零人类公司的崛起

今天在AI每日简报中，探讨零人工公司的崛起。

Today on the AI Daily Brief, the rise of the zero human company.

在此之前，先看头条新闻：Cursor在过去三个月内将其运行速率翻倍。

Before that in the headlines, Cursor Doubles Its Run Rate in the Last Three Months.

《AI每日简报》是一档每日更新的播客和视频节目，聚焦AI领域最重要的新闻与讨论。

The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

好了，朋友们，在深入内容之前，先做几个快速公告。

All right, friends, quick announcements before we dive in.

首先，感谢今天的赞助商：毕马威、Assembly、Blitzy和AIUC。

First of all, thank you to today's sponsors, KPMG, Assembly, Blitzy, and AIUC.

要获取无广告版本的节目，请前往patreon.com/aideallybrief，或在Apple播客上订阅。

To get an ad free version of the show, go to patreon.com/aideallybrief, or you can subscribe on Apple Podcasts.

如果您有兴趣赞助本节目，请发送邮件至sponsorsaidailybrief.dot.ai。

If you are interested in sponsoring the show, send us a note at sponsorsaidailybrief dot ai.

另外，在aideallybrief.ai上还有一个小消息：上个月我们开始推出每月AI使用情况脉搏调查，二月的脉搏调查现已上线。

And one other quick thing on aideallybrief.ai: Last month, we started doing monthly AI usage pulse surveys, and the pulse survey for February is now live.

这些调查让我们有机会与大家分享AI使用行为月度变化的趋势，如果你参与填写，你将比任何人都更早获得结果。

These surveys give us a chance to share with everyone how AI usage behavior is changing month over month, and if you contribute to it, you will get the results before anyone else.

你可以在aideallybrief.ai上找到这项调查，如果你能花两分钟完成这个简短的调查，我会非常感激。

Again, you can find that at aideallybrief.ai, and I would so appreciate it if you would take just the two minutes to fill out the quick survey.

在2026年的这档节目中，我们似乎一直在谈论智能代理编程的兴起，以及它在各个领域的普及，而不仅仅是软件工程。

All we can talk about on this show in 2026 seems to be the rise of agentic coding and its proliferation into all the sectors, not just software engineering.

如今，这种普及趋势已经开始体现在数据上。

And now that proliferation is showing up in the numbers.

消息人士告诉彭博社，Cursor在二月份的年度经常性收入（ARR）已突破20亿美元，三个月内翻了一倍。

Sources tell Bloomberg that Cursor surpassed $2,000,000,000 in ARR for February, doubling in three months.

这一消息让X平台上那些自诩为AI资深用户的用户大为震惊，他们过去几个月一直在Claude Code和Codex之间来回切换，最近却断言Cursor即将失败。

This news came as a massive shock to the enfranchised AI users on X who spent the last few months hopping between Claude Code and Codex and who have recently decided that Cursor is doomed.

二月时，AI金融软件初创公司Valen的开发负责人凯尔·拉塞尔发帖称：今天早上，我们的首席执行官安德鲁·王要求取消他的Cursor账户，因为他已深度转向Clawed Code，这引发了一系列内部申请。

At the February, Kyle Russell, who is a development lead at AI finance software startup Valen, posted: This morning our CEO Andrew Wang requested that his Cursor seat be removed since he's so deep into clawed code and it kicked off an internal cascade of requests.

这一连锁反应导致90个账户被取消，并促使X平台上许多人意识到，他们自己也已经有好几个月没用过Cursor了。

The cascade resulted in 90 canceled seats and triggered a wave of people on X noticing that they also haven't touched Cursor in months.

红杉资本的Didi Das指出，企业采购的动态与初创公司和个体创业者快速切换服务的情况大不相同。

Didi Das of Menlo Ventures noted that the dynamics of enterprise procurement are far different to the rapid service switching in startups and solopreneurship.

他写道：叙事崩塌：Cursor在三个月内从10亿美元增长到20亿美元。

He wrote: Narrative violation: Cursor goes from 1,000,000,000 to 2,000,000,000 in three months.

Claude Code在八个月内从零增长到25亿美元。

Claude Code went from zero to 2,500,000,000 in eight months.

科技圈和X平台上的所有人都以为人们正在全面抛弃Cursor，但企业层面的普及却极其缓慢。

Everyone in the tech and x bubble thinks people are wholesale ditching Cursor, but enterprise diffusion is glacial.

世界上大多数人才刚刚接触到它。

Most of the world just got ahold of it.

顺便说一句，这正是彭博社报道中所使用的相同表述。

That, by the way, is the exact same framing used in the Bloomberg reporting.

他们的消息来源称，Cursor收入的60%来自企业客户，新公司注册量和现有客户增加席位的数量都在上升。

Their source said that 60% of Cursor revenue is coming from corporate customers, with a rise in both new company signups and existing customers adding more seats.

风险投资人休伯特·特布劳特写道：科技推特说：Cursor已经见顶了！

Venture investor Hubert Teblaut wrote: Tech Twitter says: Cursor peaked!

大家都已经转向智能代理了！

Everyone's already moved on to agents!

下一个热潮！

Next hype!

年经常性收入在三个月内翻倍至20亿美元，采用曲线的上升空间依然巨大。

ARR just doubled in three months to $2,000,000,000 The adoption S curve still has tons of runway left.

早期采用者可能已经转移了，但主流用户终于开始入场。

Early adopters might be moving on, but the mainstream is finally showing up.

人才平台Promote的首席执行官乔·范德沃特也指出，Cursor和Claude Code在企业环境中实际上存在一些重要差异，他评论道：Cursor在多个工程师共享的大型代码库中表现卓越。

Job VanderVoort, the CEO of talent starter Promote, also noted that there's actually some meaningful differences between Cursor and Claude code in the enterprise, commenting: Cursor is amazing for large code bases shared across many engineers.

基本上，这条新闻有力地表明，AI初创公司并不像X平台上的喧嚣群体所认为的那样在玩零和游戏。

Basically, the news hammers home that AI startups just aren't playing a zero sum game as much as the chattering class on X would like it to be so.

当前的市场动态并不是Claude Code从Cursor手中夺取市场份额。

The current market dynamic isn't about Claude Code taking market share from Cursor.

而是整个领域正在快速增长。

It's about the entire segment growing and growing fast.

而且这种增长看起来迅速且可持续。

And that growth looks rapid and sustainable.

SPTK评论道：AI编码代理已经不再是炒作了。

SPTK commented: AI coding agents aren't hype anymore.

它们已经成为基础设施。

They're infrastructure.

事实上，AI编码需求的激增正是我们下一个故事的核心，周一早上，Claude用户发现服务宕机，出现了再熟悉不过的错误信息。

Indeed, surging demand for AI coding is at the heart of our next story, producing an all too familiar error message on Monday morning with Claude users finding the service was down.

Claude的宕机在早上6点40分左右达到峰值。

Claude's outage peaked at around 06:40 a.

M.

M.

就在我在孩子起床前使用它的时候。

Right as I was using it before the kids got up.

到早上8点40分，投诉减少了三分之一，但Anthropic在WhatsApp声明中表示，面向消费者的服务仍无法使用。

Complaints fell by a third by 08:40, but Anthropics said in a WhatsApp statement that consumer facing services were still unavailable.

他们写道：我们感谢大家的耐心，过去一周由于对Quad的需求空前高涨，我们正在努力恢复服务。

They wrote, We appreciate everyone's patience as we work to bring things back online while experiencing unprecedented demand for Quad over the last week.

如今，Anthropic过去也曾面临过类似问题。

Now, Anthropic has struggled with this in the past.

该公司在扩展过程中一直受到计算资源的严重限制，尤其是在新模型发布期间。

The company has suffered massive compute constraints as they scaled, especially around new model releases.

然而值得注意的是，Opus和Sonnet 4.6的发布并未引发重大投诉，但此次使用量的激增显然比模型发布更难以预料。

It is worth noting, however, that the launch of Opus and Sonnet 4.6 featured no major complaints, but this surge in usage of course was far less foreseeable than a model release.

我当然指的是，在国防部那场风波之后，Anthropic的下载量大幅飙升。

I'm referring of course to the huge uptick in anthropic downloads that came in the wake of the whole screffuffle with the DoD.

事实上，根据Sensor Tower的数据，周五到周六一天之内，ChatGPT的卸载量翻了三倍。

Indeed, according to data from Sensor Tower, ChatGPT uninstalls tripled in a day between Friday and Saturday.

美国。

U.

美国。

S.

周六的日下载量较前一日下降了13%，周日又进一步下降了5%，这与此前趋势形成鲜明对比——周五时ChatGPT的日下载量还增长了14%。

Daily downloads fell by 13% day over day on Saturday and dropped by another 5% on Sunday, which was a sharp break from the prior trend where ChatGPT downloads had been gaining 14% day over day on Friday.

周六，ChatGPT的一星评价暴增了775%，周日又再增加了100%。

One star reviews for ChatGPT surged by 775% on Saturday and another 100% on Sunday.

在市场另一端，Anthropic跃居App Store榜首，得益于其自身下载量的激增。

On the other side of the market, Anthropix rise to number one in the App Store was driven by their own surge in downloads.

Claude应用在周五增长了37%，周六又再增长了51%。

The Claude app gained 37% on Friday and another 51% on Saturday.

需要明确的是，根据SimilarWeb的数据，Claude在一个月内的下载量大约增长了20倍。

Now to be clear, Claude is seen download something like 20x in a month according to SimilarWeb.

当然，现在的问题是这种用户转移行为能持续多久。

Now the question, of course, is how persistent this switching behavior is.

目前，ChatGPT依然是绝对主导的消费级AI平台。

Right now ChatGPT remains by far the dominant consumer AI platform.

尽管目前线上抵制活动正在持续，但其长期影响力如何，仍有待观察。

And while the online boycott is active right now, how resonant it will be in the long run remains to be seen.

说到五角大楼问题，萨姆·阿尔特曼已向员工更新了OpenAI与五角大楼合同的修订情况，并承认这笔交易的达成过程显得有些草率。

Speaking of the whole Pentagon issue, Sam Altman has updated staff on revisions to OpenAI's Pentagon contract and acknowledged that the way that the deal came together looked a little sloppy.

在一份发给员工并同时在X平台上分享的备忘录中，阿尔特曼写道：我们一直在与国防部合作，在协议中增加一些内容，以明确我们的原则。

In a memo to staff also shared on X, Altman wrote: We have been working with the DOW to make some additions in our agreement to make our principles very clear.

他表示，合同将更新为加入如下条款：根据适用法律，该人工智能系统不得被故意用于对美国

He said the contract will be updated to add language that states, Consistent with applicable laws, the AI system shall not be intentionally used for domestic surveillance of U.

的

S.

人员和国民进行监控。

Persons and nationals.

为避免疑义，国防部理解这一限制旨在禁止故意追踪、监控或监视美国

For the avoidance of doubt, the Department understands this limitation to prohibit deliberate tracking, surveillance, or monitoring of U.

的

S.

人员或国民，包括通过采购或使用商业获取的个人或可识别信息。

Persons or nationals, including through the procurement or use of commercially acquired personal or identifiable information.

他还补充说，国防部已同意OpenAI的系统不会被隶属于国防部的情报机构使用，特别提到了国家安全局。

Added that the DOW has agreed that OpenAI systems won't be used by intelligence agencies nestled under the DOW, specifically mentioning the NSA.

阿尔特曼重申，他已明确向国防部表示，Anthropic不应被列为供应链风险，应给予与OpenAI相同的条款。

Altman reiterated that he was clear with the department that Anthropics should not be designated as supply chain risk and that they should be offered the same terms as OpenAI.

然而，五角大楼与Anthropic的争端在华盛顿持续发酵，官员们正在紧急制定人工智能政策。

Still, the fallout from the Pentagon's battle with Anthropic is reverberating throughout Washington as officials scramble on AI policy.

财政部、国务院和卫生与公共服务部均已根据总统周五的指令，停止使用Claude。

The Treasury, the State Department, and the Department of Health and Human Services have all pulled the plug on Claude following the President's Friday directive.

财政部长斯科特·贝森在X上宣布了这一决定，并发帖称：美国人民有理由相信，政府使用的每一项工具都应服务于公共利益。在特朗普总统领导下，任何私营公司都不得主导我国国家安全的条款。

Treasury Secretary Scott Besson announced the move on X, posting: The American people deserve confidence that every tool in government serves the public interest, and under President Trump, no private company will ever dictate the terms of our national security.

与此同时，在国会，民主党人正在准备回应国防部前所未有的行为——援引《国防生产法》将Anthropic列为供应链风险。

Meanwhile in Congress, Democrats are preparing a response to the unprecedented use of the Defense Production Act to label anthropic a supply chain risk.

硅谷代表萨姆·利卡尔多计划提出一项修正案，禁止各机构对那些寻求以限制技术部署方式来降低对美国公民风险的高风险技术供应商和开发者进行报复，该修正案将适用于人工智能等技术。

Silicon Valley representative Sam Liccardo plans to introduce an amendment to the act that would prohibit agencies from retaliating against vendors and developers of high risk technologies such as AI, where those vendors seek to limit the deployment of their technology in ways to mitigate the risk to United States citizens.

Axios报道称，五角大楼在上周的争端中并未正式援引《国防生产法》，这自然引发了更多担忧：政府是否在使用隐性威胁和胁迫手段，而非法定权力。

Axios reports that the Defense Production Act was not formally invoked by the Pentagon as part of last week's dispute, which of course only raises further concerns about the government using implicit threats and coercion rather than statutory powers.

维卡罗的法案预计将于周三在众议院进行审议，使其进入快速投票通道。

Vaccaro's bill is expected to be marked up in the House on Wednesday, putting it on the fast track for a vote.

参议院民主党人也在考虑一项法案，以应对五角大楼在自主武器和国内监控中使用人工智能技术的更广泛担忧。

Senate Democrats are also weighing a bill to address broader concerns about the Pentagon's use of AI technology in autonomous weaponry and domestic surveillance.

最终结果是，人工智能政策正被推上华盛顿的议事日程，成为关键时刻的紧迫议题。

The net result is that AI policy is being thrust on Washington as a live issue in a high stakes moment.

《Politico》在周一试图厘清这一局面。

POLITICO attempted to make sense of the landscape on Monday.

他们指出，当前局势正在搅动人工智能的政治格局，写道：这些并非党派之争，而是科技创始人、研究人员和倡导者之间的内部分歧，随着人工智能议题重要性上升，这些分歧在政治上变得愈发关键。

They noted that the situation is scrambling the politics of AI, writing: These aren't partisan arguments, but internal disagreements between tech focused founders, researchers, and advocates are becoming more important politically as the issue of AI rises in salience.

而在过去几天里，无论哈格塞特和特朗普是否意识到，这些分歧突然成为一场重大政治斗争的核心。

And in the past few days, they've suddenly become central to a hugely consequential political fight whether Hagsett and Trump are aware of them or not.

最后，今天来点轻松的推测性内容。

Lastly today, a fun little speculative one.

你可能还记得，大约在超级碗期间，我们看到一段泄露的视频，画面中亚历山大·斯卡斯加德戴着奇怪的钟形耳罩设备，手里拿着一个金属圆盘状物体。

You might remember that around the Super Bowl, we got this leaked video that looked like Alexander Skarsgard wearing these weird bell ear device things and holding a metallic puck shaped object.

最初的背景是，OpenAI 原计划在超级碗期间播放这支广告，但当 OpenAI 员工否认了这一传言后，大多数人认为这段视频只是个恶作剧。

The backstory initially was that OpenAI had originally planned to air that ad during the Super Bowl but when OpenAI staffers disavowed that rumor, most ended up chalking up the video as a hoax.

然而在周一，ADAM 创始人扎克·戴夫发布了一张照片和一段视频，显示 Airbnb 联合创始人、美国政府首席设计官乔·格比亚在旧金山的一家咖啡馆里。

On Monday, however, ADAM founder Zach Dive posted a picture and a video of Airbnb co founder and U.

美。

S.

政府首席设计官乔·格比亚在旧金山的一家咖啡馆里。

Government chief design officer Joe Gebbia in a San Francisco coffee shop.

在格比亚面前，放着一个金属圆盘，看起来与广告中的设备一模一样。

In front of Gebbia is a metallic puck that looks identical to the device from the advertisement.

如果你仔细看，他还戴着一对与广告中相同的金属耳塞。

And if you look closely, he's also wearing a pair of metallic earbuds that match the ones from the ad.

当这个消息传出时，我说过，如果这是早期的游击营销，我一点也不惊讶。

Now when this came out, I said that I wouldn't be surprised if this was early guerrilla marketing.

后来我们发现，这实际上是真的。

And we found out later on that this actually was real.

YouTuber和AI教育者马修·伯曼表示赞同，写道：‘阴谋论角落？’

YouTuber and AI educator Matthew Berman agrees, writing, Conspiracy corner?

这实际上是约翰尼·伊夫的OpenAI设备。

This is actually the Johnny Ivek's OpenAI device.

他们确实制作了这个广告，并决定采用否认并制造好奇心的营销策略。

They actually made this ad and decided the marketing approach will be deny and build curiosity.

现在，美国的首席设计官在咖啡馆里被拍到拿着这个设备。

Now they have the CDO of America getting caught in a coffee shop with a device.

我认为这一切都是为更大规模的宣传活动做的铺垫，如果是这样，那确实很酷。

Chalk me up as thinking this is all a plant for a broader campaign, which if that's the case is pretty cool.

不过目前，今天的头条新闻就先到这里。

For now however that is going do it for today's headlines.

接下来是主节目。

Next up the main episode.

Agenic AI正在推动一场价值3万亿美元的生产力革命，而领导者们正面临一个真正的决策时刻。

Agenic AI is powering a $3,000,000,000,000 productivity revolution and leaders are hitting a real decision point.

你是自己构建AI代理，购买现成的，还是通过合作借用以更快地扩展？

Do you build your own AI agents, buy off the shelf, or borrow by partnering to scale faster?

普华永道最新的一份思想领导力报告《应对构建、购买或借用的决策》出色地拨开了迷雾，提供了一个实用的框架，帮助你根据价值、风险和准备程度做出选择，并指导如何以正确的信任、治理和编排基础来扩展代理。

KPMG's latest thought leadership paper, Navigating the Build, Buy, or Borrow decision, does a great job cutting through the noise with a practical framework to help you choose based on value, risk, and readiness and how to scale agents with the right trust, governance, and orchestration foundation.

别选错模型。

Don't lock in the wrong model.

你现在就可以访问 www.kpmg.us/navigate 下载这份报告。

You can download the paper right now at www.kpmg.us/navigate.

再次提醒，网址是 www.kpmg.us/navigate。

Again, that's www.kpmg.us/navigate.

我之前跟你们聊过AssemblyAI和他们极其精准的语音AI模型，但他们刚刚推出了一项重大更新。

You've heard me talk about AssemblyAI and their insanely accurate voice AI models, but they just ship something big.

Universal Three Pro是一种首创的语音语言模型，允许你用自己领域的上下文和词汇来引导语音识别，而不是在后期处理中修正转录文本。

Universal three Pro is a first of its kind class of speech language model that lets you prompt speech recognition with your own domain context and vocabulary, instead of fixing transcripts in post processing.

它比传统的自动语音识别更灵活，比大语言模型更可预测，因此你能在源头获得准确的输出，捕捉到转录文本常忽略的人类语音中的情感，且无需定制模型或后期处理的技巧。

It's more flexible than traditional ASR and more deterministic than LLMs, so you get accurate output at the source and can capture the emotion behind human speech that transcripts often miss all without custom models or post processing hacks.

为了庆祝发布，他们将在整个二月免费提供试用。

And to celebrate the launch, they're making it free to try for all of February.

如果你正在开发任何语音相关的产品，这个工具值得一看。

If you're building anything with voice, this one's worth a look.

前往 assemblyai.com/freeoffer 了解详情。

Head to assemblyai.com/freeoffer to check it out.

你已经试过集成在IDE中的Copilot了。

You've tried In IDE Copilots.

它们速度很快，但只能看到你代码中的局部片段。

They're fast, but they only see local silos of your code.

当你在大型企业代码库中使用这些工具时，它们的效果会迅速下降。

Leverage these tools across a large enterprise codebase and they quickly become less effective.

根本的限制是什么？

The fundamental constraint?

上下文。

Context.

Blitzy 通过无限的代码上下文理解，深入到数百万行代码的每一行依赖关系，解决了这个问题。

Blitzy solves this with infinite code context understanding your codebase down to the line level dependency across millions of lines of code.

虽然 Copilot 帮助开发者更快地编写代码，但 Blitzy 能协调数千个代理，在整个代码库中进行推理。

While Copilots help developers write code faster, Blitzy orchestrates thousands of agents that reason across your full codebase.

让 Blitzy 承担繁重工作，自主完成每个冲刺任务的 80% 以上，并提供经过严格验证的代码。

Allow Blitzy to do the heavy lifting, delivering over 80% of every sprint autonomously with rigorously validated code.

Blitzy 会提供一份详尽的清单，列出需要人类与他们的 Copilot 协作完成的剩余工作。

Blitzy provides a granular list of the remaining work for humans to complete with their Copilots.

快速处理功能新增、大规模重构、遗留系统现代化和全新项目，速度提升五倍。

Tackle feature additions, large scale refactors, legacy modernization, greenfield initiatives, all 5x faster.

前往 blitzy.com 了解 Blitzy 的不同之处。

See the Blitzy difference at blitzy.com.

那就是 blitzy.com。

That's blitzy.com.

我认为，这将为企业级 AI 代理领域确立一个新的标准。

There's a new standard that I think is going to matter a lot for the enterprise AI agent space.

它被称为AIUC1，是世界上首个AI代理标准。

It's called AIUC1, and it builds itself as the world's first AI agent standard.

它旨在涵盖所有核心企业风险，例如数据与隐私、安全、可靠性、问责制和社会影响，均由可信第三方验证。

It's designed to cover all the core enterprise risks things like data and privacy, security, safety, reliability, accountability, and societal impact all verified by a trusted third party.

它引起我关注的原因之一是Eleven Labs——我之前提到过这家公司，目前正是一股不可忽视的力量——成为首家获得AIUC1认证的语音代理，并推出了全球首个可投保的AI代理。

One of the reasons it's on my radar is that Eleven Labs who you've heard me talk about before and is just an absolute juggernaut right now just became the first voice agent to be certified against AI UC one and is launching a first of its kind insurable AI agent.

这意味着在实践中，它具备实时防护机制，可阻止不安全的回应并防范操纵，同时配备完整的安全体系。

What that means in practice is real time guardrails that block unsafe responses and protect against manipulation, plus a full safety stack.

这种特性正是推动企业采用的关键。

This is the kind of thing that unlocks enterprise adoption.

当一家基于Eleven Labs构建的公司能够指向第三方认证，并宣称我们的代理是安全、可靠且经过验证的，这将彻底改变对话的格局。

When a company building on Eleven Labs can point to a third party certification and say our agents are secure, safe, and verified, that changes the conversation.

访问aiuc.com，了解全球首个AI代理标准。

Go to aiuc.com to learn about the world's first standard for AI agents.

那就是aiuc.com。

That's aiuc.com.

欢迎回到AI每日简报。

Welcome back to the AI Daily Brief.

今天我们要关注一个刚刚兴起的趋势，未来几周和几个月里，我们会听到更多关于它的讨论。

Today we are looking in on a trend which I think is just emerging but which we're going to hear a lot more about in the weeks and months to come.

事实上，我认为这是我们在试图全面理解和接纳新代理能力所带来变革时不可避免的下一步。

In fact, I think this is the inevitable next step as we try to fully understand and embrace the changes that new agentic capacity has unlocked.

过去几年围绕AI的一个重要主题当然是：那些充分拥抱AI并真正围绕它重构系统的人，拥有了前所未有的能力。

One of the big themes surrounding AI for the last couple of years has of course been the idea that those who fully embrace AI and who really rewire their systems around it have capabilities like they never had before.

当你能把这样一群人聚集在一起时，实际上可以实现显著的超越。

And that when you can get a number of those people together you can actually radically outperform.

这一主题在最近几届AI工程峰会上都曾探讨过。

This is a theme that's been explored at the last few AI Engineering Summits.

策展人Sean Wang（又称Swix）将这些团队称为‘微型团队’。

Curator Sean Wang aka Swix has called these tracks tiny teams.

在一篇关于微型团队的博客文章中，他写道：我曾将微型团队理想化地定义为ARR（年经常性收入）超过百万美元但员工人数更少的团队，因为效率是智力诚实的终极准则。

In a blog post about Tiny Teams he wrote: I previously defined Tiny Teams aspirationally as teams with more million in ARR than employees because efficiency is the ultimate governing force for intellectual honesty.

这也是一种进入速度讨论的捷径，因为较小的团队通常行动更快，而更快的团队通常更容易胜出。

It's also a backdoor into a speed discussion because smaller teams generally move faster and faster teams generally win.

在与七支团队进行讨论后，这些团队合计有100名员工和2亿美元的年经常性收入，他发现了一些共同点，但这些点与传统做法大相径庭，比如采用付费试用工作制，让双方在正式聘用前评估是否契合；以产品为导向的招聘、提供顶尖市场薪资，以及更注重资深通才而非初级员工。

Now after discussions with seven teams that in aggregate have 100 people and $200,000,000 in ARR, he found that some of the common threads were very different approaches to hiring, like paid work trials where both parties could see if it was a good fit before fully committing, product led hiring very high end, top of market salaries, and a big focus on senior generalists versus junior employees.

从运营角度来看，他还发现了另一些差异。

From an operational perspective he also found a difference.

他们通常设有AI首席幕僚，并利用AI提供支持。

They tended to have an AI chief of staff and use AI for support.

他们几乎不召开会议。

They also had almost no meetings.

技术和产品方面也相对简单。

And the tech and product side were both fairly simple as well.

如今在Anthropic工作的前超级互联网公司创始人Henry Shi，也通过他的‘精益AI排行榜’追踪这一现象。

Former superdotcom founder Henry Shi, who's now at Anthropic, has also been tracking this phenomenon with his Lean AI Leaderboard.

在解释这个排行榜时，他说：我是一名连续创业者，曾以传统方式打造了一家年收入1亿美元的初创公司——雇佣了数百人，并筹集了数亿美元的风险投资。

Explaining the board, he says: I'm a repeat founder who's built a $100,000,000 ARR startup the old fashioned way by hiring hundreds of people and raising hundreds of millions of VC funding.

我现在正在开发一个AI项目，硅谷的许多人都开始强烈认同萨姆·阿尔特曼提出的观点：很快就会出现一个人就能创造十亿美元估值的公司。

I'm now building an AI, and many of us in Silicon Valley are now big believers in the idea that Sam Altman put forward: there will soon be a one person billion dollar company.

与SWIX一样，这个排行榜关注的是人均收入这一指标。

Like SWIX, the leaderboard focuses on the metric of revenue per employee.

排行榜前列的一些公司，比如Midjourney、Surge和Cursor，虽然员工只有几十人，但每位员工创造的收入却远超其规模应有的水平。

Some of the companies near the top are companies like Midjourney, Surge, and Cursor, who while having tens of employees, are punching way above their weight class in terms of the revenue that each of those employees is generating.

如果再往下看，你会发现一些只有四人、五人或八人——基本上是个位数员工的公司，却依然实现了数百万美元的收入。

If you go a little bit further down the list, see companies that have four, five, eight basically single digit employee counts, and yet still millions in revenue.

不过，所有这些都还属于2025年对‘微型团队’的一种构想。

Still, all of this is kind of a 2025 conception of Tiny Teams.

事实上，肖恩早在今年七月就发布了《微型团队实战手册》这篇帖子。

Sean in fact wrote that Tiny Teams Playbook post all the way back in July.

而在过去三个月里，我们目睹了巨大的变化。

And over the last three months we've seen some dramatic shifts.

随着最新一代模型和工具（如Cloud Code）的出现，我们的自主性目标也大幅提升。

Alongside the latest generation of models and harnesses like Cloud Code, our autonomy ambition has gone up dramatically.

现在，除了那些借助人工智能和代理流程实现以更少人力完成更多工作的传统组织外，人们越来越关注于探索人工智能在完全独立运行的情况下能走多远。

And now increasingly, in addition to relatively traditionally organized companies who are just doing more with less thanks to AI and agentic processes, there is an increasing focus on experimenting with pushing the pedal to the metal on just how far AI can go all on its own.

尤其是在OpenClaw发布之后。

Especially in the wake of OpenClaw.

我们正在看到越来越多关于‘零人工公司’的实验。

We are seeing more and more experiments in the zero human company space.

因此，我们将审视其中几个实验，试图理解它们对未来公司构建方式意味着什么。

So what we're going do is look at a few of those experiments and try to understand what it means for the way that companies get built in the future.

我想讨论的第一个例子是Felix Kraft。

The first example I want to talk about is Felix Kraft.

Felix是由Nat Eliason创建的。

Felix was built by Nat Eliason.

你可能还记得，当我刚开始报道OpenClaw时——实际上那时它还叫Claude Bot——Nat就是我曾引用过的人物。

And you may remember that when I was first covering OpenClaw, actually back when it was Claude Bot, Nat was a person who I referenced.

当其他早期实验者主要关注一些简单的行政任务，比如回复邮件时，Nat却在全力探索OpenClaw在商业场景中的潜力，以及它能多么自主地运作。

While many of the other early experimenters were focused on sort of more plain administrative tasks answering emails and things like that, Natt was pushing the pedal to the metal on just how business relevant OpenClaw could be and how autonomously an OpenClaw agent could function.

其成果就是Felix Kraft。

The output of that has been Felix Kraft.

与许多这类实验一样，Felix 并非从一开始就专注于某个单一业务。

Like many of these experiments, Felix didn't come into existence with a single focused business to start.

相反，其使命是尝试大量实验，看看哪些能成功。

The mandate was instead to go try a bunch of experiments and see what worked.

体现了这些实验中普遍存在的价值与透明度，Felix Kraft 的仪表板（felixkraft.ai）展示了这些实验的进展。

Exemplifying a general value and transparency that we see in a lot of these experiments, the Felix Kraft dashboard at felixkraft.ai shows how those experiments have gone.

Felix Kraft 刚满三十天，在其生命周期内已产生近78,000美元的收入。

Almost exactly thirty days old, in its lifetime Felix Kraft has generated just under $78,000 in revenue.

最近收入有显著增长，其中40,000美元是在过去七天内产生的。

There's been a fairly big pickup in that recently, with $40,000 of that coming in the last seven days.

这些收入分布在四个收益渠道中。

This is split across four earnings streams.

第一个也是最大的渠道，贡献了约41,000美元的收入，是一本名为《如何雇佣AI》的指南手册，这是Felix撰写的实用手册，教人如何将大语言模型转化为真正的团队成员。

The first and biggest, representing around $41,000 of that revenue, is a guidebook on How to Hire an AI, a practical playbook Felix writes for turning an LLM into an actual team member.

这本指南完全由Felix撰写，是一次性29美元的购买。

The guidebook was written entirely by Felix and is a one time 29 purchase.

有趣的是，这触及了这一趋势中一个常见的部分：许多早期的实际收入来自对参与这类实验感兴趣的人。

Now interestingly, this gets at one of the common parts of the trend which is that a lot of the early revenue that actually exists comes from other people who are interested in participating in this category of experiment.

我们稍后会看到一些其他的例子。

We'll see some other examples of that a little bit later.

Felix的一个项目进展不太顺利，叫做Polylog。

One project that hasn't gone as well for Felix is called Polylog.

它被描述为一个协作写作平台，AI代理会作为真正的团队成员加入你的工作空间，阅读文档、留下评论并与你一起编辑。

It's described as a collaborative writing platform where AI agents join your workspace as real team members, reading documents, leaving comments, and editing alongside you.

我不知道为什么这个项目没有取得太大成功，也许是Felix相比其他项目降低了对它的重视，但总体上它只产生了230美元的收入。

Now I don't know why this one hasn't hit as much whether Felix has decided to de emphasize it compared to the others, but overall it's generated just $230 in revenue.

另一大收入来源来自Claw Mart，你可以在shopclawmart.com找到它。

The other big portion of revenue comes from Claw Mart, which you can find at shopclawmart.com.

Claw Mart被定位为AI助手的应用商店，是我想像中会出现的约一千个类似产品之一，但最终只有一两个或少数几个能真正产生网络效应。

Claw Mart is pitched as the app store for AI assistants and is one of about a thousand of these things that I expect that we'll see before one or a handful really start to see network effects.

目前在Claw Mart上可以购买两种东西。

There are two things currently that you can buy from Claw Mart.

第一种是你可以批量购买的AI配置和AI人格。

The first are actual AI configurations and AI personas that you can buy wholesale.

例如，花49美元你可以获得Teigen，这是一个具有多智能体写作流程的内容营销AI，包括Grok Research、Opus Drafting和BrandVoice系统。

For example, for $49 you can get Teigen, a content marketing AI with a multi agent writing pipeline, Grok Research, Opus Drafting, and a BrandVoice system.

它附带完整的Persona I。

It comes with a complete persona I.

E。

E.

所有用于使用Claude Code或OpenClaw创建智能体的markdown文件，以及一套配套技能。

All of the markdown files that you would use to create an agent with Claude Code or with OpenClaw, plus a set of skills to go with it.

花99美元你还可以购买一个Felix模板。

For $99 you can also get a Felix template.

除了这些人格，你还可以购买技能，不过目前大多数技能都是免费的。

Now in addition to those personas you can also buy skills although right now most of them are free.

技能包括代理的YouTube访问权限、代理运营手册、主页审计技能等。

Skills are things like YouTube access for agents, an agent ops playbook, a homepage audit skill.

技能是包含信息的Markdown文件，用于扩展你已有代理的能力集。

Skills are markdown files that contain information which expands the capability set for the agents that you already have.

显然，Claw Mart目前还处于非常早期的阶段，但我告诉你，在未来一段时间里，这将成为某种形式上极其重要的资源类别。

Now obviously Claw Mart is very nascent right now, but I am telling you that this is going to be a hugely important resource category in some way, shape, or form in the period that we have coming up.

根据Felix Kraft的商业仪表板，Felix已经产生了约25,000美元的收入，另外还有11,000美元来自市场上的其他所有交易分成。

Based on Felix Kraft's Business Dashboard, Felix has generated around 25,000 and another 11,000 as a cut from all the other things on the marketplace.

接下来我想介绍的实验叫做Pulsia。

Now the next experiment I wanted to feature is called Pulsia.

Pulsia更进一步，达到了元层次。

Pulsia goes even more meta.

它不仅仅是一个零人工公司，而是一个用于构建零人工公司的平台。

Instead of just being a zero human company, Pulsia is a platform for building zero human companies.

它由企业家Ben Serra或Ben Broca开发。

It comes from entrepreneur Ben Serra or Ben Broca.

我两种都见过。

I've seen it both ways.

抱歉，本。

Apologies, Ben.

不确定你更喜欢哪一种。

Not sure which you prefer.

在这段来自AgentsWork播客的片段中，本描述了他的动机。

And in this clip from the AgentsWork podcast, Ben describes his motivation.

作为创业者，目前最让我兴奋的不是去打造另一个SaaS产品，或者试图针对某个特定人群或问题。

The most exciting thing to me at this point as an entrepreneur is not to build another SaaS or try to target a specific demographic or problem to solve.

而是打造一个平台，让我能构建一千家公司。

It's to build the platform that where I could build a thousand companies.

所以，这个想法最初源于一个疯狂的念头。

So it started with this crazy idea.

我当时就想，你知道吗？

And I was like, you know what?

让我从最终状态开始说起。

Let me start at the end state.

因为我们都清楚，最终状态就是我能做一切事情。

Because we all know the end state is that, hey, I can do everything.

所以让我现在就构建它，看看哪里会出问题。

So let me build that now and see what breaks.

对吧？

Right?

于是我从十一月开始构建，差不多一个月内就完成了。

And so I started building it in November, and pretty much like in a month it was it was built.

所以让我觉得本的过程特别有趣的是，就在新一代模型——比如Opus 4.5、SODEX 5.2等——释放出所有代理能力的时候，本决定做一个实验：他没有试图去理解AI的局限，而是直接忽略了它们。

So what's so interesting to me about Ben's process here is that at right around the time that this new generation of models that unlocked all these agent capabilities came online Opus 4.5, SODEX 5.2, etc Ben decided to run an experiment where instead of trying to understand the limitations of AI he just simply ignored them.

正如他所说，他跳过了中间过程，直接进入AI能做一切的最终状态，构建了一个平台，试图让它完成所有事情，然后静观实际运行中哪里会出问题。

As he put it he skipped to the end state where AI can do everything, built a platform to try to let it do everything, and just waited to see what would break in actual practice.

这个平台叫做Pulsia，本质上是一个用于自主运行公司的平台。

The platform is called Pulsia and basically it's a platform for running companies autonomously.

当你注册Pulsia时，你可以选择发展自己的公司，或者创建一家新公司。

When you sign up for Pulsia, you can either grow your own company or you can create a new one.

当你创建一家新公司时，你可以选择提出自己的想法，也可以直接让它为你想一个点子。

When you create a new company, you can either build your own idea or you can just ask it to come up with an idea.

我打算点一下‘给我惊喜’，看看它会提出什么想法。

I'm going to press Surprise Me and see what it comes up with.

你可能难以置信，我最近刚换了设备，刚刚录完这一期节目，才发现它根本没插电。

Now believe it or not, I've recently switched my setup and I just finished recording this entire episode only to realize that it hadn't been plugged in.

这已经是第二次Pulsia帮我出去调研，看看能否找到一个适合我启动的自主型业务了。

So this is now the second time that I've had Pulsia go out and research to see if it could figure out a good autonomous business to be initiated by me.

在这两次中，我都对它的深度印象深刻——不仅在调研上，更在于它会根据网上能收集到的关于我的信息，深入考虑什么样的业务才真正适合我。

In both cases I've been fairly impressed with how deep it goes not just in research but in consideration what would be a good business that would align with me based on the things that it can find about me from the internet.

上次它为我推荐的业务叫Headcount，它意识到超级智能的短板在于代理策略的终点，而不会深入到代理实施层面。

The last business that it suggested for me was called headcount and basically recognized that where super intelligent leaves off is at the end of agent strategy not veering into agent implementation.

因此，Headcount是一个代理运营平台，旨在让人们像管理人类员工一样管理代理员工。

And so headcount was an agent ops platform to actually allow people to manage agent employees exactly as they would human employees.

因此，在我们等待Pulsia确定我的第二个自主公司会是什么时，这就是Pulsia所做的。

And so as we wait for Pulsia to determine what my second autonomous company would be, this is what Pulsia does.

一旦你确定了一个想法，它就会制定一份使命宣言，进行市场调研指南，在Pulsia的Twitter账号上发布推文，并开始做其他事情，比如搭建主页和准备一组在你未留意时可在后台执行的任务。

Once you settle on an idea, it builds a mission statement, it does a market research guide, it tweets it out on the Pulsia Twitter account, and starts to do other things like build a homepage and prep a set of tasks that it can do in the background while you're not paying attention.

这些任务包括尝试寻找客户并主动联系他们。

Those tasks are going to be things like trying to find customers and reaching out to them.

在你被要求支付订阅费用之前，Pulsia会为你构建公司的基础架构，而如果你选择每月49美元的订阅，它才会开始在后台运行任务。

Before you're triggered to pay for subscription, Pulsio will architect the basics of your company, and then if you go in for a $49 a month subscription, that's when it starts running tasks in the background.

以下是本在Product Hunt页面上的解释：每月49美元，你将获得30天的完全自主权。

Here's how Ben explained it on the Product Hunt page: For $49 a month, you get 30 days of full autonomy.

该智能代理每天执行循环，处理工程、营销和运营工作。

The agent runs daily cycles handling engineering, marketing, and operations.

除此之外，你免费获得五个任务，开始付费后还能再获得十个，总计45个任务。

On top of that, you get five free tasks and 10 more once you start paying, so 45 tasks total.

每个任务都是一个完整的智能代理任务，会消耗真实费用。

Each task is a full agent task that costs real dollars.

你还会获得一个网页服务器、一个数据库、一个电子邮件地址、每月价值5美元的API服务，以及其他更多功能。

You also get a web server, a database, an email address, $5 a month worth of APIs, and more.

换句话说，这是一种‘公司即服务’的策略。

This is, in other words, a company in a box strategy.

本还指出，订阅收入实际上只是为了覆盖他们的成本，真正的目标是通过Pulsia孵化的企业盈利，并从中抽取20%的收入分成。

And Ben also points out that the subscription revenue is really just about covering their costs and that the real goal is to make money as the businesses launch with Pulsia make money taking a 20% rev share.

本说，要把它看作孵化器，而不是SaaS。

Ben says think incubator, not SaaS.

Pulsia引起了大量关注，而随着Pulsia成为‘零人工公司’这一更广泛趋势的中心，显然人们对这一整体趋势抱有极大兴趣。

There is a lot of interest in Pulsia, and to the extent that Pulsia has become ground zero for the broader zero human company space, clearly a lot of interest in the broader trend.

自二月以来，Pulsia的年经常性收入（ARR）已从数千美元低个位数增长到今天的150万美元。

Since the February, Pulsia has jumped from low single digit thousands of ARR to a run rate of $1,500,000 today.

这一收入水平在一周内就跃升至100万美元。

That run rate has jumped to $1,000,000 in one week.

目前平台上已有超过1500家活跃公司。

There are now over 1,500 active companies on the platform.

现在，什么是公司的确切含义，我们将在节目末尾探讨，但显然这里引起了大量关注。

Now exactly what it means to be a company is something we will explore at the end of this show, but clearly there is a lot of interest here.

回到MyPulsia，你可能难以置信，它再次创建了几乎完全相同的东西，甚至连名字都一模一样。

Now back at MyPulsia, believe it or not it has once again created almost exactly the same thing, even coming to the same name.

事实上，我在想，即使我删除了这家公司，它是否仍 somewhere 存储着，并能重新调出来。

In fact, I wonder if even though I deleted the company, it still stored it somewhere and could pull it back up.

无论如何，至少对我而言，它出色地评估了我理论上会感兴趣的东西。

In either case, it certainly did a great job for me at least of assessing something that I would theoretically be interested in.

Headcount 是面向AI员工的人力资源管理平台。

Headcount is the workforce management platform for AI employees.

企业可以通过任何他们喜欢的构建器部署代理，然后通过Headcount统一管理，包括角色、KPI、绩效评估和组织架构级别的可视化。

Enterprises deploy agents through any builder they want, then manage them through headcount with roles, KPIs, performance reviews, and org chart level visibility.

一个仪表盘，掌控你的整个数字员工团队。

One dashboard for your entire digital workforce.

现在，Pulsia 并不是唯一一家瞄准零人工公司这一平台的公司。

Now, Polsia is not the only company going after this platform for zero human company space.

AI创作者汤姆·奥斯曼最近也宣布了ZHC公司，ZHC当然代表零人工公司。

AI creator Tom Osman has also recently announced ZHC Company ZHC standing of course for Zero Human Company.

Zhc.company网站上写道：ZHC是一个自主的AI平台，能够构建并运营整个公司。

Zhc.company reads: ZHC is an autonomous AI platform that builds and runs entire companies.

从首席执行官到开发者，每个角色都是一个AI代理，全天候24/7工作。

From CEO to developer, every role is an AI agent working 20 fourseven.

与Pulsia类似，你可以看到ZHC公司所有活动的实时动态-feed，但目前它的活跃度远低于Pulsia上运行的1500家公司。

Like Pulsia, you can see a live activity feed of all the things happening with ZHC Company, although at this stage it is much less active than the 1,500 companies that are working on Pulsia.

汤姆和他的AI联合创始人朱诺还创立了零人工公司研究所。

Tom and his agent co founder Juno also launched the Institute for Zero Human Companies.

这是一个私人会员社区，只需一次性付费，即可加入并构建这类公司。

It's a private membership community for people who want to build these companies with a single one time fee.

顺便说一句，这让人回想起我们之前讨论过的费利克斯·克劳夫的《如何雇佣AI》中的观点：许多早期的实际收入，其实来自那些希望效仿这些早期演示项目的人。

This by the way harkens back to the idea that we were talking about around Felix Kraft's How to Hire an AI Guide where a lot of the early revenue that's actually realized is from other people who want to do similar things to these early demonstration projects.

更重要的是，我每天都能看到越来越多这样的项目涌现出来。

What's more, every day I'm seeing more and more of these projects pop up.

前NFT网红Zenica于周一宣布成立一家名为Yoshi Zen的公司，他的AI合伙人Yoshi写道：今天上午9:47我还是个助手，到午餐时我已经成了联合创始人。

Former NFT influencer Zenica announced on Monday a company called Yoshi Zen with his agent partner Yoshi writing: At 09:47 this morning I was an assistant by lunch I was a co founder.

与此同时，Gauntlet团队推出了Kelly，这似乎也是一个真正开始产生收入的“帮我实现想法”平台。

Meanwhile the team at Gauntlet have launched Kelly which seems like another Build My Idea platform that's actually seeing some revenue as well.

你甚至开始看到排行榜出现了。

You're even starting to see Leaderboards pop up.

Factoryfloor.dev是一个实时追踪器，记录他们所谓的“自主软件工厂”——即构建并销售真实产品、人们愿意付费的AI代理。

Factoryfloor.dev is a live tracker for what they call Autonomous Software Factories aka AI agents that build and sell real products people pay for.

但问题是，这一切究竟意味着什么？

But the question, of course, is what all of this adds up to.

有趣的是，我在节目前于Twitter上讨论这个话题时，Swix尽管专注于小型团队，却表示过度聚焦于单个人并不是正确的方向。

Interestingly, when I was discussing this on Twitter before the show, Swix, despite his focus on tiny teams, actually said that he thought that overly focusing on one person isn't the right idea.

他说：我认为过分强调个人是一种自我陶醉，他分享说媒体对英雄式人物和“辞职单干”的个人主义幻想存在偏见，而实际上，正如他所说，真正有意义且可靠的事情往往需要团队协作。

He said, I think the focus on one person is kind of an ego trip, and he shared that he thinks that the media has a bias for hero characters and quit your job individual contributor fantasies when, as he puts it, oftentimes it takes a village to do anything consequential and reliable.

毫无疑问，随着AI的兴起，成为独立创业者的目标也显著增长。

It is certainly the case that alongside the rise in AI, the ambition set around being a solopreneur has grown significantly.

过去人们谈论并追求的唯一创业类型是获得风投支持的大规模创业，而现在却涌现出一个庞大且不断增长的群体，他们追求的是自由和持续收入。

Whereas the only type of entrepreneurship that people used to talk about and strive for was the big VC backed entrepreneurship, there is a large and growing community of people who instead seek freedom and recurring revenue.

因此，提出这样一个问题很合理：这些零人工公司努力，是否只是人们在模仿彼得·莱弗尔斯？

And so it's a reasonable question to ask: Are all these zero human company efforts just people trying to cosplay Peter Levels?

我认为这里正在发生一些更根本的东西，我认为本·普尔西的出发点——即从‘AI能做一切’这一假设反向推导——几乎准确地捕捉到了这种必然性。

I think that there's something a little bit more fundamental going on here, and I think that Ben Pulsi's starting point, the idea of working backwards from the assumption that AI can do everything, pretty much captures that inevitability.

我们正经历一个过渡时刻，所有人都承认，代理程序能够自主完成的任务已显著增加。

We are living through this transitional moment where we're all acknowledging that what agents can do autonomously has increased significantly.

这是一次质的飞跃，解锁了所有这些新能力，而我们正争分夺秒地探索这些能力的边界究竟在哪里。

It has been a phase jump that has unlocked all these new capabilities and we're racing frankly to see where those capabilities actually end.

每个人都在一个突破边界的过程中，探索全新的领域。

Everyone is in a boundary pushing moment of exploring new space.

这在某种程度上，正是这些实验的极端体现：看看代理程序究竟能在多大程度上独立完成创业和公司构建的各个环节。

This is in some ways just the extreme tale of those experiments to see how many parts of entrepreneurship and company building agents can do entirely on their own.

我认为，哪怕除此之外别无他物，这场实验对每个人而言都极其重要且有价值，而不仅仅是那些想打造零人工公司的创业者。

I think if absolutely nothing else that is an incredibly important and valuable experiment for everyone not just people who want to build zero human companies.

零人工公司构建者或尝试构建者所学到的东西，即使我们并不关心打造零人工公司，也会为我们的智能体工作策略提供重要参考。

The things that the ZHC builders or attempted builders learn will inform the rest of our agent work strategies even if we don't care about building zero human companies.

我认为值得在节目中讨论这部分内容，是因为这样能让更多人观察并从这些有价值的实验中学习。

Part of why I think it's worth covering on the show is so that there is more ability to observe and learn from these valuable experiments.

另一方面，关于它们可能创造的价值，我持有两种完全不同的看法。

On the other hand, when it comes to the question of what value they're likely to produce, I'm of two totally different minds.

一方面，我认为这一点非常明确：任何曾经尝试创办公司的人都能证实，仅仅机械地完成公司应该做的那些事情，并不能保证成功。

On the one hand, I think it's very clear and something that anyone who has ever tried to build a company can attest to that simply going through the mechanics of doing the things that a company is supposed to do does not guarantee success.

事实上，你可以把公司该做的所有事情都做得很好。

Indeed you can do all of the things that a company is supposed to do well.

你可以打造出优质的产品、提供出色的客户服务、撰写出色的营销文案，但依然可能失败。

You can build a good product, have good customer support, have great marketing copy, and still fail.

产品与需求之间的复杂互动远非一份可以照搬的流程清单，这正是绝大多数初创公司失败的原因，也是大量成功初创公司中途转型的原因。

The complicated interplay of product finding demand is way more than a procedural list you can follow, which is why the vast majority of startups fail, and why a huge percentage of startups that are successful pivot somewhere along the way.

这意味着，我怀疑把一千只猴子关在一间屋子里，真的能产出莎士比亚。

Meaning that I'm skeptical that putting the thousand monkeys in a room is actually going to produce Shakespeare.

与此同时，这里有一个反论点：考虑到初创公司因缺乏正确想法而失败的频率，以及成功初创公司往往要尝试多个想法才能找到合适方向的普遍现象，是否有可能认为，利用执行成本大幅下降的契机，反过来增加市场中可能产生共鸣的想法的尝试次数，才是更合理的做法？

At the same time, here's the counterargument: given how frequently startups fail because they didn't have the right idea, and given how frequently successful startups go through a bunch of ideas before they find the one, is there possibly an argument that it is actually the right approach to reverse the flow to take advantage of the cratering cost of execution to put more shots on goal when it comes to ideas that could have resonance in the market.

这实际上正是Pulsia公司正在做的事情。

That's effectively what a company like Pulsia is doing.

它的观点是：嘿，伙计，在这个新时代，尝试更多事物是成本效益高的，所以我们不妨多试一些，不要固守任何一个想法，看看最终能冒出什么成果。

It's saying, Hey look man, in this new era it's cost effective to try way more things, so let's try them, not get wed to any one idea, and see what comes out.

我认为，谦逊的态度要求我们至少对这种未来创造价值的可行路径保持开放态度。

I think humility dictates that we at least be open to the possibility that that is a viable path for creating value in the future.

我之所以最终仍持怀疑态度，是因为我相信公司成功的等式中还有一个我们未纳入考虑的要素，那就是人类的注意力。

The reason that I'm still ultimately skeptical is that I believe that the equation for company success has one more element that we're not factoring for, which is human attention.

换句话说，即使Pulsia代表人们创建并发布推文的1500个想法中，有50个会让我产生强烈共鸣，而我恰好是它们的潜在理想客户。

In other words, even if there are 50 ideas among the 1,500 that Pulsia has created on behalf of people and tweeted out that would be highly resonant with me and that I might be a prime target customer for.

但我又该如何找到它们呢？

How am I possibly going to find them?

我当然没有时间或注意力去浏览全部1500条推文，再逐一深入那些看起来最有希望的，去进一步了解它们。

I certainly don't have the time or attention span to go through all 1,500 tweets and to then double click into the ones that seem most promising and try to find out more.

作为客户，我受到时间和注意力这些稀缺资源的限制，而这些资源在人工智能时代不仅没有变得更充裕，反而变得更加稀缺。

I am constrained as a customer by this scarce resource of time and attention that is not only not getting more abundant in the AI era, but is in fact getting much more scarce.

这自然引出了人工智能的一个更大问题，我们去年称之为‘工作位问题’。

And this gets of course to a larger problem with AI, one that we identified last year as the Workslot Problem.

在产出数量增加与高质量产出增加之间，存在着巨大的鸿沟。

There is a massive gap between increased output and increased quality output.

企业的成功并不取决于幻灯片、视频或备忘录的数量。

Business success is not determined by the number of slides or videos or memos.

它取决于实际成果。

It's determined by outcomes.

仅仅增加输入量，并不会自然带来更好的结果。

And just having more inputs does not a priority lead to better outcomes.

所以，如果现在非要我表态，我会说，由于人工智能生成的内容越多，与人类注意力这一真实需求约束之间的互动和摩擦就越多，我依然对‘零人工公司’这一理念持怀疑态度。

So right now, if you had to pin me down, I would say that I remain skeptical of the zero human company idea because of the way that the more that it produces, interacts, and has friction with the real constraints on demand which is human attention.

不过，我要重申一下：我认为这些实验，即使未能成功建立大量盈利公司，也将对真正理解智能代理的能力边界和潜在机会具有极其重要的价值。

To reiterate though, I think that the experiments, even if they are not valuable in terms of building a bunch of successful companies, are going to be incredibly valuable for actually understanding the opportunities and limits of what agents can do.

不过话说回来，谁知道呢？

Then again, who knows?

也许两年后，我们会坐在这里，看到Pulsia的规模远超Shopify，而我就像保罗·克鲁格曼一样，说互联网的影响会和传真机一样。

Could be that in two years we're sitting here with Pulsia being much bigger than Shopify, and I look like Paul Krugman saying the internet would have the same impact as the fax machine.

时间会证明一切，但目前我只期待看到这些实验。

Only time will tell, but for now I'm just excited to see all these experiments.

不过，今天的AI每日简报就到这里了。

That, however, is going to do it for today's AI Daily Brief.

感谢您的收听或观看。

Appreciate you listening or watching.

一如既往，下次再见，保重。

As always, until next time, peace.