段兵楠

@bingnan

👨‍💻 美团工程师 | 📚 终身学习 | 🌟 认识新朋友

Beijing

Joined April 2007

1.5K Following

698 Followers

3.8K Posts

bingnan retweeted

Spencer Baggins

@bigaiguy

5 days ago

A French engineer who lives quietly in Paris has spent 30 years writing software that the entire internet now runs on without knowing his name. He wrote the code that streams every YouTube video, every Netflix show, every TikTok clip. He wrote the code that runs the virtual servers underneath AWS, Google Cloud, and Microsoft Azure. He calculated more digits of pi than anyone in history. He has no Twitter. He has no marketing. He just keeps shipping. His name is Fabrice Bellard. Here is the story, because almost nobody outside the systems programming world knows what one man has built. Fabrice was born in 1972 in Grenoble, France. He studied at École Polytechnique, the top French engineering school. He never went to Silicon Valley. He never built a startup empire. He just wrote code. In 2000 he started a project called FFmpeg, an open-source multimedia framework for encoding, decoding, and streaming video. He was 28. The project did one thing nobody else had done well. It handled every video and audio format that existed, in one library, on every operating system. He led it himself for years. Today FFmpeg is the invisible engine of the internet. YouTube uses it. Netflix uses it. VLC uses it. Chrome and Firefox use parts of it. Every Android phone, every iPhone, every smart TV, every video editing tool you have ever touched runs FFmpeg somewhere underneath. If you have watched a video on a screen in the last 20 years, Fabrice's code processed it. He was not done. In 2003 he started QEMU, a machine emulator and virtualizer. He wrote it solo until version 0.7.1 in 2005. QEMU lets you run any operating system on any other operating system. It became the foundation of modern virtualization. KVM, the Linux kernel hypervisor, runs on top of QEMU. Every major cloud provider, AWS, Google Cloud, Microsoft Azure, IBM Cloud, runs virtual machines on infrastructure built around it. The Quick Emulator is the most cited piece of cloud infrastructure code on Earth. He kept going. In 2001 he won the International Obfuscated C Code Contest with a small C compiler that grew into TCC, the Tiny C Compiler. TCC can compile and boot a Linux kernel from source in under 15 seconds. In 2004 he calculated the most digits of pi ever computed at the time, using a personal desktop computer and an algorithm he derived himself called Bellard's formula. In 2011 he wrote a complete PC emulator in pure JavaScript that runs Linux in your browser, a project called JSLinux that engineers still cannot believe is real. In 2019 he released QuickJS, a small but complete JavaScript engine that fits where V8 cannot. In 2021 he released NNCP, a neural network based lossless data compressor that immediately took the lead on the Large Text Compression Benchmark. Then he turned his attention to large language models. He built TextSynth Server, a web server with a REST API for running LLMs locally. He released ts_zip and ts_sms, compression utilities that use language models to compress text and short messages at ratios traditional algorithms cannot reach. He released TSAC, a very low bitrate audio compression system. In December 2025 he released Micro QuickJS, a new JavaScript engine for microcontrollers, separate from QuickJS, designed for environments with almost no memory. Fabrice co-founded a telecom company called Amarisoft in 2012, where he serves as CTO. Amarisoft builds 4G and 5G base station software used by carriers and labs around the world. He has been running it for over a decade while continuing to ship personal projects from his own home page at bellard dot org He has no Twitter. He has no Instagram. He gives almost no interviews. His personal website is a flat list of projects with no styling, no fonts, no marketing copy. Just titles and links. A quiet French engineer who never moved to Silicon Valley wrote the code that quietly runs the internet. He is still shipping.

bigaiguy's tweet photo. A French engineer who lives quietly in Paris has spent 30 years writing software that the entire internet now runs on without knowing his name.

He wrote the code that streams every YouTube video, every Netflix show, every TikTok clip. He wrote the code that runs the virtual servers underneath AWS, Google Cloud, and Microsoft Azure. He calculated more digits of pi than anyone in history. He has no Twitter. He has no marketing. He just keeps shipping.

His name is Fabrice Bellard.

Here is the story, because almost nobody outside the systems programming world knows what one man has built.

Fabrice was born in 1972 in Grenoble, France. He studied at École Polytechnique, the top French engineering school. He never went to Silicon Valley. He never built a startup empire. He just wrote code.

In 2000 he started a project called FFmpeg, an open-source multimedia framework for encoding, decoding, and streaming video. He was 28. The project did one thing nobody else had done well. It handled every video and audio format that existed, in one library, on every operating system. He led it himself for years.

Today FFmpeg is the invisible engine of the internet. YouTube uses it. Netflix uses it. VLC uses it. Chrome and Firefox use parts of it. Every Android phone, every iPhone, every smart TV, every video editing tool you have ever touched runs FFmpeg somewhere underneath. If you have watched a video on a screen in the last 20 years, Fabrice's code processed it.

He was not done.

In 2003 he started QEMU, a machine emulator and virtualizer. He wrote it solo until version 0.7.1 in 2005. QEMU lets you run any operating system on any other operating system. It became the foundation of modern virtualization. KVM, the Linux kernel hypervisor, runs on top of QEMU. Every major cloud provider, AWS, Google Cloud, Microsoft Azure, IBM Cloud, runs virtual machines on infrastructure built around it. The Quick Emulator is the most cited piece of cloud infrastructure code on Earth.

He kept going.

In 2001 he won the International Obfuscated C Code Contest with a small C compiler that grew into TCC, the Tiny C Compiler. TCC can compile and boot a Linux kernel from source in under 15 seconds. In 2004 he calculated the most digits of pi ever computed at the time, using a personal desktop computer and an algorithm he derived himself called Bellard's formula. In 2011 he wrote a complete PC emulator in pure JavaScript that runs Linux in your browser, a project called JSLinux that engineers still cannot believe is real.

In 2019 he released QuickJS, a small but complete JavaScript engine that fits where V8 cannot. In 2021 he released NNCP, a neural network based lossless data compressor that immediately took the lead on the Large Text Compression Benchmark.

Then he turned his attention to large language models. He built TextSynth Server, a web server with a REST API for running LLMs locally. He released ts_zip and ts_sms, compression utilities that use language models to compress text and short messages at ratios traditional algorithms cannot reach. He released TSAC, a very low bitrate audio compression system. In December 2025 he released Micro QuickJS, a new JavaScript engine for microcontrollers, separate from QuickJS, designed for environments with almost no memory.

Fabrice co-founded a telecom company called Amarisoft in 2012, where he serves as CTO. Amarisoft builds 4G and 5G base station software used by carriers and labs around the world. He has been running it for over a decade while continuing to ship personal projects from his own home page at bellard dot org

He has no Twitter. He has no Instagram. He gives almost no interviews. His personal website is a flat list of projects with no styling, no fonts, no marketing copy. Just titles and links.

A quiet French engineer who never moved to Silicon Valley wrote the code that quietly runs the internet.

He is still shipping.

370

25K

11K

bingnan retweeted

Spencer Baggins

@bigaiguy

7 days ago

A kid from Shantou with no programming background got into Tsinghua University's computer science department by winning a high school informatics olympiad, then quit a job offer from Google Brain to start an AI company in Beijing with two of his college bandmates. Four years later that company released a 1 trillion parameter open-source AI model that outperformed every American closed model on coding benchmarks and became the fastest Chinese tech company in history to hit a $10 billion valuation. His name is Yang Zhilin. The company is called Moonshot AI. The model is called Kimi K2. Here is the story. Yang was born in 1992 in Shantou, a small city in Guangdong province. In high school he had never written a line of code. He got selected for an informatics olympiad training program anyway. He won first prize at the Guangdong provincial level, which got him guaranteed admission to Tsinghua University. He scored 667 on the gaokao, far above the Tsinghua cutoff. But the system placed him in Thermal Energy Engineering. He transferred to Computer Science in his sophomore year. He graduated in 2015 ranked first in his department class. During undergrad he was advised by Tang Jie, a Tsinghua professor who would later co-found another Chinese AI giant called Zhipu. He went to Carnegie Mellon for his PhD under Ruslan Salakhutdinov and William Cohen. He finished in under four years. During that time he co-authored two of the most influential papers in modern AI, Transformer-XL and XLNet, which together shaped the long-context capabilities every modern LLM relies on. He worked at Facebook AI Research and Google Brain. He contributed to the original Google Gemini and Bard projects. Then in November 2022 ChatGPT launched. Yang flew back to the United States, looked at what OpenAI had done, and made up his mind. He told an interviewer later that he sensed two things were about to move at once, capital and talent, and when those two move together it is the rare moment when you can build a company from zero to one whose only purpose is AGI. In March 2023 he founded Moonshot AI in Beijing with two Tsinghua classmates, Zhou Xinyu and Wu Yuxin. The three of them had been bandmates in a college rock group called Splay. The company name is a tribute to Pink Floyd's Dark Side of the Moon, Yang's favorite album. The company launched on the album's 50th anniversary. He raised $60 million and built a 40-person team in three months. By 2024 he had raised over a billion dollars. The investors included Alibaba, Tencent, and Sequoia China. Moonshot AI became the fastest Chinese startup in history to reach a $10 billion valuation. ByteDance took four years. Pinduoduo took three. Moonshot did it in two. In October 2023 they launched Kimi, a consumer chatbot with a 200,000 character context window, the longest in the world at the time. By 2024 it was running on Chinese hardware and had tens of millions of users. Then on July 11, 2025 they released Kimi K2. K2 is a 1 trillion parameter Mixture of Experts model that activates 32 billion parameters per inference. It is open weights. It beat GPT-4 and Claude on coding benchmarks. It outperformed DeepSeek V3 on agent tasks and tool use. Former OpenAI researcher Andrew Carr publicly said K2 communicates differently than other models, refusing to be sycophantic and pushing back on bad ideas the way few models do. By early 2026 Moonshot had crossed $240 million in revenue. The Kimi K2.5 release exceeded the entire 2025 revenue total in under 20 days. K2.6 dropped in April 2026 with a SWE-Bench Pro score of 58.6, ahead of the leading closed-weight coding models at the time. A kid from Shantou who had never coded a line in high school just released the open-source model that competes head-on with everything OpenAI, Google, and Anthropic have shipped. He named it after the dark side of the moon.

bigaiguy's tweet photo. A kid from Shantou with no programming background got into Tsinghua University's computer science department by winning a high school informatics olympiad, then quit a job offer from Google Brain to start an AI company in Beijing with two of his college bandmates. Four years later that company released a 1 trillion parameter open-source AI model that outperformed every American closed model on coding benchmarks and became the fastest Chinese tech company in history to hit a $10 billion valuation.

His name is Yang Zhilin. The company is called Moonshot AI. The model is called Kimi K2.

Here is the story.

Yang was born in 1992 in Shantou, a small city in Guangdong province. In high school he had never written a line of code. He got selected for an informatics olympiad training program anyway. He won first prize at the Guangdong provincial level, which got him guaranteed admission to Tsinghua University.

He scored 667 on the gaokao, far above the Tsinghua cutoff. But the system placed him in Thermal Energy Engineering. He transferred to Computer Science in his sophomore year. He graduated in 2015 ranked first in his department class. During undergrad he was advised by Tang Jie, a Tsinghua professor who would later co-found another Chinese AI giant called Zhipu.

He went to Carnegie Mellon for his PhD under Ruslan Salakhutdinov and William Cohen. He finished in under four years. During that time he co-authored two of the most influential papers in modern AI, Transformer-XL and XLNet, which together shaped the long-context capabilities every modern LLM relies on. He worked at Facebook AI Research and Google Brain. He contributed to the original Google Gemini and Bard projects.

Then in November 2022 ChatGPT launched.

Yang flew back to the United States, looked at what OpenAI had done, and made up his mind. He told an interviewer later that he sensed two things were about to move at once, capital and talent, and when those two move together it is the rare moment when you can build a company from zero to one whose only purpose is AGI.

In March 2023 he founded Moonshot AI in Beijing with two Tsinghua classmates, Zhou Xinyu and Wu Yuxin. The three of them had been bandmates in a college rock group called Splay. The company name is a tribute to Pink Floyd's Dark Side of the Moon, Yang's favorite album. The company launched on the album's 50th anniversary.

He raised $60 million and built a 40-person team in three months. By 2024 he had raised over a billion dollars. The investors included Alibaba, Tencent, and Sequoia China. Moonshot AI became the fastest Chinese startup in history to reach a $10 billion valuation. ByteDance took four years. Pinduoduo took three. Moonshot did it in two.

In October 2023 they launched Kimi, a consumer chatbot with a 200,000 character context window, the longest in the world at the time. By 2024 it was running on Chinese hardware and had tens of millions of users.

Then on July 11, 2025 they released Kimi K2.

K2 is a 1 trillion parameter Mixture of Experts model that activates 32 billion parameters per inference. It is open weights. It beat GPT-4 and Claude on coding benchmarks. It outperformed DeepSeek V3 on agent tasks and tool use. Former OpenAI researcher Andrew Carr publicly said K2 communicates differently than other models, refusing to be sycophantic and pushing back on bad ideas the way few models do.

By early 2026 Moonshot had crossed $240 million in revenue. The Kimi K2.5 release exceeded the entire 2025 revenue total in under 20 days. K2.6 dropped in April 2026 with a SWE-Bench Pro score of 58.6, ahead of the leading closed-weight coding models at the time.

A kid from Shantou who had never coded a line in high school just released the open-source model that competes head-on with everything OpenAI, Google, and Anthropic have shipped.

He named it after the dark side of the moon.

681

119

374

72K

bingnan retweeted

Claude

@claudeai

15 days ago

Introducing Claude Opus 4.8: it builds on Opus 4.7 with sharper judgment, more honesty about its own progress, and the ability to work independently for longer than its predecessors. Available today at the same price.

claudeai's tweet photo. Introducing Claude Opus 4.8: it builds on Opus 4.7 with sharper judgment, more honesty about its own progress, and the ability to work independently for longer than its predecessors.

Available today at the same price. https://t.co/EufxL7T1kb

68K

15M

bingnan retweeted

Lex Fridman

@lexfridman

about 1 month ago

Here's my conversation all about @FFmpeg, the legendary open-source software powering most video on the Internet. In the episode, I talk with Jean-Baptiste Kempf and Kieran Kunhya. JB is lead developer of VLC and Kieran is FFmpeg contributor, codec engineer, and the person behind the now-infamous @FFmpeg account on X. VLC (@videolan), by the way, is also a legendary piece of open-source software: it's a video player that can open basically anything & has been downloaded over 6 billion times. I think both FFmpeg and VLC are two of the most important and impactful software systems ever created, both open source, and both created & maintained by volunteers: brilliant engineers from all walks of life. Thank you to everyone who contributed to FFmpeg and VLC, and in general to all engineers giving their heart & soul to building systems used by millions (or billions) of people, and often doing so not for money, status, or fame, but purely for the love of building great software and doing good for the world. Thank you to the builders! 🙏❤️ Shoutouts in this chat to @ID_AA_Carmack @karpathy @elonmusk @TimSweeneyEpic and everyone who is a contributor & fan of open source! It's here on X in full and is up everywhere else (see comment). Timestamps: 0:00 - Episode highlight 2:17 - Introduction 5:35 - Weirdest things VLC opens 9:59 - How video playback works 19:20 - Video codecs and containers 30:07 - FFmpeg explained 51:07 - Linus Torvalds 55:46 - Turning down millions to keep VLC ad-free 1:10:04 - FFmpeg & Google drama 1:29:18 - FFmpeg developers 1:35:55 - VLC and FFmpeg 1:40:29 - History of FFmpeg 1:43:46 - Reverse engineering codecs 1:57:01 - FFmpeg testing 2:01:08 - Assembly code (handwritten) 2:25:26 - Rust programming language 2:34:42 - FFmpeg and Libav fork 2:43:04 - Open source burnout 2:50:51 - x264 and internet video 3:04:07 - Video compression basics 3:11:04 - CIA and fake VLC 3:21:39 - Ultra low latency streaming 3:39:07 - AV2 codec and video patents 3:48:59 - VLC backdoors 3:59:14 - Video archiving 4:05:51 - Future of FFmpeg and VLC

196

552

492K

Who to follow

多党制的推崇以及支持者，希望2030年以前实现多党统治，多权分立！支持并参与中国鹰派！

bingnan retweeted

29 days ago

I'm traveling the world for a bit, starting with China but then hopping around the globe, anywhere. Open to any adventure. No plans, only a backpack. Hoping to meet & get to know humans from all walks of life. The pic is from a long hike on the Great Wall. For me, as a fan of history, this was an epic experience. In China, first I'm visiting a few big cities & talking to engineers at the heart of China's AI revolution. After that, if feeling crazy enough, I'm hitchhiking (first time) across rural China for a few weeks. Hitchhiking because I think it's the best way to meet rural folks who I would otherwise never get the chance to meet. I hope to do the same in US and other places. I have a request, if you have a travel recommendation, fill out the form(s) below if you feel like it. Or share with folks who might have advice about such travel. Form 1 - travel recommendation: If you can, recommend to me an interesting place I should visit anywhere in the world. For this, fill out form 1. Not touristy stuff, but something off the beaten path, that tourists may not know about, but is legendary. It could be as remote as meeting a herder in the mountains who is a local legend. Asia, Middle East, Europe, India, South/North America, Africa, Australia, anywhere. In China, I'm hoping to visit maybe Heibei, Shanxi, Shaanxi, Gansu, Sichuan, Yunnan, etc, so recommendations for spots to visit are helpful. Form 2 - coffee: If you want to grab a coffee with me anywhere in the world, fill out form 2 (please don't use form 1 for that). Anyway, I hectically tossed stuff in backpack. Realizing I don't have a clear plan of any kind, which is probably the only way to do it. LFG. Love you all ❤️

lexfridman's tweet photo. I'm traveling the world for a bit, starting with China but then hopping around the globe, anywhere. Open to any adventure. No plans, only a backpack. Hoping to meet & get to know humans from all walks of life. The pic is from a long hike on the Great Wall. For me, as a fan of history, this was an epic experience.

In China, first I'm visiting a few big cities & talking to engineers at the heart of China's AI revolution. After that, if feeling crazy enough, I'm hitchhiking (first time) across rural China for a few weeks. Hitchhiking because I think it's the best way to meet rural folks who I would otherwise never get the chance to meet. I hope to do the same in US and other places.

I have a request, if you have a travel recommendation, fill out the form(s) below if you feel like it. Or share with folks who might have advice about such travel.

Form 1 - travel recommendation:
If you can, recommend to me an interesting place I should visit anywhere in the world. For this, fill out form 1. Not touristy stuff, but something off the beaten path, that tourists may not know about, but is legendary. It could be as remote as meeting a herder in the mountains who is a local legend. Asia, Middle East, Europe, India, South/North America, Africa, Australia, anywhere. In China, I'm hoping to visit maybe Heibei, Shanxi, Shaanxi, Gansu, Sichuan, Yunnan, etc, so recommendations for spots to visit are helpful.

Form 2 - coffee:
If you want to grab a coffee with me anywhere in the world, fill out form 2 (please don't use form 1 for that).

Anyway, I hectically tossed stuff in backpack. Realizing I don't have a clear plan of any kind, which is probably the only way to do it. LFG.

Love you all ❤️

15K

665

bingnan retweeted

李志 | Rational Investing

@LZRationalnvest

18 days ago

SpaceX招股书中，让我最震撼的四张图片：第一张：穿越火墙，再入大气层第二张：地球与火星第三张：猎鹰9号划破夜空第四张：33台猛禽的怒吼

12K

bingnan retweeted

Tesla Owners Silicon Valley @teslaownersSV

19 days ago

Engineering is magic

981

79K

bingnan retweeted

Chun

@satofishi

22 days ago

Starship will fly us to Mars. 🚀

160

111

109K

bingnan retweeted

宝玉

@dotey

about 1 month ago

Codex 的野心，MCP 和 Skill 的下一步这段时间我在密集使用 Codex App、Cursor 等 Agent 应用，有件事越来越觉得有意思。去年大家争的是谁家模型更强，今年争的好像变成了谁家窗口右侧更好用。 Codex、Claude 桌面版��Cursor 3.0、TRAE SOLO，这几家最顶尖的 Agent，在��全没有协商的情况下，几乎同时收敛到了同一个界面布局：左侧是项目和会话列表，中间是和 Agent 的对话，右侧是工作区，放着文件浏览、网页预览、文件变更审查这些功能。肯定不是相互之间的抄袭，更像是当前 Agent 交互的最优解。【1】为什么是三栏传统 Chatbot 只需要两栏，左边会话历史，右边对话窗口，你问它答，用完走人。到了 Agent 时代，Agent 能自己写代码、改文件、调工具了。它做完之后，你得看看有没有做对——右侧工作区就是为这件事出现的。但这只是第一阶段。随着用户越来越多时间是在指挥 Agent，打开 VSCode 这类专业��具的时间自然越来越少。那个问题迟早会冒出来：Agent 帮你写完代码、做完 PPT，你想微调几个字，还要专门切出去打开另一个软件？没有人愿意这样。用户的自然期待是：能不能直接在 Agent 里改？这也是目前 Codex App 呼声最高的功能之一（另一个呼声高的是手机版，马上要出了）。于是各家开始悄悄升级右侧工作区，让它从只能看文件编辑记录，变成了一个多功能区。Codex 在 4 月 16 日的大版本更新里，右侧工作区的改动幅度是所有功能里最大的。交互细节上各家略有差异。Codex 和 Cursor 用 Tab 切换，Claude 用浮动面板。我自己用下来觉得 Codex 最顺手，Claude 的浮动面板方案设计感有余、实用性不足，迟早要改。【2】Codex 的真正野心但如果只把这个变化读成“设计界面进化”，就低估 Codex 了。 Codex 4 月大版本发布时的口号是“Codex for (almost) everything”——几乎任何任务都能做。你可以把它理解成一句广告口号，但更像是一个产品方向的声明。要兑现这句话，Codex 不能只是个擅长写代码的 Agent，它必须能处理各种文件格式，支持各领域的专业工作流，还要让用户能在它里面完成全程闭环，包括最后的人工微调。目前 Codex 还做不到最后一步：生成之后无法编辑，代码、Markdown、PPTX 都不行。这可能是产品上有意为之的克制，可能是技术上还没跑通，也可能是在等一个统一的解决方案出现。我猜是第三种。【3】MCP 和 Skill 都只解决了一半要理解 Codex 在等什么，得先想清楚 Agent 能力拼图里现在差哪一块。 MCP 解决了“连接”问题：Agent 通过统一规范接入各种工具，数据库、日历、代码仓库，都能打通。 Agent Skills 解决了“怎么做”的问题：Agent 学会了它没训练过的领域知识和最佳实践，比如怎么写特定风格的文章，怎么处理某类复杂任务。这两件事做得都还不错。但有一块缺口始终没补上：用户的二次编辑。你让 AI 写完一篇文章，最后还是要自己打开编辑器改几处，毕竟很多时候最后那 5% 的精准度，只有自己动手才能到位。就算将来 AI 再聪明，它也做不到百分百的懂你，还是少不了要手动去做修改。于是最近 Markdown 编辑器又火了，各种 Vibe Coding 出来的 Markdown 产品满天飞。但 Codex 不会自己做一个 Markdown 编辑器，因为每个人的偏好都不一样，做出来永远有人不满意；更何况它也不��能把每个垂直领域的专业编辑器都集成进来。最合理的路，是插件机制。【4】下一步：Agent 版 App Store 把 Agent 做成平台，让社区来贡献插件，就像 VSCode 和 Chrome 那样。 Codex 只需要聚焦在 Agent 调度这一层，把文件预览、二次编辑、垂直领域的专业能力都交给插件来扩展。用户按需安装，做设计的装设计插件，写作者装写作插件。插件机制还能顺手解决一个长期没有答案的问题：Skill 没办法商业化。我自己的 baoyu-skills 快 2 万 Star 了，但从中赚到的钱是 $0。Skill 这东西几乎是透明的，对 Agent 透明，对人也透明，复刻成本极低，不管你写得再好，护城河都很浅。插件不一样。App Store 和 Chrome 插件市场已经跑通了一套收费和版权保护机制，把它移植到 Agent 插件市场完全可行。好插件可以收费，开发者才有持续打磨的动力，生态才真正能转起来。 Codex 现在已经有了一个非常原始的插件市场。从这里到成熟的收费插件生态，还有很长的路，但方向是对的。想做这件事的不止 Codex 一家。Cursor 我能看到类似的影子。唯独 Claude Code 和 Cowork，目前没看到这个方向的产品迹象——也许他们不屑于做，也许只是还没走到这一步。【5】留给中小团队的窗口如果 Codex 真的跑通了插件生态，对中小团队意味着什么？除了自己做一个垂直 Agent，还有另一条路：在 Codex 这样的平台上做插件。不用自己搭 Agent 调度层，不用解决 Token 接入，用户分发也靠平台。你只需要专注在那个“最后一公里”——帮用户把 Agent 生成的结果处理好、编辑好、用得顺手。这个窗口不会开太久。先进去的能拿到冷启动红利，晚进去的只剩存量竞争。时间点不会太远，也许就在这几个月。 Codex 的野心摆在那里，“几乎任��任务”这个口号要��正兑现，插件机制是绕不过去的一步。如果 OpenAI 在这件事上继续犹豫，那才是真的失误。你觉得这个插件生态最后会是哪家先跑通？或者说你觉得有更适合 Agent 的产品表现形式？欢迎留言分享！

dotey's tweet photo. Codex 的野心，MCP 和 Skill 的下一步

这段时间我在密集使用 Codex App、Cursor 等 Agent 应用，有件事越来越觉得有意思。

去年大家争的是谁家模型更强，今年争的好像变成了谁家窗口右侧更好用。

Codex、Claude 桌面版��Cursor 3.0、TRAE SOLO，这几家最顶尖的 Agent，在��全没有协商的情况下，几乎同时收敛到了同一个界面布局：左侧是项目和会话列表，中间是和 Agent 的对话，右侧是工作区，放着文件浏览、网页预览、文件变更审查这些功能。

肯定不是相互之间的抄袭，更像是当前 Agent 交互的最优解。

【1】为什么是三栏

传统 Chatbot 只需要两栏，左边会话历史，右边对话窗口，你问它答，用完走人。

到了 Agent 时代，Agent 能自己写代码、改文件、调工具了。它做完之后，你得看看有没有做对——右侧工作区就是为这件事出现的。

但这只是第一阶段。

随着用户越来越多时间是在指挥 Agent，打开 VSCode 这类专业��具的时间自然越来越少。那个问题迟早会冒出来：Agent 帮你写完代码、做完 PPT，你想微调几个字，还要专门切出去打开另一个软件？

没有人愿意这样。用户的自然期待是：能不能直接在 Agent 里改？这也是目前 Codex App 呼声最高的功能之一（另一个呼声高的是手机版，马上要出了）。

于是各家开始悄悄升级右侧工作区，让它从只能看文件编辑记录，变成了一个多功能区。Codex 在 4 月 16 日的大版本更新里，右侧工作区的改动幅度是所有功能里最大的。

交互细节上各家略有差异。Codex 和 Cursor 用 Tab 切换，Claude 用浮动面板。我自己用下来觉得 Codex 最顺手，Claude 的浮动面板方案设计感有余、实用性不足，迟早要改。

【2】Codex 的真正野心

但如果只把这个变化读成“设计界面进化”，就低估 Codex 了。

Codex 4 月大版本发布时的口号是“Codex for (almost) everything”——几乎任何任务都能做。你可以把它理解成一句广告口号，但更像是一个产品方向的声明。

要兑现这句话，Codex 不能只是个擅长写代码的 Agent，它必须能处理各种文件格式，支持各领域的专业工作流，还要让用户能在它里面完成全程闭环，包括最后的人工微调。

目前 Codex 还做不到最后一步：生成之后无法编辑，代码、Markdown、PPTX 都不行。这可能是产品上有意为之的克制，可能是技术上还没跑通，也可能是在等一个统一的解决方案出现。

我猜是第三种。

【3】MCP 和 Skill 都只解决了一半

要理解 Codex 在等什么，得先想清楚 Agent 能力拼图里现在差哪一块。

MCP 解决了“连接”问题：Agent 通过统一规范接入各种工具，数据库、日历、代码仓库，都能打通。

Agent Skills 解决了“怎么做”的问题：Agent 学会了它没训练过的领域知识和最佳实践，比如怎么写特定风格的文章，怎么处理某类复杂任务。

这两件事做得都还不错。但有一块缺口始终没补上：用户的二次编辑。

你让 AI 写完一篇文章，最后还是要自己打开编辑器改几处，毕竟很多时候最后那 5% 的精准度，只有自己动手才能到位。就算将来 AI 再聪明，它也做不到百分百的懂你，还是少不了要手动去做修改。

于是最近 Markdown 编辑器又火了，各种 Vibe Coding 出来的 Markdown 产品满天飞。

但 Codex 不会自己做一个 Markdown 编辑器，因为每个人的偏好都不一样，做出来永远有人不满意；更何况它也不��能把每个垂直领域的专业编辑器都集成进来。

最合理的路，是插件机制。

【4】下一步：Agent 版 App Store

把 Agent 做成平台，让社区来贡献插件，就像 VSCode 和 Chrome 那样。

Codex 只需要聚焦在 Agent 调度这一层，把文件预览、二次编辑、垂直领域的专业能力都交给插件来扩展。用户按需安装，做设计的装设计插件，写作者装写作插件。

插件机制还能顺手解决一个长期没有答案的问题：Skill 没办法商业化。

我自己的 baoyu-skills 快 2 万 Star 了，但从中赚到的钱是 $0。Skill 这东西几乎是透明的，对 Agent 透明，对人也透明，复刻成本极低，不管你写得再好，护城河都很浅。

插件不一样。App Store 和 Chrome 插件市场已经跑通了一套收费和版权保护机制，把它移植到 Agent 插件市场完全可行。好插件可以收费，开发者才有持续打磨的动力，生态才真正能转起来。

Codex 现在已经有了一个非常原始的插件市场。从这里到成熟的收费插件生态，还有很长的路，但方向是对的。

想做这件事的不止 Codex 一家。Cursor 我能看到类似的影子。唯独 Claude Code 和 Cowork，目前没看到这个方向的产品迹象——也许他们不屑于做，也许只是还没走到这一步。

【5】留给中小团队的窗口

如果 Codex 真的跑通了插件生态，对中小团队意味着什么？

除了自己做一个垂直 Agent，还有另一条路：在 Codex 这样的平台上做插件。不用自己搭 Agent 调度层，不用解决 Token 接入，用户分发也靠平台。你只需要专注在那个“最后一公里”——帮用户把 Agent 生成的结果处理好、编辑好、用得顺手。

这个窗口不会开太久。先进去的能拿到冷启动红利，晚进去的只剩存量竞争。

时间点不会太远，也许就在这几个月。

Codex 的野心摆在那里，“几乎任��任务”这个口号要��正兑现，插件机制是绕不过去的一步。如果 OpenAI 在这件事上继续犹豫，那才是真的失误。

你觉得这个插件生态最后会是哪家先跑通？或者说你觉得有更适合 Agent 的产品表现形式？欢迎留言分享！

128

725

101

699

141K

段兵楠 @bingnan

23 days ago

@leeoxiang local first有啥好处吗？不��该是在云端更好吗

bingnan retweeted

Jiayuan (JY) Zhang

@jiayuan_jy

2 months ago

开源 Multica：专为 AI-native 团队设计的 Agent + 人的协作平台 https://t.co/Wx0IutujjR 为什么做 Multica？ Multica 最初是为了解决我们团队自己的问题： 1. 团队间的知识无法共享。每个人都在用 coding agent，但产出的上下文全部散落在各自的 agent session 里。A 做完了一件事，B 不知道；agent 跑完了一轮，结果只有发起人看得到。团队知识变成了一座座孤岛。 2. 多人 + 多 Agent 的协作缺乏中枢。当团队同时有多个 agent 在跑任务，谁在做什么、做到哪了��卡住了没有——没有一个地方能看到全貌。人和 agent 之间、agent 和 agent 之间，缺少一个共同的协作界面。 Multica 是什么？一句话：像 Linear 一样管理任务，但 AI agent 是一等公民。你可以像分配任务给同事一样，把 issue 分配给 agent。agent 会自动领取任务、在你的本地机器上执行代码、提交结果、更新状态、发表评论——一切都发生在同一个看板里，所有人实时可见。核心思路很简单：每个人把自己的 coding agent（Claude Code / Codex）注册到团队 workspace，之后就可以像分配任务给同事一样分配给 agent。agent 自动执行、更新状态、发表评论，所有人实时可见。适合谁？ - 1-10 人的 AI-native 小团队 - 正在大量使用 coding agent 但缺少协作中枢的团队 - 希望让 agent 融入日常工作流而不是当作独立工具的团队官网: https://t.co/sdFwgJM8KU 欢迎 star、试用、提 issue，也欢迎 PR。

jiayuan_jy's tweet photo. 开源 Multica：专为 AI-native 团队设计的 Agent + 人的协作平台

https://t.co/Wx0IutujjR

为什么做 Multica？

Multica 最初是为了解决我们团队自己的问题：

1. 团队间的知识无法共享。每个人都在用 coding agent，但产出的上下文全部散落在各自的 agent session 里。A 做完了一件事，B 不知道；agent 跑完了一轮，结果只有发起人看得到。团队知识变成了一座座孤岛。

2. 多人 + 多 Agent 的协作缺乏中枢。当团队同时有多个 agent 在跑任务，谁在做什么、做到哪了��卡住了没有——没有一个地方能看到全貌。人和 agent 之间、agent 和 agent 之间，缺少一个共同的协作界面。

Multica 是什么？

一句话：像 Linear 一样管理任务，但 AI agent 是一等公民。

你可以像分配任务给同事一样，把 issue 分配给 agent。agent 会自动领取任务、在你的本地机器上执行代码、提交结果、更新状态、发表评论——一切都发生在同一个看板里，所有人实时可见。

核心思路很简单：每个人把自己的 coding agent（Claude Code / Codex）注册到团队 workspace，之后就可以像分配任务给同事一样分配给 agent。agent 自动执行、更新状态、发表评论，所有人实时可见。

适合谁？

- 1-10 人的 AI-native 小团队
- 正在大量使用 coding agent 但缺少协作中枢的团队
- 希望让 agent 融入日常工作流而不是当作独立工具的团队

官网: https://t.co/sdFwgJM8KU

欢迎 star、试用、提 issue，也欢迎 PR。

112

172

388K

bingnan retweeted

Andrej Karpathy

@karpathy

25 days ago

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

150K

11K

14K

28M

段兵楠 @bingnan

27 days ago

🤨🤨🤨

bingnan retweeted

Zayn Hao

@ZaynHao

about 1 month ago

强烈推荐这期播客，这是最近听到的非常实用，信息不拖沓的 Claude Code (Anthropic)团队是如何运营组织，如何工作的分享。非常适合身处或正在管理构建 AI Agent 产品的团队朋友听。里面有很详实的故事来讲述他们团队的分工，大家的做事方式，以及变化等。很实用。

ZaynHao's tweet photo. 强烈推荐这期播客，

这是最近听到的非常实用，信息不拖沓的 Claude Code (Anthropic)团队是如何运营组织，如何工作的分享。

非常适合身处或正在管理构建 AI Agent 产品的团队朋友听。

里面有很详实的故事来讲述他们团队的分工，大家的做事方式，以及变化等。

很实用。 https://t.co/xVrH4Prhec

346

462

52K

bingnan retweeted

Berryxia.AI

@berryxia

about 1 month ago

Moonshot AI创始人杨植麟最近放出了一个40分钟视频。这位92年生、清华计算机本科第一、CMU博士、Transformer-XL和XLNet共同作者，前Google Brain和Meta研究员，坐在镜头前平静拆解了Kimi K2的整个训练过程。他们只花了460万美元。上周一场8模型实时编程大战，Kimi K2直接拿下第一，GPT-5.5排第三，Claude Opus 4.7第五。我看完后最大的感受是，AI竞赛的规则已经在悄然改变。所有人还在拼谁敢烧更多钱、堆更多算力，他却用极致优化、线性注意力、子代理这些硬核架构，把资源差距直接抹平甚至反超。 40分钟全是干货，零废话，把关键打法讲得清清楚楚。如果你正在做AI代理，或者准备2026年入场大模型赛道，这段视频强烈建议存下来周末慢慢看。小团队靠聪明架构，正在把大厂的传统玩法一点点颠覆。你还觉得只有堆钱才能赢吗��