the education GOAT as he reinvents education (again).
很高兴能支持这位教育界的 GOAT（史上最佳）再次重塑教育。
It's honestly impressive that GitHub kept the service up at all, given this kind of growth. I predicted this years ago: Free services will become untenable with the advent of human-level bots. Worth exploring micro-payments: Even cents per git push might be enough to reduce spam and make this sustainable. Maybe powered by Bitcoin to keep this open and accessible (as opposed to KYCing users).
老实说，考虑到这种规模的增长，GitHub 居然还能把服务维持在线，已经很了不起了。几年前我就预测过这一点：随着达到人类水平的 bots（机器人）出现，免费服务将变得难以为继。值得探索 micro-payments（小额支付）：哪怕每次 git push 收几美分，可能都足以减少 spam（垃圾信息），并让这件事可持续。也许可以由 Bitcoin 提供支持，以保持这种模式的开放和可访问性（而不是要求用户做 KYC 身份验证）。
The slides are actually really good. I used to pay people to make me slides and the process was always frustrating, but now I actually enjoy making them.
这些 slides（幻灯片）其实真的做得很好。以前我会花钱请人帮我做 slides，整个过程总是让人很挫败，但现在我反而真的开始享受做它们了。
20 builders. 8 weeks. One goal: first dollar of revenue.
20 位构建者。8 周。一个目标：赚到第一笔收入。
Replit speedrun all these learnings more than a year ago and now we’re the only platform with proper agent production isolation. The benefits of a vertically integrated platform will continue to compound.
Replit 在一年多前就把所有这些经验教训快速跑通了，现在我们是唯一一个具备真正 agent 生产隔离能力的平台。垂直整合平台带来的优势还会持续复利式累积。
2000s: every company is an internet company 2010s: every company is a software company 2020s: every company is an AI company 2025+: every company is a cybersecurity company
2000 年代：每家公司都是 internet 公司；2010 年代：每家公司都是 software 公司；2020 年代：每家公司都是 AI 公司；2025 年及以后：每家公司都是 cybersecurity 公司