面向所有人的 AI 智能体平台

AI代理的承诺可能终于要变成现实了。

The promise of AI agents might finally be becoming a reality.

今年年初，很明显2026年将是代理之年。

At the beginning of this year, it was clear that 2026 was going to be the year of agents.

模型和工具链的进步相结合，意味着在年初左右，很明显已经达到了一个关键的转折点，人们在一月份就开始争相探索像Cloud Code和OpenAI的Codex这样的工具所带来的新功能。

The combination of the advancement of models plus harnesses meant that around the turn of this year, it was clear that some critical inflection point had been reached, and people came into January racing to uncover all of the new capabilities that tools like Cloud Code and OpenAI's Codex made available to them.

然而，代理热潮真正爆发是在OpenClaw推出的时候。

When the agent excitement really popped off, however, was with the introduction of OpenClaw.

有了OpenClaw，人们可以组建整个代理团队，包括幕僚长、研究员、写手，你能想到的任何角色，让它们真正协调互动，完成你大部分的工作，而且全部通过像Telegram或WhatsApp这样的简单聊天界面来协调。

With OpenClaw, people were able to spin up entire teams of agents, chief of staffs, researchers, writers, anything you could imagine and have them actually coordinate and interact with one another, doing big chunks of your work, and coordinated all through easy chat interfaces like Telegram or WhatsApp.

当然，问题在于，这样做在技术上极其复杂和困难。

The problem was, of course, that it was extremely technically complex and difficult to do so.

我们甚至推出了整个课程ClawCamp，就是为了帮助人们解决这个问题。

We even dropped an entire course ClawCamp just to help people figure that out.

而且在整个一年中，已经有一些尝试，让这类代理工作的界面更简单，以便更广泛地采用。

And throughout the year, there have been some attempts to make those sort of interfaces for agentic work more straightforward for broader adoption.

但到目前为止，还没有一个真正达到目标。

Yet none of them have really hit the mark.

有些人认为，随着本周Grok Bot的推出，这一切都改变了。

Some think that with this week's introduction of Grok Bot, that has all changed.

Grok Bot允许用户为不同任务创建多个代理。

Grok Bot allows users to spin up multiple agents for different tasks.

你可以给它们访问所需系统的权限，它们会在你观看或做其他事情时去完成工作。

You can give them access to whatever systems they need, and they go off and do work while you watch or while you're doing other things.

这是Cursor和SpaceX AI合作的首批产品之一，两家公司甚至表示，这些机器人会随着时间推移不断学习和改进。

It's one of the first products from the combined efforts of Cursor and SpaceX AI, and the companies even say that the bots will learn and get better over time.

那么，这是否就是我们一直在等待的代理平台，一个能将代理工作能力带给更广泛人群的平台？

So is this the agent platform that we've been waiting for, the agent platform that can bring the capabilities of agentic working to a much wider array of people?

让我们一探究竟。

Let's find out.

《AI每日简报》是一个每日播客和视频节目，涵盖人工智能领域最重要的新闻和讨论。

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

好的，朋友们。

Alright, friends.

在开始之前，先快速说几个通知。

Quick announcements before we dive in.

首先，感谢今天的赞助商：KPMG、Blitzy、Section 和 Hyperagent。

First of all, thank you to today's sponsors, KPMG, Blitzy, Section, and Hyperagent.

想收听无广告版本，可以去 patreon.com/aideallybrief，或者到 Apple Podcasts 上订阅。

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

如果想了解赞助方式，请发送邮件到 sponsors@aideallybrief.ai，或者直接访问官网 aideallybrief.ai，点击赞助板块。

To learn more about sponsoring the show, send us a note at sponsors@aideallybrief.ai, or you can actually visit the website at aideallybrief.ai and click around to the sponsor section.

在 aollby 上，你还能找到每集的完整配套内容。

While you are on aideallybrief.ai, you can also find the complete companion to each episode.

每一集都被拆分成可分享的小片段，也就是说，如果你想分享其中的某句话或某个数字，那里很可能已经为你准备好了。

Every single episode is broken down into shareable chunks, meaning that if there's just one quote or some number that you want to share with a colleague, it's probably there waiting for you.

你还可以从那里订阅我们的电子通讯，所以再说一遍：aideallybrief.ai。

You can also sign up to our newsletter from there, so again, aideallybrief.ai.

好了，不过我们还是先快速浏览一下今天的头条新闻，然后再聊聊这个令人兴奋的Grok Bot新版本。

With that, though, let's first cruise through the headlines and then talk about this exciting new release of Grok Bot.

今天我们一开始就讲一个备受关注的话题，它的热度高到感觉这周晚些时候可能会成为大鱼。

We kick off today with a story that has gotten enough attention that it honestly might make it into a main later this week.

Anthropic因为给AI生成的文本加入水印而引发了巨大的争议。

Anthropic has stirred up huge amounts of controversy by introducing watermarks for AI generated text.

从本月开始，Anthropic所有新模型都会在生成的内容里嵌入隐形水印。

As of this month, all new Anthropic models will now include invisible watermarks in their generations.

Anthropic模型将能够检测这些水印，帮助识别AI生成的内容，同时也会支持第三方AI检测工具来帮忙。

Anthropic models will be able to detect the watermarks to help identify AI generated content, and Anthropic will support third party AI detectors to do the same.

这项新政策是Anthropic对欧盟关于AI生成内容行为准则的承诺的一部分，而这个行为准则是欧盟《人工智能法案》的一部分。

This new policy is part of Anthropic's commitment to the EU Code of Practice on AI generated Content, part of the EU AI Act.

争议的关键在于，Anthropic不像其他公司那样，只给图片或视频加水印来对抗深度造假的危害。

Now, the controversial part is that Anthropic is not just including watermarks in images or videos to combat the harms of deepfakes as other companies have done.

他们反而是在所有生成的文本中都嵌入了水印。

Instead, they're embedding the watermarks in all generated text.

重要的是，它不在元数据里，而是以某种方式嵌在文本本身。

And importantly, it is not in the metadata, but instead, the watermark is somehow included in the text itself.

Anthropic声称：你看不到它，而且它不会改变Claude回复的含义、质量或可读性。

Anthropic claims: You won't see it, and it doesn't change the meaning, quality, or readability of Claude's response.

由于水印是文本的一部分，当文本被复制粘贴到别处时，水印也会跟着走，而且即使经过一些编辑，它可能仍然存在。

Because the watermark is part of the text, it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing.

这项政策适用于所有地区的所有模型，所以即使你不在欧盟，你的Claude输出也照样会被打上水印。

The policy is being applied to all models in all regions, so even if you're not in the EU, your clawed outputs will still be watermarked.

人们很怀疑Anthropic能不能在不降低输出质量的情况下做到这一点。

Folks were very skeptical that Anthropic can achieve this without degrading the quality of the outputs.

奥尔夫写道，水印基本上就是模型的统计特征或词语选择。

Writes Orf, The watermark will basically be the statistical signature or word choice of the models.

Anthropic说这不会影响质量，但这显然不太可能，因为他们要限制模型的采样，偏向某些词语，这可能会让模型在后续输出中变得更不具创造力和创新性。

Anthropic claims this won't affect quality, but this seems non obvious given that they'll be constraining the model sampling and biasing token choice, and potentially making the model less creative and inventive downstream.

欧盟又搞这一套了。

The EU does it again.

现在，光靠几条推文很难说清楚大家对此有多不满。

Now, it would be hard to describe in just a couple of tweets how upset people are about this.

意识到这不仅适用于普通写作，也适用于代码编写，开发者尼克·诺博斯写道：Claude在我的代码库里添加隐形水印？

Recognizing that this is not just for regular word writing but also for code writing, developer Nick Nobos wrote: Claude adding invisible watermarks inside my codebase?

完全是胡扯。

Total BS.

这开创了一个恶劣的先例。

Diabolical precedent to be setting.

我们到底在搞什么鬼？

What TF are we even doing here?

西蒙·史密斯写道：关于Anthropic的水印器，有个问题：如果我让Claude做研究，而研究需要引用像法律文件这样的内容，引文会不会被微妙地改动以嵌入水印？

Simon Smith writes: Question about Anthropic's watermarker: If I task Claude with doing research, and that research requires quoting something like a legal document, will the quote be subtly changed from the source to introduce the watermark?

因为那太可怕了。

Because that's terrifying.

如果不是这样，我们又怎么能确定呢？

If not, how can we be sure?

虽然安德鲁·柯伦指出，Gemini从2024年起就在做类似的事情，但这并没有让人们感觉好多少。

And while Andrew Curran points out that Gemini has been doing something like this since 2024, that didn't make people feel much better.

这个问题本身值得深入讨论，不只是上头条那么简单，但今天就先介绍到这里，我相信以后还会有更多相关消息。

Now, this is a debate that's worthy of more than just the headlines, but for now, that's the primer on the topic and I am sure it is not the last that we will be hearing about this.

说到谷歌，他们宣布Gemini应用的用户数已经达到了十亿。

Speaking of Google, the company has announced that they have reached a billion users for the Gemini app.

CEO桑达尔·皮查伊在周二宣布了这个里程碑，他写道：现在每月有超过十亿人在使用Gemini应用，激发新想法、完成任务。

CEO Sundar Pichai announced the milestone on Tuesday, posting: 1,000,000,000 plus people are now using the Gemini app every month to spark new ideas and get things done.

这是我们有史以来增长最快的产品，也是第十四个达到十亿用户的产品。

It's our fastest growing product ever and our fourteenth to hit the 1,000,000,000 user mark.

不过，考虑到Gemini已经集成到了搜索、YouTube和其他平台里，有些人对这个数字持怀疑态度，但The Verge确认了这个周的十亿用户里程碑确实专门针对Gemini应用。

Now, some are a bit skeptical of this, given that Gemini is integrated into Search, YouTube, and other platforms, but The Verge confirmed that this week's billion user milestone applies specifically to the Gemini app.

谷歌已经开始在Android手机上预装Gemini，但用户只有真正使用这个应用才会计入这个数字。

Google has begun pre installing Gemini onto Android handsets, but those users still need to engage with the app to be included in these numbers.

谷歌还分享了一些有趣的数据，说明用户是如何使用他们的聊天机器人的。

Google also shared some interesting details about how users are interacting with their chatbot.

63%的用户都在用语音界面，这说明大家有多想摆脱键盘。

63% of users are using the voice interface, demonstrating how eager people are to ditch the keyboard.

Gemini的图像模型也保持了很大的使用量，用户每天生成1.5亿张图片。

Gemini has also retained massive volume for their image model, with users generating 150,000,000 images per day.

真正有意思的是，这反映了模型领域的现状。

What's really interesting is what this says about the state of models.

Olga Zirkorn指出，Google目前没有一款AI模型进入前十，这说明在消费市场获胜的关键不是产品，而是分发渠道。

Olga Zirkorn points out that Google doesn't have an AI model in the top 10 right now suggesting this means what wins in consumer markets isn't the product, it's distribution.

但你也得想想，那十亿用户中有很多人根本不知道自己错过了什么，因为他们所有的AI使用都困在一个至少落后前沿六个月的模型里。

But you also have to think that a lot of those billion users have no idea what they're missing, given that all of their AI use is stuck in a model that is at this point at least six months behind the frontier.

现在，我们继续聊聊产品方面。Manus在完成与Meta的分拆后，重新成为一家独立公司。

Now staying on the product side for a minute Manus is returning as an independent company after finalizing their split with Meta.

Manus在去年12月被Meta以20亿美元收购，但随后几个月，中国官方对该交易进行了审查，并在4月下令将其撤销。

Manus was acquired by Meta for $2,000,000,000 in December, but in the following months, Chinese officials scrutinized the deal and in April ordered it to be unwound.

国家安全方面的担忧似乎是，这笔巨额交易会鼓励更多中国初创企业和人才寻求在西方退出。

The national security concern, it seemed, was that the big payday would encourage more Chinese startups and talent to seek an exit in the West.

在这期间，大家一直在讨论马努斯要怎么筹钱还给Meta，因为那些钱都已经分给投资者了。

In the interim, there has been swirling discussion on how Manus will raise the money to repay Meta, given that the funds had already been distributed to investors.

不过到了周二，马努斯宣布他们很快就会重新作为一家独立公司回归，继续服务他们的数百万用户。

On Tuesday, however, Manus announced that they'll soon return as an independent company serving their millions of users.

但是，要完成和Meta的分离，他们需要删除12月29日之后生成的一些用户数据。

However, to complete their separation from Meta, they need to delete some user data generated after December 29.

他们建议所有用户在8月23日过渡日期之前备份数据，并表示独立系统会在8月25日上线。

They advised all users to back up their data ahead of the transition date on August 23, and said that independent systems will be online from August 25.

现在关键的问题是，这会不会是马努斯的一个全新开始，让他们回到智能体竞赛的赛道上来。

Now the big question is whether this is a fresh start for Manus that will let them get back in the agent race.

他们在X上发了一篇文章，写道：展望未来，我们对前景感到无比兴奋。

In a post on X, they wrote: As we look ahead, we couldn't be more excited by the future.

我们正在准备一系列新功能，将再次推动通用AI智能体可能性的边界。

We're preparing a series of new features that will push the boundaries of what's possible for general AI agents once again.

说实话，虽然持怀疑态度是合理的，但你得想想，他们在背水一战、保持独立的情况下走的创新之路，比起在Meta那种巨头内部能看到的要多得多。

And frankly, while it's reasonable to be skeptical, you gotta think that the innovation that they will pursue now as their backs are against the wall, and as they remain independent, is a lot more than what we would have seen from inside the behemoth that is meta.

正如Peter Corbett所指出的：回到零年经常性收入，重新开始，这将会成为一个有趣的案例研究。

As Peter Corbett points out: Going back to $0 ARR and starting again is going to make for an interesting case study.

现在，说到AI行业的热门领域，Open Router的100亿美元估值显然在路由器领域引发了一场竞购战。

Now, on the topic of hot segments of the AI industry Open Router's $10,000,000,000 price tag has apparently triggered a bidding war across the router segment.

据The Information报道，对token路由器收购的兴趣异常高涨，多家大型软件公司希望将这种基础设施纳入自己的技术栈。

The Information reports that interest in token router acquisitions is off the charts, with multiple large software companies looking to add the infrastructure to their stack.

据报道，Snowflake、Base ten、Cloudflare和Vercel都在积极寻找收购目标，而且这种兴趣已经延伸到了该领域一些非常小的初创公司。

Snowflake is reportedly in the market alongside Base ten, Cloudflare, and Vercel, and the interest is reaching some very small startups in the space.

Requestly的CEO Thibault Jigu表示，他们这家只有五人的初创公司已经收到了25家不同公司的投资、收购或合作意向。

Thibault Jigu, the CEO of Requestly, said that his five person startup has fielded interest from 25 different companies looking to invest, acquire, or partner with Requestly.

Myriad Ventures Partners的Dean Mai认为现在是卖方市场，他评论说：对于数据基础设施公司和超大规模云服务商来说，现在不考虑收购是不明智的。

Dean Mai of Myriad Ventures Partners, which has a small token router play in his portfolio, thinks it's a seller's market, commenting: It would be unwise for data infrastructure players and hyperscalers not to entertain acquisitions right now.

token路由器公司Concentrate AI的联合创始人Ari Jacoby表示，本月已有七家公司与他接洽，他评论说：过去两周我们异常受欢迎，这完全超出了我的想象。

Ari Jacoby, the co founder of token router Concentrate AI, said that he's been approached by seven companies so far this month, commenting: We've been unbelievably popular for the past two weeks in a way that I could have never imagined before.

换句话说，目前看来，大型公司和行业巨头已经决定，即使他们能自己开发这种功能，速度的需求压倒一切，现在是时候收购了。

In other words, it seems at this moment that the large companies and incumbents have decided that even if they could build this sort of functionality, the need for speed trumps all, and it is time to buy.

再聊聊市场方面，还有个故事简直能单独当个大新闻。

Moving over to markets for a moment, another story which could easily be an entire main.

英伟达正在组建一个5000亿美元的平台，用来改善数据中心的融资环境。

Nvidia is putting together a $500,000,000,000 platform to help improve the financing landscape for data centers.

这个平台周一宣布成立，将由Apollo、贝莱德和黑石等投资银行和私人信贷巨头提供资金支持。

Announced on Monday, the platform will provide financing from investment banking and private credit giants including Apollo, BlackRock, and Blackstone.

这些机构将向新兴云服务商提供信贷，助力数据中心的大规模建设。

These firms will provide credit to neo clouds to help fuel the data center build out.

具体细节还不太清楚，但看起来这个方案可能会把多家贷款方聚在一起，来统一数据中心债务的处理方式。

The details of the arrangement aren't entirely clear, but it appears the approach could bring together multiple lenders to standardize data center debt.

另外看来英伟达的GPU也会被接受作为抵押品，而且收益分成可能也是方案的一部分。

It also appears that Nvidia GPUs will be accepted as collateral, and revenue sharing could be part of the arrangement.

在一份新闻稿里，黄仁勋说：英伟达达到了一个重要的里程碑。

In a press release, Jensen Huang said: Nvidia has reached an important milestone.

我们是从制造芯片起步的。

We began by building chips.

今天，我们在帮助创造一种新的、可投资的生产性基础设施。在人工智能领域，算力就是收入。

Today, we are helping create a new class of productive investable infrastructure In AI, compute is revenue.

英伟达的算力特别适合这个角色。

Nvidia compute is uniquely suited for this role.

它被广泛采用，灵活适用于各种模型和工作负载，可互换，并且可以在客户和运营商之间转移。

It is broadly adopted, flexible across models and workloads, fungible, and transferable across customers and operators.

这些融资平台将帮助客户大规模获取稀缺的算力，并建设人工智能工厂，这些工厂将在人工智能时代为每个行业和国家提供动力。

These financing platforms will help customers access scarce compute at scale, and build the AI factories that will power every industry and country in the age of AI.

在接受CNBC采访时，黄仁勋将这一举措描述为人工智能融资的新范式，他说：“这确实是第一次，技术芯片成为一种可投资的资产类别。”

In a CNBC interview, Huang presented this as a new paradigm for AI financing, saying: This is really the first time that technology chips have become an investable asset class.

这些现在都是能产生收入的资产。

These are revenue generating assets now.

它们是生产性的。

They're productive.

它们寿命长。

They're long lived.

它们可以互换。

They're fungible.

它们很灵活。

They're flexible.

当然了，对于那些担心循环融资的看空者来说，这简直是往他们的担忧火上浇油。

Now of course, for the bears who are worried about circular funding, this is just pouring absolute gasoline on the fires of their concerns.

但对于其他更中立的市场参与者来说，这一举措似乎正在带来回报。

But for other more neutral market actors, the move seems to be paying dividends.

周二，英伟达的信用利差收窄，表明违约风险降低，其债券也出现上涨，从而降低了下一轮借款的隐含利率。

Tuesday saw Nvidia's credit spreads close implying a lower risk of default, and their bonds also rallied, reducing the implied interest rate for the next round of borrowing.

所以至少就英伟达本身而言，市场基本上将这一新工具解读为英伟达将亏损风险分散给了多方。

So at least when it comes to Nvidia themself, essentially the market interpreted the new vehicle as Nvidia spreading the risk of losses across multiple other parties.

如果出现AI放缓，投资者不再期望英伟达会因为收入下降和数据中心坏账而遭受双重打击。

If we see an AI slowdown, investors no longer expect Nvidia to take a double hit from reduced revenue and bad debt from data centers.

Coherence Credit Strategy的首席投资官Sal Naro表示：没人知道这五千亿美元的潜在融资意味着什么。

Sal Naro, the CIO at Coherence Credit Strategy, said: Nobody knew what the $500,000,000,000 potential financing meant.

现在大家觉得他们把所有相关方都拉进来了，他们的风险敞口并没有投资者最初担心的那么严重。

Today you have an idea that they re getting everybody involved and that their exposure isn t as serious as investors originally feared.

最后，今天要关注华盛顿的消息。参议员伯尼·桑德斯正式加入了暂停运动，并呼吁山姆·奥特曼、达里奥·阿莫德和马克·扎克伯格也这样做。

Lastly today, keeping track of things in Washington, Senator Bernie Sanders has officially joined the Pause Movement and is calling on Sam Altman, Dario Amade, and Mark Zuckerberg to do the same.

在给这三人的信中，他写道：几乎每天都有新消息说你们的公司正在失去对你们开发的AI技术的控制，这可能会带来灾难性的后果。

In a letter to the trio, he wrote: Almost every day there is a new story about how your companies are losing control of the AI technology you are developing with potentially cataclysmic results.

他引用了最近关于利用AI制造新型病毒的研究作为主要例子，声称：这种发展如果落入坏人之手，可能导致新的生物武器，造成数千万人死亡。

He referenced the recent scientific research about using AI to create novel viruses as the prime example, claiming: This type of development, in the wrong hands, could lead to new bioweapons that result in the deaths of tens of millions of people.

当然，尽管这项研究实际上与这三家公司毫无关系，而且使用的AI类型与他们的LLM完全不同。

This is, of course, despite the actual research having no connection to any of these three companies and using a completely different type of AI to their LLMs.

但为了桑德斯的论点，这都算AI。

But for the sake of Sanders' argument, it's all AI.

桑德斯还提到了Hugging Face被黑客攻击的事件，认为这明显违反了联邦法律。

Sanders also referenced the Hugging Face hack, arguing it was a clear violation of federal law.

最后，桑德斯在信中以略带威胁的语气总结道：让我说清楚：如果你们现在不采取适当行动，我和参议院的同事们会采取行动。

And leaving the letter on a slightly threatening note, Sanders concluded: Let me be very clear: If you do not take appropriate action now, my colleagues and I in the Senate will.

这只是华盛顿气氛升温的又一个例子，不过今天要讲的头条新闻就到这里。

Just another example of the temperature rising in Washington for now though that is going do it for today's headlines.

接下来是主节目。

Next up, the main episode.

如果你在企业里负责AI，你很清楚现在的差距不在能力，而在落地执行。

If you're leading AI inside an enterprise, you already know that the gap right now isn't capability but execution.

所以毕马威的《You Can With AI》带着新一季回来了，我们邀请了Emma的Sarojit Chatterjee、Writer的Mehbib、McKesson的CIO Ellery Fisher等高管，一起聊实践落地。

That's why KPMG's You Can With AI is back with a new season featuring conversations with leaders like Sarojit Chatterjee of Emma, Mehbib of Writer, McKesson CIO, Ellery Fisher, and others focused on practical execution.

哪些方法有效，哪些不行，以及从试点到真正规模化落地，在战略、数据准备、治理、人才和价值观方面，到底需要付出什么。

What's working, what's not, and what it actually takes to move from pilots to real scaled impact across strategy, data readiness, governance, workforce, and value.

当然，这档节目是我和Nathan Whitimore一起共同主持的。

And of course, it's co hosted by me, Nathaniel Whitimore.

去 www.kpmg.us/aipodcasts 收听并订阅吧。

Go listen and subscribe at www.kpmg.us/aipodcasts.

这就是大多数传统系统现代化项目失败的原因：做这项工作的AI无法理解大规模代码库。

Here's why most legacy modernization projects fail: the AI doing the work can't understand codebases at scale.

它只能看到一小段上下文，检查语法，却错过了分布在全球应用生态系统中多年的决策。

It sees a small slice of context, examines syntax, and misses years of decisions distributed across the global application ecosystem.

Blitzy解决这个问题的方式和它解决其他问题一样。

Blitzy solves this the way it solves everything.

在任何迁移开始之前，Blitzy的代理就基于你的代码，将整个遗留系统逆向工程成一个持续更新的知识图谱。

Grounded in your code before any migration begins, Blitzy's agents reverse engineer the entire legacy system into a persistent knowledge graph.

每一个依赖、每一个限制、每一条原本只存在某位工程师脑子里的隐性知识。

Every dependency, every constraint, every piece of tribal knowledge that used to live in one engineer's head.

基于这个理解，Blitzy自主执行语言迁移、框架升级、以及单体到微服务的改造，全程端到端验证。

From that understanding, Blitzy autonomously executes language migrations, framework upgrades, and monolith to microservices transformations, all validated end to end.

一位Blitzy客户用编码代理，在十六周内完成了一个价值一千万美元的单体保险系统的现代化升级，原本的工期是一百三十七周。

One Blitzy customer modernized a $10,000,000 monolithic insurance stack in sixteen weeks against a one hundred and thirty seven week baseline with coding agents.

这相当于压缩了九倍的时间。

That's 9x compression.

在加速你的路线图的同时，把技术债清掉。

Retire technical debt while accelerating your roadmap.

想了解详情，就去 blitzy.com 看看吧。

See how at blitzy.com.

没错，就是 blitzy.com。

That's blitzy.com.

听起来可能有点扎心：你们公司很可能在AI工具上花了成千上万美元，可这些工具却被严重浪费了。

Here's a harsh truth: Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized.

有一半的公司都配置了AI工具，但真正用它们创造商业价值的只有12%。

Half of companies have AI tools, but only 12% use them for business value.

大多数员工还只是拿AI来总结会议纪要。

Most employees are still using AI to summarize meeting notes.

如果你在公司里负责AI的落地推广，那你需要的是 Section。

If you're the one responsible for AI adoption at your company, you need Section.

Section 是一个平台，能够帮你管理整个组织的AI转型。

Section is a platform that helps you manage AI transformation across your entire organization.

它会引导员工在真实业务场景中使用AI，追踪谁在真正用AI带来业务影响，同时让你准确看到AI在哪些地方创造了价值，又在哪些地方没有起作用。

It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value.

结果呢？

The result?

你不再只是推出工具，而是能推动可衡量的AI价值。

You go from rolling out tools to driving measurable AI value.

你的员工从写会议总结转向解决实际业务问题，而且你能证明投资回报率。

Your employees move from meeting summaries to solving actual business problems, and you can prove the ROI.

别再猜测你的AI投资是否有效了。

Stop guessing if your AI investment is working.

去sectionai.com看看Section吧。

Check out Section at sectionai.com.

就是sectionai.com。

That's sectionai.com.

本期AI每日简报由Hyperagent赞助播出。Hyperagent让你能运行成群的智能体，你的团队可以一起管理。

This episode of the AI Daily Brief is brought to you by Hyperagent, where you run fleets of agents your team can manage together.

新用户可以获得1000美元的推理额度。

New users get $1,000 in inference.

别再依赖本地代理和聊天工作流，等着你在笔记本电脑上发指令了。

Forget local agents and chat workflows waiting on your laptop to be prompted.

Hyperagent 在云端部署了全天候运行的代理，在你团队已经使用的工具中完成实际工作。

Hyperagent deploys always on agents in the cloud, doing real work across the tools your team already uses.

市场部的代理能把竞争对手的动作变成落地页。

Marketing's agent turns competitor moves into landing pages.

销售部的代理能丰富潜在客户信息、起草邮件，并更新客户关系管理系统。

Sales's agent enriches leads, drafts emails, and updates the CRM.

运营部的代理负责跟进文书工作和预算追踪。

Ops agent chases the paperwork and tracks the budget.

每个代理都能访问共享上下文，并遵循你关于范围和审批的规定。

Every agent has access to shared context and follows your rules about scope and approvals.

是时候让代理像团队成员一样为你工作了。

It's time you add agents that feel like teammates.

去 Hyperagent 雇佣你的代理吧，它由 Airtable 团队打造。

Hire yours at Hyperagent, built by the team at Airtable.

到hyperagent.com/aidailybrief领取你的一千美元推理额度。

Claim your $1,000 in inference at hyperagent.com/aidailybrief.

欢迎回到AI每日简报。

Welcome back to the AI Daily Brief.

说实话，我真不记得上次看到大家对一个产品发布这么兴奋是什么时候了，人们这次对刚发布的Grok Bot简直是热情高涨。

I genuinely don't remember the last time I saw people as excited about a product announcement as people have been about the newly announced Grok Bot.

说到底，我觉得原因很简单。

And when push comes to shove, think the reason why is pretty simple.

自从今年年初Open Claw出来之后，大家一直在找办法替自己构建和部署智能体，最好能比那种技术上很复杂的系统简单一些。

Ever since Open Claw came out at the beginning of the year, people have been looking for ways to build and deploy agents on their behalf, ideally in a simpler way than what that sort of technically complex system required.

这期间有过不少尝试，但都没真正火起来，不过从第一印象来看，Grok Bot可能就是那个答案。

There have been a variety of shots on that goal but nothing has really stuck, but first impressions suggest that that's what Grok Bot might be.

Grok Bot的交互界面看起来简直就像Telegram，而Telegram当然也是OpenClaw刚出来时大多数人使用它的地方。

The interface for interacting with Grok Bot looks frankly like Telegram, which is of course where most people were interacting with OpenClaw when it first came out.

要开始使用的话，你可以创建一个新机器人，也可以跟已经建好的机器人直接互动。

To get started, you can either create a new bot, or you can interact with a bot that you've already initiated.

你通过聊天窗口和Grok Bot交互，方式和之前用OpenClaw完全一样，说实话，也跟用Clawd或Codex一样。Grok Bot在自己的虚拟电脑里后台运行，所以当任务在云端执行时，你能继续做其他事情。

You interface with your Grok Bots through the chat window, exactly as you had with OpenClaw, and frankly exactly as you do with Clawd or Codex, and Grok Bots work in the background in their own virtual computer, meaning you can keep working while the task is running on the cloud.

它还能通过人类的操作界面使用计算机，所以它能够登录网页应用、操作软件，而不需要依赖API接口。

It's also capable of using its computer through human interfaces, so it can sign into web apps and operate software without APIs.

当它需要你授权某些操作时，它会直接弹出窗口，让你在它的虚拟机器上登录，就像你在自己电脑上登录一样。

When it needs you to authorize something, it'll simply bring up that window, and have you sign in on its virtual machine in the same way that you would on yours.

当然，Grok Bot是由SpaceX的AI模型家族驱动的，这些模型现在又变得有竞争力了。

Grok Bot is of course powered by the family of SpaceX AI models, which are once again getting competitive.

也就是说，虽然在某些任务上它可能不如Fable或者Five Six Soul，但这个代理很可能已经足够处理多种多样的实际工作任务，做到有始有终。

Meaning that while it might not be as good as Fable or Five Six Soul on certain tasks, the agent likely will be good enough to handle a wide range of work tasks end to end.

实际上，就像我们看到的，SpaceX内部一直在测试Grok Bot，它基本上已经成了日常工作中的核心工具。

In fact, as we'll see, SpaceX has they've been testing Grok Bot internally, and it has basically taken over as a core work tool.

不过，这些并不是真正全新的功能，但他们把各个元素整合在一起，再加上对易用性的把握，这一步迈得非常大。

Now, none of this is net new functionality, but the way that they've put together the elements and the ease of use has made a huge leap.

这个界面把所有的复杂操作都屏蔽掉了，还直接集成了很多在其他产品上要么难以使用、要么让人失望的高级功能。

The interface abstracts all the complexity away, and that includes the native integration of a lot of advanced features that have been difficult to use or disappointing in other products.

比如，你可以同时运行多个机器人，但不会让人感到手忙脚乱，因为机器人之间的消息系统和简洁的界面，让你管理一队代理感觉轻松多了。

For example, you can run multiple bots at the same time, but rather than being overwhelming, the interbot messaging system and clean interface makes it feel way more achievable to manage a team of agents.

你甚至可以组建一个完整的代理团队，每个代理都有自己专门的角色和名字。

You can even build a whole team of agents, each with a specific role and individual name.

你的Grok Bot们可以彼此协作，像一个统一的智能代理系统那样共同运转。

Your Grok Bots can coordinate with one another, allowing them to function like a single integrated agentic system.

虽然Grok Bot目前还不能接入Slack或Teams这样的办公聊天应用，但你可以设置它们跨个人账户协同工作。

And while Grok Bot doesn't currently live in workplace messaging apps like Slack or Microsoft Teams, you can set them up to work together across individuals' accounts.

这实现了RPA时代以来的承诺。由于Grok Bot本质上是用虚拟机操作电脑的机器人，你能很容易地训练它处理日常流程——比如下次你做某件事时，让它跟着学习就行。

Integrating a promise that's been around with us since the days of RPA, because the Grok Bot is at core a computer use bot using its virtual machine, you can really easily train it on common workflows by asking it simply to follow along the next time you do a particular task.

SpaceX AI表示，机器人会记录下每一步操作，记住工作是怎么完成的，把流程存成一套例行程序。

SpaceX AI says that the bot will watch the steps and remember how the work gets done, saving the workflow as a routine.

你还能在这种过程中不断学习加修正，改进机器人的工作方式，这有点像利用屏幕录制来创建技能，但比那要自然得多。

You can also iterate on this process providing corrections that improve the bot's workflow, kind of like being able to create a skill with a screen recording but much more native.

公司表示，这些Grok机器人会随着时间不断学习，在调整写作风格、应付特殊案例，甚至在决定什么时候停下来向你确认求助等方面，都会越来越顺手。

The company says that the Grok Bots will learn over time and get better adjusting things like writing style, or handling of edge cases, or even knowing when to stop and ask for clarification.

每次有新产品发布的时候，我总会特别关注开发团队是怎么谈论这个产品的。

Now, one thing that I always watch when a new product gets launched is how the team that built it is talking about it.

你可能会觉得这没什么参考价值，毕竟开发团队肯定会夸自己的产品，对吧？

And while you might wonder how valuable that is because, of course, the team that built it is going to shill it, right?

但我觉得，从团队谈论自己作品的方式里，其实能看出很多门道。

I think you can actually get a lot of signal for How a team talks about the thing that they've built.

Cursor和SpaceX AI的团队，这是他们第一次大规模合作的产品之一，他们对这个产品简直是赞不绝口。

And the team at Cursor and SpaceX AI, for whom this is one of their first major collaborative products, are absolutely raving about this thing.

彭政写道：我们和AI的关系正在从来回聊天，转变为把真正的工作交给一个代理团队。

Peng Zheng writes: Our relationship with AI is shifting from chatting back and forth to entrusting a team of agents with real work.

Grok Bot就是那个未来的一个缩影。

Grok Bot is a glimpse of that future.

Sam Sokolin写道：这是我参与过的最酷的项目之一。

Sam Sokolin writes: This is one of the coolest projects I've worked on.

Grok Bot在内部几乎立刻就展现出了惊人的产品市场契合度。

Grok Bot had insane product market fit internally almost instantly.

对公司里很多人来说，机器人已经成了他们所有工作的操作界面。

For much of the company, bots have become the interface for all of their work.

Cursor的Ricky Dora写道：Grok Bot不是什么新概念，但它的执行真是太完美了。

Cursor's Ricky Dora writes: Grok Bot is not a new concept, but its execution is flawless.

好的工具框架、云计算、超强的电脑使用能力和顶级模型的组合，能做到的事情简直令人难以置信。

It's mind blowing what a good harness, cloud computing, massively advanced computer use, and state of the art models are capable of.

我每天都会把更多的工作自动化，这样我就能找出接下来该专注的新领域。

Every day I automate 20% more of my job so I can discover the next frontier 20% to focus on.

社区的第一印象也是同样积极的。

First impressions of the community were similarly positive.

Vishal Singh看到各种不同的使用场景，脑子都爆炸了。

Vishal Singh's brain exploded with all sorts of different use cases.

他把Grok Bot称为一种相当疯狂的东西，他说：它能自己打开电脑，登录你的应用，像人一样使用网站，你睡觉的时候它还在工作，清理你的收件箱，发邮件，更新你的客户管理系统，研究人和公司，按你的风格写领英和邮件的草稿，模仿你的工作方式。

Calling Grok Bot actually kind of ridiculous, he says: It can open a computer by itself, log into your apps, use websites like a human, work while you sleep, clean your inbox, send emails, update your CRM, research people and companies, write LinkedIn and email drafts in your style, how you work.

看你操作一次就能学会你的工作流程。

Learn a workflow after watching you once.

然后永远跑那个工作流程。

Run that workflow forever.

还能和其他机器人协调合作。

Coordinate with other bots.

只有需要做决定的时候才来打扰你。

Only bother you when it needs a decision.

还有好多我还没搞明白的功能。

And so much more that I m yet to decipher.

这不太像是AI助手，更像是一夜之间升成COO的实习生。

This is less AI assistant and more the intern who somehow became COO overnight.

Prahsanjit写道：第一次用Grok Bot，我给了它一个GitHub链接，让它读代码，不只是读README。

Prahsanjit writes: First impression with Grok Bot, I gave it a GitHub link and told it to read the code, not only the readme.

它通过API把每个文件都拉下来，然后分析了整个代码库。

It pulled every file through the API and broke down the entire codebase.

这玩意儿不只是一个聊天机器人。

This thing is not just a chatbot.

Mike P写道：我和这个机器人聊了聊，了解它是怎么工作的，我的第一印象是，这是目前所有领先公司里最好的主流Agentic AI界面。

Mike P writes: My first impression after chatting with the bot to understand how it works is that this is currently the best mainstream interface for Agentic AI that I've seen from any leading company.

他解释说，它使用虚拟机，但也可以访问你的本地文件系统。

He explains that it uses a virtual machine but can also access your local file system.

他说，关键的区别在于，你实际上不需要切换一堆UI选项或进行任何设置。

The magic he says in the key differentiator is that you don't actually have to toggle a bunch of UI options or set anything up.

Mike写道，我觉得很多人打开Cloud Cowork、Code、GPT Work或Codex，不知道怎么回事，就放弃了，因为他们不想费劲去搞清楚。

Writes Mike, I think a lot of people open Cloud Cowork or Code or GPT Work or Codex and don't know WTF is going on and just bail because they don't feel like figuring it out.

我刚刚安装了Grok Bot，连接了我的连接器，然后就开始问它问题，它就把剩下的都处理好了。

I just installed Grok Bot, hooked up my connectors, and just started asking it stuff and it just handled the rest.

当它仅仅因为我的要求就跳到我Mac上的一个本地文件夹时，我真的很震惊。

I was actually shocked when it jumped into a local folder on my Mac just from me asking.

没有UI切换或任何你需要做的来指向你想让它去的地方，你只要告诉它做什么，它就会请求你的权限，并在后台处理一切。

There's no UI toggle or anything you need to do to point it where you want it to go, you just tell it what to do and it asks for your permissions and handles everything in the background.

它还声称，当我使用iOS应用时，如果我提示它需要用到我的Mac，它会将我的Mac作为主要机器来使用。

It also claims that when I use the iOS app, it reaches for my Mac as a primary machine if I prompt it with something that requires using my Mac.

所以基本上，这就像一个遥控器，但你不必像之前用Dispatch那样去UI里的特定部分，它自己就搞定了。

So, basically, it's a remote control, but you don't have to go into a specific part of the UI like you did with Dispatch it just does it.

他总结说，实现大规模普及的关键就在于——把Agentic AI包装成一个简单到令人只需打开就能直接上手的易用产品。

This, he concludes, is what it's going to take to get mass adoption a Genetic AI wrapped up in a product so easy to use that people will be able to just unwrap it and start cooking.

早期讨论中另一个让我注意的点是，很多时候你听到人们大力夸赞它，却不具体说他们是怎么用的，但我确实看到很多人在具体分享他们的使用场景以及哪些点让他们印象深刻。

Now another notable thing in the early Discourse to me is that a lot of times you hear people raving about it without being specific about how they used it, but I saw a ton of folks actually talking about their use cases and what had impressed them in specific.

特斯拉的Yuntu Tsai写道：Grok Bot能够做到：a）查看日历，找出我需要提前预约但还没处理的事情；b）确定最佳的预约时间；c）导航并完成网站上的预约操作。

Tesla's Yuntu Tsai writes: Grok Bot was able to: a) go through the calendars and find anything I need to make reservations for beforehand that I hadn't done yet b) Determine the best time to make reservations and c) Navigate the reservations on a website.

当我在停车场里走向我的车时，我用中英混着跟它说话。

While I was walking in the parking lot before getting to my cars, I was talking to it in mixed Chinese and English.

真是让我刮目相看！

Color me impressed!

Matt Schumer写道：正是这些细节让Grok Bot与众不同。

Matt Schumer writes: The little details are what makes Grok Bot special.

比如，我设置了一个研究员机器人和一个写作机器人，然后做了一个参谋机器人，让它协调另外两个在同一个项目上协作。

For example, I set up a researcher bot and a writer bot, then made a chief of staff bot, and asked it to get the other two working together on a project.

我当时觉得这肯定会出问题，因为不可能开箱即用。

I checked in fully expecting that to fall apart because there was no way it would work out of the box.

结果它真的开箱即用了。

It worked out of the box.

老实说，他说，感觉这可能会让数百万普通人第一次真正用上智能体。

Honestly, he says, this feels like it could be the thing that gets millions of normal people using agents for the first time.

这确实是很多人的直观感受。

And that was really the gut sense for a lot of folks.

我们在OpenClaw、后来又是在Hermes那里看到的前景，现在可能终于要靠Grok Bot这样的东西迎来规模化的时刻了。

That the promise we first saw in places like OpenClaw and then Hermes might finally be getting its moment to scale with something like Grok Bot.

A16z的Martin Casado说：这是我用过的第一个真正把虚拟同事做到极致的产品。

A16z's Martin Casado says: This is the first product I've used that really nails the virtual coworker.

我觉得这次发布会被视为在职场中搞定AI抽象层的关键转折点。

I suspect we'll view this launch as a pivotal moment in getting the abstraction for AI in the workplace right.

Heaton Shaw，如果你关注他的话，从Open Claw出来以后他就一直在深入研究智能体的构建，他写道：我花了好几个月用最笨的方法去构建智能体。

Heaton Shaw, who if you follow him has been going deep on agent building ever since Open Claw came about, wrote: I spent months building agents the hard way.

我给了它们服务器、内存、技能、工具和循环。

I gave them servers, memory, skills, tools, and loops.

我把它们放进了Slack，并围绕它们建立了恢复机制。

I put them in Slack and built recovery around them.

其中一个帮我六天就交付了一个产品。

One helped me ship a product in six days.

但同一个系统也会卡住、丢失上下文，然后把没做完的部分又丢回给我。

The same system also stalled, lost context, and handed the unfinished edges back to me.

我反而成了那个基础设施。

I became the infrastructure.

所以Grok Bot才让我这么震撼。

That's why Grok Bot hit me so hard.

我早就开始测试它了。

I've been testing it early.

它的机器人之间能互相协作，在一个持久化的电脑上工作，而且在你离开的时候，它们还能继续处理你的文件和已登录的应用。

Its bots coordinate with each other, work from a persistent computer, and keep going across your files and logged in apps while you are away.

我认得出来团队建了什么，因为那些东西很多都是我自己亲手拼起来的。

I recognize what the team built because I had assembled so much of it by hand.

Grok Bot 把围绕智能体的那些隐蔽工作更多地搬进了产品本身。

Grok Bot moves more of the invisible work around the agent into the product.

这正是我等了许久的那个能干活的工作智能体。

This is the missing work agent I have been waiting for.

但没有任何产品一发布就完美无缺，对吧？

But no product can be perfect right out of the box, right?

那大家主要在抱怨什么呢？

So where are people's complaints?

有一个点是关于 Cursor 和 SpaceX 之间的命名风格。

One small one is around the naming conventions between Cursor and SpaceX.

Mike P.

Mike P.

写道：Cursor 和 SpaceX 的 AI 重叠，很有那种 Venmo 和 PayPal 的既视感。

Writes: The Cursor SpaceX AI overlap is giving Venmo PayPal vibes.

Cursor和Grok会一直分开吗？

Are Cursor and Grok going to stay separate things?

这明明叫Grok Bot，却把我导到Cursor的登录页面了。

This is called Grok Bot but I'm getting routed to a Cursor login portal.

我得上Grok账号关联到Cursor才能启动一个Grok产品。

I need to connect my Grok account to Cursor to launch a Grok product.

这是在搞什么啊，朋友们？

What are we doing here, fam?

确实，对一些人来说，Grok的包袱太重了，对这个新产品提不起兴趣。

And indeed for some, Grok has too much baggage to be excited about this new product.

Ben Barry写道：我本来会饶有兴趣地去试试Cursor Bot的，但对Grok Bot几乎没一点兴趣。

Ben Barry writes: I would have excitedly tried Cursor Bot but have little to no interest in trying Grok Bot.

它带着太多负面包袱了，我根本不敢让它碰任何东西。

Way too much negative baggage for me to ever trust it with access to anything.

我还是等着OpenAI和Anthropic推出的同类产品吧。

I'll wait for the inevitable offerings from OpenAI and Anthropic.

除此之外，有些人就是体验不佳。

Beyond that, some people just didn't have a great experience.

Gerbach Shahal写道：抱歉各位，但Grok Bot感觉就像个光鲜的玩具，演示时惊艳，但根本还不到能正式上场的时候。

Gerbach Shahal writes: Sorry guys, but Grok Bot feels like a shiny object that looks incredible in demos but simply isn't ready for prime time.

经济账完全不合理。

The economics are broken.

光是初始化它就会烧掉大量token，然后为了继续用下去又被推着花更多钱。

You burn through tokens just trying to onboard it, then get pushed towards spending more just to keep going.

这可不是可持续的工作流，就像一台坏掉的老虎机。

That's not a sustainable workflow, it feels like a broken slot machine.

更大的问题在于记忆。

The bigger issue is memory.

它会忘记上下文，跟丢任务，在较长的项目里也难以保持连贯性。

It forgets context, loses track of tasks, and struggles to maintain continuity across longer projects.

AI编程助手得记住任务，而不是每隔几小时就跟失忆似的。

An AI coding agent needs to remember the mission, not act like it has amnesia every few hours.

现在，我想呈现各种不同的看法，所以我把这个也加进来了。

Now, I want to provide a full range of views, so I'm including this.

不过，这绝对不是我看到的普遍观点。

However, this was definitely not a common take that I saw.

所以大家自己判断吧，或者随便怎么理解都行。

So contextualize it, or give it whatever grain of sand you want.

其他抱怨还包括来自June的一条，他说虽然Grok Bot的界面非常简洁，但引用原话：“它还是严重依赖集成。”

Other complaints include this one from June, saying that while Grok Bot has a really clean UI, quote: It still heavily relies on integrations.

没人愿意在刚开始使用时就去连接20多个工具。

No one wants to connect 20 plus tools during onboarding.

而且，我不确定人们是否愿意为每一个任务都创建一个专门的代理。

Plus, I'm not sure people want to create a specialized agent for every task.

Matt Schumer写道：我真正唯一的抱怨是模型路由器，我测试时它表现不太好。

Matt Schumer writes: My only real complaint was the model router, which wasn't great when I tested it.

你不能为你的Grok Bot随意挑选模型。

You don't choose a model for your Grok Bot.

这是在后台自动完成的。

It's done automatically on the back end.

做得好对普通用户来说很棒，但做得不好对高级用户来说就很烦。

Incredible for regular users when it's done well, but frustrating for power users when done poorly.

舒默补充说，他听说自他测试以来已经改进很多了。

Schumer did add I'm told they've made it much better since I tested.

马克斯·布莱德指出了通过Grok Bot的虚拟电脑登录服务的一个功能性问题。

Max Blade points out a functional issue of logging into your services through Grok Bot's virtual computer.

他说现在最大的问题是数据中心的IP地址会导致日常网站出现机器人拦截，比如在沃尔玛上买杂货都会有问题。

He says the biggest issue right now is the data center IP address creates bot blockages on everyday websites which make things like ordering groceries from Walmart problematic.

不过他提到，这个问题可以通过扩大内置连接器和插件来解决，到处都能得到官方支持。

Although he notes, quote: This will be solved by expanding their built in connectors and plugins to give official support everywhere.

可能除此之外，还有一个更广泛的信任问题，这并不只是Grok Bot特有的，但随着这些工具变得更强大，更深入地融入我们的工作和个人生活，这个问题就变得更加尖锐了。

Maybe beyond that, there's just a broader question of trust that is not unique to Grok Bot, but becomes more poignant the more powerful and the more deeply integrated into our work and personal lives these tools get.

彼得·杨写道：这个挑战，我相信其他主要实验室即将推出的产品也一样，就是：第一。

Peter Yang wrote: The challenge with this, and I'm sure upcoming products from the other major labs, is: one.

怎么让普通用户信任把自己的账号密码交给一个远程电脑？

How do you get regular users to trust sharing their credentials and logins with a remote computer?

第二点。

Two.

怎么让他们放心这个远程电脑是安全的，真正是属于他们可以随便用的？

How do you reassure them that this remote computer is secure and truly theirs to play with?

说实话，这一点我挺有共鸣的。

And honestly, this one resonates with me.

我当时在玩和测试Grok Bot的时候，确实有那一刻，我正要登进我的Spotify创作者账户，但突然停住了，心想要是出了什么岔子，那得多糟糕。

When I was playing around and testing Grok Bot I did have this moment where I was about to log it into my Spotify for Creators account but then paused, thinking about just how devastating it would be if something went wrong.

按照计算机使用方式的特点，我不仅仅是在给它访问某个分析API的权限。

By nature of the computer use paradigm, I wouldn't just be giving it access to, for example, some analytics API.

我等于在让它直接进入我的真实账户，在里面点来点去、做各种操作。

I would be letting it have access to my actual account to click around and do things.

我肯定不觉得我某个错误的指令会让它去删掉我们这档节目的所有过往剧集，但那种可能性，哪怕只是想想，也让我犹豫了。

Now I certainly don't think that some errant command of mine would lead it to go delete all past episodes of this show, but the fact that that's even possible gave me pause.

就像Peter提到的，这当然不只是Grok Bot的问题，但确实是全面转向新工作方式的一大障碍。

And like Peter noted, this is certainly not a problem for Grok Bot alone, but it is a major barrier to fully transitioning to new ways of working.

不过，也别太夸大这个问题。

Still, it's important not to overstate this.

比如，我给它权限访问我的邮箱就没什么顾虑，说实话，就算它乱发一堆邮件或者删掉一堆邮件，也不会像牵扯到节目的事那么严重。

For example, I had no problem giving it access to my email frankly, it's sending a bunch of emails that it shouldn't or even deleting a bunch of emails would not be nearly as devastating as anything having to do with the show.

到目前为止，这个工具一个很实际的问题就是只有超贵的账户才用得上。

One very practical knock on the tool so far is that it is only for extremely expensive accounts.

要么是每月300美元的Grok重度账户，包含了这个功能，要么是每月200美元的Cursor Ultra账户，但那个还访问不了顶配的Grok模型。

You're either talking about a $300 a month Grok heavy account that includes it, or a $200 a month Cursor Ultra account which doesn't have access to the top Grok model.

所以至少眼下，这个功能基本被价格挡在门外了。

So at least for the moment, this is pretty well price gated.

话虽如此，我倒是很怀疑这会是长期的设计，而不是现阶段优先为了节省推理资源而加的机制。

That said, I would be very surprised if that was a permanent feature, rather than an inference preservation mechanism, first and foremost, at this stage.

有个挺有意思的话题，其实是我这周晚些时候想做的一期节目，就是关于‘AI队友’这个比喻到底是不是合适的思考模式。

One interesting conversation, which is actually an episode that I was thinking about doing later this week, is about whether the AI teammate metaphor is actually the right mental model.

我一直都在琢磨，也打算深入探讨一下，AI队友这个概念在种种因素下可能并不合适，或许像顾问这样的模式会更好。

I've been wondering and plan on exploring whether, in fact, AI teammates are, for a variety of reasons, the wrong model, and a better might be something like consultants.

显然，我不是唯一这么想的人。Type.com的弗莱彻·里士满就写过：我们也曾认为AI队友是最合适的范式，但从客户那里我们了解到，拥有几十个AI队友反而适得其反。

But clearly, I'm not the only person thinking about this, as type.com's Fletcher Richmond wrote: We also thought a bunch of AI teammates was the right paradigm, but we've learned from customers that having dozens of AI teammates is actually counterproductive.

这纯粹是个虚荣指标。

It's a vanity metric.

相反，团队需要一个共享的工作空间，让他们能够：使用任意模型、构建包含技能、集成和上下文记忆的“公司大脑”并映射到相应的权限中、开发和托管自定义应用，以及在Slack、邮件或他们工作的任何地方进行交互。

Instead, teams need a shared workspace where they can: Work with any model Build a company brain of skills, integrations, and contextmemory mapped to their permissions build and host custom apps, and interact from Slack, email, or wherever they work.

Grok Bot 确实挺好用，但它并不是大家未来工作的常态。

Grok Bot is pretty slick, but it's not how people are going to work.

不过，我觉得 Fletcher 把它说得那么绝对，未必准确。

Now, I'm not convinced that it's as binary as Fletcher is posting it.

但我同意，我们最终会发现，在这种代理式辅助方面存在两种截然不同的模式：个人使用的工具和与团队互动、集成的工具，这两者可能会有很大差异。

But I would agree that I believe we are going to discover that there are two very different modes when it comes to this sort of agentic support, the things that you use personally and the things that interact with and integrate your team, and they might be pretty significantly different.

尽管如此，我预计在未来几周，我们会看到很多令人兴奋的实验，让人回想起 OpenClaw 早期那些精彩的时光，当时人们都在组建完整的 AI 团队。

Still, I expect that over the next couple of weeks, we are going to see a lot of exciting experiments that harken back to those early awesome days of OpenClaw, when people were spinning up entire teams.

推特上的N2Parco组建了一个核心团队，包括一名幕僚长、一名工程经理、五名工程师、一名数据分析师和一名产品经理，并分享了他们如何互动和协作。

N2Parco on Twitter has built a core team that includes a chief of staff, an engineering manager, five engineers, a data analyst, and a product manager, and shares how they interact and work together.

Farzad给团队成员起了名字。

Farzad has given his team names.

Webby是他的网页设计师。

Webby is his web designer.

Chatri是他的短视频内容创作者。

Chatri is his short form content creator.

Writy是他的文章和通讯稿写手。

Writy is his articlenewsletter writer.

你明白这个意思了吧。

You get the idea.

我确实打算做一系列实验，然后回来向你汇报我在哪些方面发现了价值。

I fully intend to do a bunch of experiments and come back and report to you on where I'm finding value.

但目前，我同意Kettlebell Dan的说法，他说：大家对Grok Bot的热情已经爆棚了。

But for now, I agree with Kettlebell Dan when he says, The excitement around Grok Bot has been off the charts today.

我好久没见过新产品这么受欢迎了。

I haven't seen this kind of reception for a new product in a while.

我个人非常期待深入研究，到时候会再跟大家汇报。

I, for one, am extremely excited to dig deeper and will report back when I do.

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.

一如既往，感谢你的收听或收看，我们下次再见，祝一切安好。

Appreciate you listening or watching as always, and until next time, peace.