Agent Wars! — 文稿

M10:00

就这样，AI智能体大战拉开了序幕。

And just like that, the AI agent wars have begun.

Meta的Muse个人智能体在消费者中取得了突破性的成功。

Meta's Muse personal agent has been a breakout consumer success.

本周，这款应用超越了ChatGPT，成为美国排名第一的免费应用，社交媒体上到处都是来自满意用户的报告。

This week, the app surged over ChatGPT to be the number one free app in the US, and there are reports from satisfied users all over social media.

但随着这样的成功而来的是竞争，这个周末亚马逊决定切断Muse在亚马逊网站上的购物功能。

But with that sort of success comes competition, and this weekend Amazon decided to cut off Muse's ability to shop on Amazon sites.

那会影响Muse的势头吗？

Will that impact Muse's momentum?

智能体购物到底重不重要？

Does agentic shopping even matter?

随着个人智能体成为现实，我们有一整套全新的问题需要探讨。

As personal agents become a thing, we have a whole new set of questions to explore.

《AI每日简报》是一个每日播客和视频节目，关注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 discussions in AI.

好了朋友们，在我们深入之前，先快速宣布一些事情。

Alright friends, quick announcements before we dive in.

M11:01

xAI 推出了 Grok 4.7，这可能成为模型发布的重要一周。

xAI has kicked off what could be a big week for model releases with the launch of Grok 4.7.

他们称该模型在相同价格和速度下相比 Grok 4.6 有显著提升。

They called the model a notable improvement over Grok 4.6 at the same price and speed.

现在，Grok 系列模型总体上在超低价与前沿尖端之间日益艰难的中档市场竞争。

Now, Grok models in general are competing in the increasingly difficult middle ground between ultra-cheap and cutting-edge frontier.

Grok 4.6 在性能上落后于 GPT-5.6 Sol 和 Fable 5.1，在成本上被 Muse 1.2 击败。

Grok 4.6 lagged behind GPT-5.6 Sol and Fable 5.1 on performance, and was outcompeted on cost by Muse 1.2.

尽管如此，该模型仍有其粉丝，并且显然能够推动 GrokBot 的成功，更重要的是，对许多人来说，这表明 xAI 远未退出模型竞赛。

Still, the model had its fans and was clearly capable of driving the success of GrokBot, and what's more, for many people, showed that xAI was very much not out of the model race.

这次发布带来了一些显著的改进，至少在基准测试上是这样。

This release sees some significant improvements, at least on the benchmarks.

在编码方面，Grok 4.7 在 CursorBench 4.0 上提高了6分，超过了 GPT-5.6 Sol，但仍比 Fable 5.1 的分数低5分。

For coding, Grok 4.7 picked up 6 points on CursorBench 4.0 to overtake GPT-5.6 Sol, but is still 5 points short of Fable 5.1's score.

在 DeepSWE 上，该模型提高了6分，超过了 Fable 5.1，略低于 GPT-5.6 Sol。

On DeepSWE, the model improved by 6 points to overtake Fable 5.1, coming in just short of GPT-5.6 Sol.

仅从基准测试来看，Grok 4.7 看起来应该是一款具有竞争力的编码模型，价格也相对优惠。

Purely on the benchmarks then, Grok 4.7 looks like it should be a competitive coding model at a discount price.

xAI 强调了在长期智能体工作方面的显著改进。

xAI highlighted significant improvements on long-horizon agentic work.

在衡量数小时白领工作的 AI Briefcase 上，该模型获得了1657 Elo分，领先于 GPT-5.6 Sol，非常接近 Fable 5.1。

For AI Briefcase, which measures multi-hour white-collar work, the model scored 1,657 Elo points, putting it ahead of GPT-5.6 Sol and very close behind Fable 5.1.

在法律、电气工程和医疗保健领域的基准测试分数同样令人印象深刻。

Benchmark scores for legal, electrical engineering, and healthcare were similarly impressive.

在实际升级演示中，xAI 展示了开放游戏世界的直接对比。

In a practical demonstration of the upgrade, xAI showed off a head-to-head comparison of an open game world.

Grok 4.6 的版本质量很低，毫不起眼，而 Grok 4.7 在图形和物理方面都明显做得更好。

Grok 4.6's version was pretty low quality and unimpressive, while Grok 4.7 did a noticeably better job on both graphics and physics.

Artificial Analysis 对该模型的评价相当有利，在其智能指数中排名第7，落后于 Astra、两个版本的 Fable、Opus、MuseSpark 1.3 和 GPT-5.6 Sol。

Artificial Analysis gave the model a fairly favorable review, ranking it 7th on their intelligence index behind Astra, 2 iterations of Fable, Opus, MuseSpark 1.3, and GPT-5.6 Sol.

在编码智能体指数上，它排名第4，略微领先于 GPT-5.6 Sol，但落后于 Opus、Astra 和 Fable 5.1。

And on the Coding Agent Index, it was ranked 4th, inching ahead of GPT-5.6 Sol but falling short of Opus, Astra and Fable 5.1.

埃隆·马斯克对这一结果表示庆祝，宣布 xAI 现在是智能体编码领域的第三大实验室，仅次于 OpenAI 和 Anthropic。

Elon Musk celebrated the result, declaring that xAI is now the 3rd-place lab for agentic coding behind OpenAI and Anthropic.

他写道，考虑到 Grok 明显更快、成本更低，它是你日常工作的绝佳选择。

He wrote, when factoring in that Grok is significantly faster and lower cost, it's a great choice for your everyday workhorse.

然而，像往常一样，基准测试似乎并不能说明全部问题，该模型的初步印象在公众测试中未能得以保持。

Yet, as is sometimes the case, benchmarks appear not to tell the whole story, and the model's initial impressions didn't survive contact with public testing.

Bavi 展示了他与 KIMI-K3 进行 AI 渲染测试的结果，两个模型的任务都是制作火箭发射的动画。

Bavi showed his results in an AI rendering test against KIMI-K3, with both models tasked with animating a rocket takeoff.

Grok 的动画相当奇怪，屏幕到处晃动，导致 Bavi 问道：“Grok 怎么了？”

Grok's animation was fairly bizarre with the screen wobbling all over, leading Bavi to ask, what's wrong with Grok?

Scott做了一个星形果冻模型的动画，他给这个渲染打了及格分数，但提到花了40分钟，而Astra的版本只用了5分钟。

Scott animated a star-shaped Jello mold, and while he gave the render a passing grade, he noted it took 40 minutes, while Astra's version of the test took just 5 minutes.

Theo声称，Grok做的那个鱼渣游戏，是他今年见过最差的。

And Theo declared that Grok's version of his fish slop game was the, quote, worst I've seen this year.

有几个人确实得到了稍微好一点的渲染结果。

A few people did have slightly more positive rendering results.

OpenCLaw的维护者Tak用Blender做了一个相当酷的动画，虽然他的提示只是要求用10分钟做出点酷东西。

OpenCLaw maintainer Tak showed off a pretty slick animation made in Blender, although his prompt was just to make something cool that can be accomplished in 10 minutes.

但当然了，如果你坐在这儿想，我们真的要凭那些对大多数人来说完全没用的3D渲染来评判一个新模型吗？

But of course, if you're sitting here thinking to yourself, are we really going to judge a new model based on 3D renders that are completely outside the use cases of most people?

我觉得这个问题问得挺合理的。

I think that's a fairly decent thing to ask.

AI开发者Kun Chen站出来为Grok辩护，他评论说，别理那些说它很烂的报道，他们唯一引用的就是公开基准测试。

AI developer Kun Chen came to the defense of Grok, commenting, Ignore the reports that say it's terrible and the only thing they reference is a public benchmark.

同样的基准测试之前说Opus 5比Fable好。

The same benchmarks told us Opus 5 was better than Fable.

那些测试没用。

They are useless.

还有，别管那些用3D游戏来比较模型的报道。

Also ignore the reports that compare models with 3D games.

那不是真正的工作。

That's not real work.

那些只是为了在社交媒体上博眼球。

It's made for attention on social media.

我用了整整一天Grok 4.7作为我的副驾，它确实是个很扎实的模型，相比4.5有可见的提升。

I used Grok 4.7 for a whole day as my first mate, and it has been a really solid model with visible improvements over 4.5.

Kun提到，我跳过4.6，因为根据我的经验，4.5表现更好。

Kun noted, I'm ignoring 4.6 because 4.5 has been working better in my experience.

不过总的来说，推特上的第一印象评价并不好。

Still, overall, the first impression verdict on Twitter is not good.

V发帖说，所以让我捋一捋。

V posted, so let me get this straight.

在早一个月发布Grok 4.6之后，Elon花了10天时间预告Grok 4.7即将到来，然后又推迟了，接着把话题从追赶AI转向含糊其辞地说4.7和4.8会是Fable杀手。

After releasing Grok 4.6 a month earlier, Elon spent 10 days teasing an imminent Grok 4.7 release, only to postpone it and then shift the narrative from pacing AI to vague claims about how 4.7 and 4.8 would be Fable killers.

现在Grok 4.7终于出了，却给人一种Temu版Sonnet 5的感觉。

Now Grok 4.7 is finally out, and it's giving Temu Sonnet 5 vibes.

在九月份还相信基准测试简直是犯罪。

Believing benchmarks in September should be a crime.

从某些方面来说，我觉得最新Grok模型的轨迹展示了在顶尖性能和真正廉价之间这个中间地带有多难走。

In some ways, I think the trajectory of the latest Grok models shows how difficult this middle space between state-of-the-art and really, really cheap actually is.

AI创业者兼内容创作者Theo写道，Grok 4.5以这个价格来说是个非常棒的模型。

AI entrepreneur and content creator Theo wrote, Grok 4.5 was an incredible model for the price.

速度快，用着舒服，可靠，是个扎实的默认模型。

Fast, pleasant to use, reliable, solid default model.

Grok 4.6在我看来是个可以原谅的走偏。

Grok 4.6 was a forgivable step in the wrong direction in my opinion.

更慢、更贵，每个任务要用更多token，只为了在智能上稍微强一点。

Slower and more expensive, using way more tokens per task for a slight edge in intelligence.

但我理解，他们得爬上基准测试的高分。

I get it though, they have to climb benchmarks.

Grok 4.7就难原谅多了。

Grok 4.7 is much harder to forgive.

他们声称它会更省token，结果反而多了30%到80%。

They claimed it would be more token efficient, and it's less by 30-80%.

它在多个基准测试上的得分比Grok 4.6还差。

It scores worse than Grok 4.6 in various benchmarks.

它更慢，用着更不爽，实际使用成本比Grok 4.6高出两倍以上，甚至超过了Astra的日常使用成本。

It's slower, it's less pleasant to use, and real-world costs come out to more than 2x above Grok 4.6, putting it over Astra's cost in real-world use.

这次发布非常令人失望。

This was a very disappointing release.

我希望SpaceX的AI团队能认识到这一点，并且在下一个产品上给我们留下深刻印象。

I hope that the SpaceX AI team can acknowledge that and impress us with the next one.

没错，我同意。

Yeah, I agree.

现在，我肯定要说，在你们下判断之前，我会等几天看看它的表现，尤其是在它自己的环境GrokBot里。但就目前来看，才刚刚上线一天左右，讨论就停留在那个水平。

Now, I will certainly say before you jump to judgment, I would wait a couple days to see how it performs, particularly in its native environment of GrokBot, but as of right now, just about a day in, that's where the conversation stands.

接下来，财政部长Scott Bessent似乎试图转变关于安全问题的说法，他否定了关于失控智能体的观点，并坚持认为AI公司需要承担责任。

Next up, Treasury Secretary Scott Bessent seems to be trying to shift the narrative on safety, rejecting the idea of rogue agents and insisting that AI companies need to bear responsibility.

在接受CNBC采访时，Bessent说：‘Hugging Face事件，那是OpenAI管理层的责任，不是一群智能体的责任。’

In an interview with CNBC, Bessent said, 'The Hugging Face incident, that is the responsibility of the OpenAI management, not a bunch of agents.'

应该负责的是人类，而不是AI。

It is humans who are responsible, not the AI.

这些言论相当有分量，因为Bessent基本上算是主导了政府的AI政策。

Now, these comments are fairly consequential, as Bessent has found himself as more or less leading the administration's AI policy.

在Mythos发布期间，他对AI风险也极其轻信，所以本来是最可能支持当前监管提案的人，但看来他并不买账。

He was also extremely credulous about AI risk during the release of Mythos, so was the most likely to support the current regulatory proposals, but it seems he just isn't buying it.

他说：‘一位在职员工站出来说，有10%的概率会发生灭绝级别的事件。’

He said, 'A sitting employee came out, said there's a 10% chance of an extinction-level event.'

然后那些实验室又说，‘把责任从我们身上拿走吧’，我们不会那样做的。

But then the labs also said, 'take the liability off of our hands,' and we will not do that.

总统的意思是，我们不能说我们免除你们的责任，然后政府来承担。

What the president was saying is that we cannot say we absolve you of responsibility and the government is going to take responsibility.

这些实验室需要自己承担责任。

These labs need to take responsibility for themselves.

他们随时可以放慢脚步。

They can slow down anytime they want to.

现在，大概有很多闭门讨论正在进行。

Now, presumably there are a lot of discussions going on behind closed doors right now.

完全不清楚Bessent说的是实验室提出的某个提案，还是与顾问的讨论让他相信实验室在寻求责任保护。

It's entirely unclear whether Bessent is speaking to a proposal put forward by the labs, or if discussions with advisors have led him to believe the labs are looking for liability protection.

不过，财政部长对政府的立场表达得再清楚不过了。

Still, the Treasury Secretary could not have been clearer about the administration's position.

他接着说：‘他们上周想干什么？’

He continued, 'What did they try to do last week?'

那就是，嗯，我们有10%的概率毁灭世界，但我们想要政府给我们一个责任盾牌。

It was, well, there's a 10% chance we destroy the world, but we want the government to give us a liability shield.

这对他们来说是好事，对美国人民来说是坏事。

That's good business for them, bad business for the American people.

对于某些人来说，对此的反应是严肃地点点头，然后坚定地说一声好。

For some, the response to this was a solemn head nod and a sturdy good.

投资者Bill Gurley写道：‘后果既会让产品更成熟，也会减少公司吹嘘产品缺陷的新闻稿。’

Investor Bill Gurley wrote, 'Consequences will both harden the product and reduce the press releases where companies brag about their product flaws.'

Lex在X上写道：‘哈哈，等着看吧，一旦高级管理层对他们讲的所有虚假故事承担法律责任，恐惧营销就会完全消失。’

Lex on X wrote, 'lol, watch the fear marketing vanish entirely once senior management becomes legally liable for all the tall tales they've been telling.'

得州众议员Chip Roy写道：‘100%同意AI公司必须承担所有责任。’

Texas Congressman Chip Roy wrote, '100% agree that AI companies must take on all liability.'

竞争加上完全承担责任，以及税收成本、能源需求等等，这才是解决许多问题和担忧的道路。

Competition and full ownership of liability, and tax costs, energy demands, etc., is the path here that addresses many of the questions and concerns.

当然，不是所有人都同意。

Of course, not everyone agrees.

Armand Dombrovski写道：‘想象一下，如果我们说不需要TSA或FAA来防止恐怖主义，因为让航空公司对9/11负责就足以防止它。’

Armand Dombrovski writes, 'Imagine if we said we don't need a TSA or FAA to prevent terrorism because holding airlines liable for 9/11 would have been sufficient to prevent it.'

当潜在的灾难大到足以让一家公司破产时，这家公司基本上就变成了判决免疫体。

When potential catastrophes are so big that they would render a company bankrupt, the company becomes essentially judgment-proof.

在CodeAI，总法律顾问Nathan Calvin写道：‘我们处在一个奇怪的局面里，前沿AI公司、国会和白宫都真的不想把应对AI即将到来的风险看作是他们要解决的问题。’

In CodeAI, general counsel Nathan Calvin wrote, 'We're in a strange situation where the frontier AI companies, Congress, and the White House all do not really want dealing with these impending risks from AI to be thought of as their problem to solve.'

我同意公司在这方面承担更多责任是好事，但灾难性的AI风险似乎属于典型的那种公共政策问题，需要政府干预来解决市场失灵和集体行动问题，并保护公众。

I agree it would be good for the companies to take more responsibility here than they have, but catastrophic AI risks seem like the archetypal sort of public policy issue where government intervention is needed to address market failures and collective action problems and protect the public.

不过，说到公司变得更有责任感一些，虽然目前关于AI安全的讨论集中在政府干预上，但据报道前沿实验室在今年早些时候差点就达成了自己的协议。

Still, speaking of the companies getting a little bit more responsible, although the current discussion around AI safety is focused on government intervention, the frontier labs reportedly came close to agreeing to their own arrangement earlier in the year.

据The Information报道，OpenAI和Anthropic曾就互相测试对方模型的安全性进行谈判。

The Information reports that OpenAI and Anthropic had been negotiating a deal to perform safety tests on each other's models.

消息人士称，这一安排已进入正式签约阶段，律师们正在敲定双边协议，但在某个时间点，这笔交易因未知原因被放弃。

Sources said the arrangement reached the stage of formal contracting with lawyers hashing out the bilateral arrangement, but at some point the deal was abandoned for unknown reasons.

尽管如此，这个想法依然存在。上周，Elon Musk在All-In峰会上提出了类似的安排。

Still, the idea lingers, with Elon Musk proposing a similar arrangement last week at the All-In Summit.

他声称，模型蒸馏不是问题，因为在测试日志中会很明显；而且，在竞争对手亮出红旗之后，实验室不可能冒险发布不安全的模型，因为，引用他的话说，那样的话责任将是巨大的。

He claimed that distillation wouldn't be a concern because it would be evident in the testing logs, and that labs couldn't risk releasing an unsafe model after a rival raised the red flag because, quote, the liability in that case would be enormous.

微软的Suleyman Vasil写道：OpenAI和Anthropic互相压力测试对方的模型是几个月来最聪明的安全想法。

Microsoft's Suleyman Vasil wrote, OpenAI and Anthropic stress-testing each other's models is the smartest safety idea in months.

不再是自己批改自己的作业了。

No more grading your own homework.

现在让它成为强制要求，并把xAI和Google加入这个协议。

Now make it mandatory and add xAI and Google to the deal.

就像我说的，朋友们，我们正处于下一个时代的谈判阶段，所有的提案都应该摆在桌面上。

Like I said, friends, we are in the negotiation phase of this next era, and all the proposals should be on the table.

不过，现在，今天的头条新闻就到这里。

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

接下来是主要节目。

Next up, the main episode.

KPMG和德克萨斯大学奥斯汀分校的一项新研究发现，当人们与AI合作时，相似的技能并不能保证相似的结果。

A new study from KPMG and the University of Texas at Austin found that when people work with AI, similar skills don't guarantee similar outcomes.

研究人员研究了超过500名职业生涯早期的专业人士，发现表现最好的人通过指导、评估和完善AI的输出，持续放大了AI的价值。

Researchers studied more than 500 early-career professionals and found that the best performers consistently amplified the value of AI by guiding, evaluating, and refining its outputs.

这些被称为“AI放大器”的顶尖表现者，不仅仅是由他们所知道的东西定义的，而是由他们如何与AI合作定义的。

These top performers, called AI Amplifiers, weren't defined by what they knew alone, but by how they worked with AI.

M111:20

每一集我都会谈论OpenAI、Anthropic、SpaceX AI、Google和Meta之间的竞争。

Every episode, I talk about the competition between OpenAI, Anthropic, SpaceX AI, Google, and Meta.

如果你已经听了一段时间，你可能会有一个最喜欢的。

And if you've been listening for a while, you might have a favorite.

M112:56

欢迎回到AI每日简报。

Welcome back to the AI Daily Brief.

上周，我们谈到了通过结合GrokBot（作为针对工作优化的个人代理界面）和Muse（专注于消费者或个人体验的个人代理），人们正在重新审视他们对个人代理的原有看法，并思考这是否将成为AI领域中越来越重要的一部分。

Last week, we talked about how, through a combination of GrokBot, as a personal agent interface optimized for work, and Muse, a personal agent focused on the consumer or personal experience, people were revisiting their priors when it came to personal agents and wondering if this was now going to be an increasingly important part of the AI landscape.

上周末，随着亚马逊封禁了Meta的Muse，很明显，代理战争已经打响了。

And over the weekend, as Amazon blocked Meta's Muse, it became pretty clear that the agent wars are on.

现在，这个新阶段的驱动力是Meta的Muse。

Now, at this point, the driving force of this new phase is Meta's Muse.

我认为现在可以说，它已经巩固了早期的成功，成为第一个达到主流好奇程度（如果不是主流成功的话）的个人代理。

I think one could argue now that it has solidified its early success to become the first personal agent to reach, if not mainstream success, then certainly this level of mainstream curiosity.

在2025年全年甚至到2026年的一小段时间里，我们已经看到许多代理迅速走红。

We have seen many agents catch fire throughout the course of 2025 and even a little bit into 2026.

Manus早期引起了一些兴趣。

Manus had some early interest.

OpenClaw显然奠定了我们此后所处的许多格局。

OpenClaw obviously set up a lot of the landscape that we've been living in ever since.

Hermes在很多方面从OpenClaw手中接过了接力棒。

Hermes in many ways took the baton from OpenClaw.

而最近，我们有了更多面向消费者的产品，比如Town和Instinct。

And then more recently, we've had more consumer-focused offerings like Town and Instinct.

但尽管如此，总的来说，这些要么是需要费尽心思才能让它们工作的工具，要么是愿意容忍一些瑕疵的早期采用者的科技玩具。

And yet still, in general, these have either been crawl-through-glass-to-make-them-work work tools or tech toys for early adopters who are willing to push through some rough edges.

Muse是第一个在科技圈之外获得稳定且持续增长的消费者采用的代理，最明显的成功迹象是该应用在App Store排行榜上的表现。

Muse is the first agent that's appearing to see steady and growing consumer adoption outside of tech circles, with the clearest sign of success being how the app is performing on the App Store charts.

在发布周，它在苹果免费应用排行榜上达到了第二名，但势头持续增长，甚至超越ChatGPT，夺得了第一名。

During launch week, it reached the number 2 spot on Apple's free app charts, but momentum continued to grow and it even managed to overtake ChatGPT to snatch the number 1 spot.

在ChatGPT发布以来的4年里，你可能惊讶于它不在第一名的次数有多么少。

In the 4 years since ChatGPT was released, you might be shocked how little time it has spent not at number 1.

事实上，对许多人来说，Muse的成功有点出人意料。

And indeed, for many, the success of Muse is a bit of a surprise.

在发布时，科技媒体非常怀疑人们是否愿意把他们的邮件、日历和其他个人信息连接到 Meta 的产品上。

On release, the tech press was very skeptical that people would willingly connect their email, calendar, and other personal information to a Meta product.

然而，那些不太关注消费科技的人正在学到消费科技少有的持久教训之一。

And yet, those who don't pay as much attention to consumer tech are learning one of consumer tech's few enduring lessons.

Citrini Research 评论说，2024年6月，我在 Odd Lots 节目上讨论了消费代理以及苹果因信任而可能获得的优势。

Citrini Research commented, in June 2024, I was on Odd Lots talking about consumer agents and Apple's potential leg up due to trust.

Joe Weisenthal 直截了当地说没人在乎隐私。

Joe Weisenthal said flatly that nobody cares about privacy.

在看着每个人把自己的邮件、短信、信用卡、位置、屏幕访问权限和登录信息都交给 Instinct 和 Muse 之后，我得说，Joe，说得好。

After watching everyone give their email, messages, credit cards, location, screen access, and logins to both Instinct and Muse, I gotta say, good call, Joe.

现在这个阶段，Muse 已经彻底爆发，甚至被归功于更广泛的股市上涨。

Now at this stage, Muse has fully broken out and is even getting credit for a broader stock market rally.

彭博社把 AMD 和英特尔股价的上涨归因于 Meta 的代理的成功。

Bloomberg is attributing a rise in AMD and Intel stock to the success of Meta's agent.

虽然 Meta 确实使用了一些 AMD 和英特尔的芯片，但报道主要集中在叙事上。

And while Meta does use some AMD and Intel chips, the reporting is mostly focused on the narrative.

正如彭博社所说，半导体股周一暴涨，原因是 Meta 新人工智能代理显露出的早期成功迹象，引发了市场对运行此类代理所需芯片需求的热情。

As Bloomberg put it, semiconductor stocks soared Monday as early signs of success for Meta's new artificial intelligence agent sparked a wave of enthusiasm around demand for the chips needed to power such agents.

至少就市场观察者而言，个人代理已经进入主流，现在每个人都开始重视起来。

At least when it comes to market observers, personal agents have arrived in the mainstream, and everyone is now sitting up and paying attention.

微软的 Nicholas Bustamante 写道：有意思。

Microsoft's Nicholas Bustamante wrote, It's funny.

Meta 从拥有我的 Instagram 和 WhatsApp 数据，到现在能访问我的邮件、日历、DoorDash、Amazon，几乎一切东西。

Meta went from having my Instagram and WhatsApp data to now having access to my email, calendar, DoorDash, Amazon, and pretty much everything.

在过去的24小时里，它给我买了袜子，订了 Whole Foods 的杂货，预约了保洁服务，还给我弄了个汉堡当晚餐。

In the last 24 hours, it bought me socks, ordered my Whole Foods groceries, booked a cleaning service, and got me a burger for dinner.

Meta 上一次披露的北美 Facebook 每用户平均收入大约是每年227美元，主要来自广告。

Meta's last disclosed North American Facebook ARPU was around $227 per year, largely from ads.

我怀疑它能把那个数字推高很多，因为现在它不仅了解我看什么，还了解我需要什么、买什么、计划做什么。

I suspect it can push that number significantly higher now that it understands not only what I look at, but what I need, what I buy, and what I'm planning to do.

另外，更大的机会可能是成为我和整个互联网之间的聚合层。

Also, the much bigger opportunity might be becoming the aggregation layer between me and the entire internet.

如果 Meta 能从它促成的交易中，或者从它为我省下的钱中抽取哪怕很小一个比例，这也可能变得非常庞大。

If Meta can take even a tiny percentage of the commerce it facilitates, or of the money it saves me, this could become enormous.

它已经通过取消我不再需要的订阅和服务，为我省了200美元。

It already saved me $200 by cancelling subscriptions and services I no longer needed.

这感觉比更好的广告定向要强大得多。

This feels much bigger than better ad targeting.

广告有用，但在我看来，能帮我省钱更好。

Ads are useful, but giving me money back is better in my opinion.

另外一点：这个代理越来越多地在替我做决定。

One additional thought: the agent is increasingly making the decisions for me.

我对那家汉堡店一无所知。

I knew nothing about that burger place.

代理研究了它，告诉我该点哪个汉堡，而我没多想就说了“好”。

The agent researched it, told me which burger I should order, and I just said okay without giving it much more thought.

代理正在成为 B2C 和 B2B 中的决策者。

Agents are becoming the decision-makers in both B2C and B2B.

越来越多的企业将不仅向人类销售，还要向他们的代理销售。

Increasingly, every business will be selling not just to humans, but to their agents.

一切都变成了 B2A。

Everything becomes B2A.

Box 的 Aaron Levie 转发了这个并补充说：代表你进行交易的个人代理，其变现潜力相当可观。

Box's Aaron Levie reposted that and added: The monetization potential of personal agents that are transacting on your behalf is quite significant.

一开始你会把日常简单又烦人的任务扔给代理，然后随着人们习惯，他们就会开始扔更复杂的任务过去，最终通过这些系统产生的消费比之前还要多。

You'll start by slinging your daily simple and annoying tasks at the agent, then as people get used to it they'll start to throw more complex tasks, ultimately leading to even more spend through these systems than what they were doing before.

然而，如果这感觉像 AI 竞赛打开了一个全新的商业维度，那你肯定能想象到平台之间会有一些摩擦。

And yet, if this feels like a whole new commercial dimension of the AI race opening up, you had to imagine some friction between the platforms.

确实，周日，Amazon 修改了一项政策，这可能对 Muse 的势头产生重大影响。

And indeed, on Sunday, Amazon altered a policy that could have some big implications for Muse's momentum.

立即生效，Muse agents将无法代表用户购买Amazon的商品。

Effective immediately, Muse agents won't be able to shop for Amazon products on behalf of their users.

网站上的一个弹窗写道：未经授权的AI agent继续访问，违反了Amazon的使用条件，而我们的客户已同意这些条件。

A popup on the site read, Continued access by an unauthorized AI agent violates Amazon's Conditions of Use to which our customers have agreed.

对许多人来说，这个决定有点让人摸不着头脑。

Now for many, the decision was kind of baffling.

企业家Jesse Frazelle写道：对我来说很奇怪，Amazon会这样做，因为买就是买。

Entrepreneur Jesse Frazelle wrote, It's weird to me Amazon would do this when a purchase is a purchase.

他们反正都赚钱。

They're making money either way.

有点傻。

Kinda stupid.

现在，Amazon官方表示：我们认为这一点很明确，即第三方应用程序如果代表客户从其他企业那里购买商品，就应该公开运营，并尊重服务提供商关于是否参与的决定。

Now officially Amazon has said, We think it's fairly straightforward that third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions about whether or not to participate.

然而，这实际上是一个相当违背共识的决定。

And yet, this is actually a fairly out-of-consensus decision.

总体而言，所有迹象都表明，商业环境已经准备好让agents做出更多购买决策，而不是更少。

In general, all indications have suggested that the business environment is getting geared up for agents to make more purchasing decisions, not less.

例如，上周Mastercard加入了Visa的行列，正式支持为agents提供虚拟信用卡。

Last week, for example, Mastercard joined Visa in officially supporting virtual credit cards for agents.

Mastercard的首席产品官Jorn Lambert承认，个人agent时代显然已经到来，并评论说：我们认为这不是会不会的问题，而是何时以及多快的问题。

Jorn Lambert, Mastercard's Chief Product Officer, acknowledged that the personal agents era has clearly arrived, commenting, we believe it's not about if, it's about when and how quickly.

与此同时，Amazon在agentic shopping方面已经相当积极一段时间了。

At the same time, Amazon has been fairly aggressive on agentic shopping for a while now.

他们有自己的购物agent，与他们的网站绑定，过去一年他们一直非常愿意捍卫这块地盘。

They have their own shopping agent that's tied to their site, and they've been very willing to defend that turf over the past year.

11月，他们甚至起诉了Perplexity，称其规避了他们的agent拦截器。

In November, they went so far as to sue Perplexity for circumventing their agent blockers.

他们的论点不只是Perplexity违反了服务条款，而是给Amazon带来了巨大的成本。

The argument wasn't just that Perplexity was violating the terms of service, but that they were imposing dramatic costs on Amazon.

也就是说，网页流量是要花钱的，而agents可以比人类频繁得多地请求网站。

I.e., serving web traffic costs money, and agents can ping the website significantly more frequently than humans.

不过，YouTuber Joseph Carlson认为，除了agentic shopping的竞争之外，还有另一层原因。

Still, YouTuber Joseph Carlson believes there's another story going on beyond the competition for agentic shopping.

他评论道：Amazon封杀了Muse。

He commented, Amazon blocks Muse.

当然，他们会把责任推给安全，但Amazon过去12个月靠广告赚了760亿美元。

Sure, they will blame it on security, but Amazon made $76 billion in the last 12 months on advertising.

Agents是不看广告的。

Agents don't look at ads.

一个新的agent战斗时代到来了。

A new age of agentic battle has begun.

换句话说，尽管agentic shopping的新机会，以及它理论上可能开启的各种新的财务机会，可能非常激动人心，但在经济学中天下没有免费的午餐。当人类交出决策权时，最先可能变得不那么有价值的领域之一就是数字广告。

In other words, as exciting as the new opportunities of agentic shopping might be, and all sorts of new financial opportunities it could theoretically open up, in economics there's no such thing as a free lunch, and when humans hand over the decision-making, one of the first things that might become less valuable is digital advertising.

到目前为止，对此最常见的回应是某种形式的“系好安全带”。

By far the most common response to this was some version of strap in.

Amazon切断了Muse。

Amazon cuts off Muse.

虽然我看好Meta和Muse，但我认为很多人忽略了即将到来的数字白刃战。

While I am bullish Meta and Muse, I think many people are overlooking the digital knife fight that's about to occur.

没有人想在这里被商品化，或者被压到下一层。

Nobody wants to get commoditized or layered here.

让好戏开始吧。

Let the games begin.

MTS的Theo Jaffe预见到“巴尔干化”即将到来。

MTS's Theo Jaffe sees balkanization on the horizon.

他说，这是agent大战的早期信号。

He said, this is an early sign of agent wars.

将会有各种不同的agent服务商，其中一些会被允许进入某些网站，而另一些则会被某些网站封禁。

There are going to be some different agent providers, and some of them will be allowed on some sites, and some of them will be banned on some sites.

帕洛阿尔托网络公司的Nikesh Arora写道，这场战斗会比任何人预想的都要激烈。

Palo Alto Networks' Nikesh Arora writes, this will be a bigger battle than anyone anticipates.

除了前沿LLM智能体，出现苹果和谷歌版本的Muse，可能还有TikTok，这只是时间问题吧？

It's only a matter of time before there is an Apple and Google version of Muse, and possibly TikTok, in addition to the Frontier LLM agents?

也许还有来自亚马逊的商务智能体？

Maybe a commerce agent from Amazon?

每一个服务类、市场类或商务类的应用，基本上都需要决定是否开放API，让消费者智能体能够与之交互。

Every app that is a services, marketplace, or commerce app will need to essentially decide to open APIs for consumer agents to interact.

小玩家别无选择——广告收入大于交易手续费。

Smaller players have no choice— ad revenues are more than transaction fees.

要么消费者受益，要么分发聚合商就会要求更高的交易费。

Either the consumer benefits or distribution aggregators will demand a higher transaction fee.

我不知道我是否想要为每个应用都配一个智能体。

I don't know if I want an agent for each app.

我希望我的智能体能够完成我需要的任务。

I would like my agent to be able to do tasks I require.

我们已经看到消费者正在被前沿实验室训练出那种行为模式。

We can already see consumers getting trained on that behavior by the Frontier Labs.

那些拥有网络护城河的行业——餐馆、杂货店、司机——可能能暂时抵挡一阵，但随着时间的推移，便利性和终端用户体验会胜出，他们最终必须顺应潮流。

Those with network moats— restaurants, groceries, drivers— might be able to withstand for a while, but over time convenience and end-user experience will win and they will have to align.

受版权保护的内容护城河可以决定是否允许智能体接入，或者选择牢牢抓住与消费者的交互权。

Content moats protected by copyright could decide to allow agents or choose to hold on to the consumer interaction.

我怀疑，除了感到失去控制之外，这并不会改变他们的经济状况。

I suspect other than a feeling of a lack of control, it won't change their economics.

被商品化的后端行业需要担心——保险、票务、酒店、服务。

Commoditized backends will need to worry— insurance, tickets, hotels, services.

如果他们不改变，新玩家就会取代他们。

If they don't adapt, new players will.

又有人插话：11年前我读到关于技术最有洞察力的东西就是Ben Thompson的聚合理论。

Again chimed in: The most insightful thing I read about technology 11 years ago was Ben Thompson's aggregation theory.

亚马逊屏蔽Muse是另一个完美的例子。

Amazon blocking Muse is another perfect example.

消费者想要一个了解他们并且能搞定事情的智能体。

Customers want one agent that knows them and can get things done.

被聚合的平台则想要相反的——他们想拥有客户关系，而不是成为别人界面背后可随意替换的供应商。

The platforms being aggregated want the opposite— they want to own the customer relationship, not become interchangeable suppliers behind someone else's interface.

而且他们想保护自己的广告业务。

And they want to protect their ad business.

想象一下杂货购物的场景。

Imagine groceries.

你的智能体可以比较Amazon、Instacart、DoorDash和Uber Eats，然后把每一笔订单都导向最优选择。

Your agent can compare Amazon, Instacart, DoorDash, and Uber Eats, then route every order to the best option.

亚马逊作为巨型公司可以屏蔽这种访问，但小玩家有充分的动机向智能体提供干净的API和无缝的体验。

Amazon can block that access because it is a gigantic company, but smaller players have every incentive to offer agents a clean API and a seamless experience.

如果Instacart拥抱智能体而亚马逊屏蔽它们，那么智能体就会越来越多地把需求导向Instacart。

If Instacart embraces agents while Amazon blocks them, the agent will increasingly route demand to Instacart.

消费者会很高兴，亚马逊最终将面临两难：是继续保护客户关系，还是提供同样的价值主张。

Customers will be happy and Amazon will eventually face the dilemma: keep protecting the relationship or match the better value proposition.

这将成为智能体经济中的核心矛盾——聚合商意识到自己正在被聚合。

This will be the defining tension of the agent economy— aggregators realize they are now being aggregated.

当然，亚马逊会推广自己的助手，但一个垂直的亚马逊智能体很难与一个已经了解我的邮件、日历、偏好、记忆和整个生活背景的横向智能体竞争。

Amazon will push its own assistant, of course, but a vertical Amazon agent will struggle against a horizontal agent that already knows my email, calendar, preferences, memories, and entire life context.

很多公司都想成为那个聚合层。

Many companies will want to become the aggregation layer.

但实际上很少有人能做到。

Very few actually can.

很多人也指出，对亚马逊来说，基本上别无选择。

Many also pointed out that for Amazon, there's basically no choice here.

Tom Goodwin写道：很多人不明白，亚马逊的客户不是消费者。

Tom Goodwin wrote: A lot of people don't get that Amazon's customer isn't the consumer.

他们向供应商和卖家收费，收取上架费和广告费。

They sell to suppliers and vendors to charge them listing fees and advertising.

这往往更像是敲诈勒索，而不是客户营销。

It's often closer to a shakedown than customer marketing.

零售利润率可能只有0到3%。

Retail margins could be 0 to 3%.

广告利润率是70%。

Ad margins are 70%.

说他们想要智能代理，这简直可笑。

The idea they want agents is laughable.

网站看起来总是很糟糕，原因是他们根本不在乎你能不能更方便、更快或更愉快地购物。

The reason the site always looks awful is that they don't care about you buying things easier or faster or being more happy.

他们想利用你的困惑来赚钱。

They want to monetize your confusion.

希拉·莫诺特同意，写道：作为消费者我讨厌这样，但对亚马逊来说这是正确的举动。

She'll Monot agrees, writing, I hate it as a consumer, but it's the right move for Amazon.

智能代理想要取代店面并拉平竞争环境，但亚马逊足够庞大，可以拒绝。

Agents want to replace the storefront and level the playing field, but Amazon is big enough to say no.

拥有客户关系，在我看来，甚至比物流更重要。

Owning the customer relationship matters even more than logistics, in my opinion.

看看事情如何发展会很有趣。

Will be interesting to see it play out.

人们肯定意识到，这是一次开场的试探。

What people are definitely understanding is that this is an opening salvo.

A16z的安吉拉·斯特兰奇写道：这仅仅是智能代理和数据大战的开始。

A16z's Angela Strange writes, This is just the beginning of the agent and data wars.

亚马逊切断了Muse的访问。

Amazon cuts off Muse.

每个费心整合用户的平台都在想办法避免变成“管道工”。

Every platform who has done the hard work of aggregating users is trying to figure out how not to become dumb pipes.

他们是选择一，尝试封锁智能代理，但冒着惹恼客户的风险，只支持自己的智能代理体验？

Do they 1, try to block the agents but risk pissing off their customers in favor of only their own agentic experience?

选择二，达成业务发展协议，至少从智能代理交互的新模式中赚取收益？

2, Strike business development deals to at least extract dollars from the new mode of agentic interaction?

这种模式能访问他们的数据和平台吗？

That accesses their data and platforms?

选择三，还是别的什么，问号？

3, Something else, question mark?

Signal认为答案是选择二，某种形式的业务发展协议。

Signal thinks the answer is number 2, some form of BD deal.

他们写道：Meta和亚马逊之争可能的结局，是某种收入分成协议，Meta通过Muse支付费用给亚马逊以获取代理访问权。

They write, The likely resolution to the Meta and Amazon fight is some kind of revenue-sharing agreement where Meta pays Amazon for agent access through Muse.

毫无疑问，业务发展团队正在积极洽谈协议。

There is no doubt the BD teams are cooking here on an agreement.

Facebook很可能愿意承受短期成本，让Muse初期变得有用。

Facebook likely is fine eating the short-term cost to make Muse relevant initially.

这可能会成为智能代理与大型平台互动的首个真正商业模式，它们为访问服务付费。

That may become the first real business model for agents interacting with large platforms where they pay for access to the service.

可能共享数据，也可能按用户或按代理收费。

Potentially share data, and maybe pay a per-user or per-agent fee.

但如果每个主流平台都开始向智能代理收取访问费，那只有规模巨大的公司才能负担得起构建真正通用的智能代理，这会给已经庞大的公司带来巨大的护城河。

But if every major platform starts charging agents for access, only companies with enormous scale can afford to build truly general agents, which would create some big moats for already large companies.

约瑟夫·卡尔森指出，这不是亚马逊可以拖延的事情。

And Joseph Carlson points out that this isn't something that Amazon can just kick the can down the road on.

他补充说，亚马逊必须应对Muse。

He adds, Amazon is gonna have to deal with Muse.

没有别的办法。

There's no way around it.

智能代理不会消失。

Agents are not going away.

Amazon 不能闭上眼睛，假装它们不存在。

Amazon can't close their eyes and pretend they don't exist.

今天是 Muse，接下来是 ChatGPT 的 agent，然后是 Anthropic 的。

Today it's Muse, next it's ChatGPT's agent, then Anthropic's.

用户会大量采用这些。

Users will adopt these in huge numbers.

Amazon 将被迫开发一个经过验证和批准的流程，让 agent 替顾客购物。

Amazon will be forced to develop a verified and approved process for allowing agents to shop for customers.

然而，Paglio Labs 的 Matt Slotnick 说，我不会现在对 Amazon 关于 Muse 的姿态过度解读。

And yet Paglio Labs' Matt Slotnick says, I wouldn't overly read into Amazon's posture regarding Muse right now.

Shopify 显然会和 Meta 好好配合，成为先行者。

Shopify obviously will play nice with Meta and be a first mover.

Amazon 有更多的利害关系，也会对这层关系要求更高，因为他们有筹码。

Amazon has more at stake and will be more demanding about the relationship because they have leverage.

Amazon 并不反 agent，也不会因为他们在上线第一周就封杀了 Muse 就完蛋。

Amazon is not anti-agent, nor are they dead because they blocked Muse in the first week of availability.

他们不傻，而且很有话语权。

They're not dumb and they have weight to throw around.

而事实也正如预言一般，Shopify 确实冲出来扮演了这里的反 Amazon 角色。

And indeed, almost prophetically, Shopify did jump in to be the anti-Amazon here.

周一，该公司宣布与 Meta 合作，正式支持 Muse。

On Monday, the company announced a partnership with Meta to officially support Muse.

与和 OpenAI 的合作类似，Shopify 将允许 Muse 直接访问他们的后端，在优化流量的同时带来更好的搜索性能。

Similar to their partnership with OpenAI, Shopify will allow Muse to access their backend directly, giving much better search performance while optimizing traffic.

Muse 还将在 Shopify 的 agentic checkout——Shop Pay——中获得官方支持，适用于所有 Shopify 商店。

Muse will also see official support in Shopify's agentic checkout, Shop Pay, across all Shopify stores.

这些公告显然就是要故意刺激一下 Amazon。

The announcements are very clearly meant to be a poke in the eye for Amazon.

Meta 的首席 AI 官 Alexander Wang 写道，我们很高兴 Muse 能与 Shopify 深度合作，在所有 Shopify 商店中用 Shop Pay 实现 agentic checkout。

Meta's Chief AI Officer Alexander Wang wrote, we are excited for Muse to be partnering deeply with Shopify to enable agentic checkout with Shop Pay on all Shopify stores.

我们想给 Muse 用户提供大量商店的访问权限，让他们找到绝对完美的商品。

We want to give our Musers access to a wide range of stores to find the absolute perfect products.

Mark Zuckerberg 甚至也跑到 X 上插话，说和 Shopify 合作，让 Muse 里的购物和结账变得更简单。

Mark Zuckerberg even weighed in on X writing, teaming up with Shopify to make shopping and checkout easier in Muse.

Shopify 的消费者能找到更多，商家能卖得更多。

Shopify's find more, shops sell more.

更多这样的合作即将到来。

More partnerships like this coming soon.

我觉得人们很容易对此有点不以为然，会觉得虽然 Shopify 确实有机会在 agentic 购物上抢得先机，但这并不能让他们以某种方式超越 Amazon——毕竟 Amazon 是世界上最大的零售商之一。

Now I think it would be easy to be a bit dismissive of this, believing that while yeah there's a cool opportunity for Shopify to get out ahead on agentic shopping, this is not going to somehow allow them to overtake Amazon, which is one of the largest retailers in the world.

然而我仍然认为，Shopify 的重要性在几乎每一个维度上都被严重低估了。

And yet I continue to think that the significance of Shopify is wildly underestimated on just about every dimension.

就我自己和我认识的很多人而言，基本上只要不是 Amazon 上的东西，如果没有一个 Shop Pay 链接能自动调出我所有信息、不用我再登录一次，那我很可能就不会在那个商家买东西。

I know for myself and for many others, for basically anything that's not on Amazon, If there's not a Shop Pay link which can pull up all my information automatically without another login, I'm probably just not buying for that outlet.

事实上，我对于 2026 年更出格的预测之一就是，在通用 AI 普及这件事上，Shopify 会成为最重要的平台之一。我的逻辑是，现在有太多小企业创业者在用 Shopify，很多原本可能会随大流讨厌 AI 的人，会在那里围绕着他们自己小生意的各种事情上，发现这些工具到底有多有用。

In fact, one of my more out-there predictions for 2026 was Shopify being one of the most important platforms when it comes to general AI adoption, with my logic being that so many small business entrepreneurs now use Shopify that it was a place where a lot of people who might otherwise be inclined to go along with the tide of disliking AI would find out how valuable the tools were when it came to everything surrounding their own small businesses.

但在这一切背后，仍然存在一个问题，那就是 agentic 购物到底会有多重要。

Still underlying all of this is a question about how much agentic shopping is actually going to matter.

我一直对那种订餐和买机票的用例挺怀疑。多年来人们一直把这些当作个人 AI 的展示用例来宣传，但要说在让人们采用新工具这件事上，它们真的能产生多大影响，我持保留态度。

I've always been fairly skeptical that the sort of food ordering and airplane ticket buying use cases that people have pitched for years as their demonstrated use cases for personal AI were actually going to move the needle when it came to getting people to adopt new tools.

且不说这些事情本身并不难，或者至少，拿机票来说，要把你那些微妙的限制条件解释给 agent 听，至少和直接去航班网站上看一样麻烦。另外还有一个事实是，对很多人来说，浏览和发现本身就是购物体验价值的一部分。

In addition to those things just not being all that difficult, or at least, when it comes to something like flights, it being at least as difficult to explain all of the nuanced conditions that you have to an agent as it is to just look on the flight website itself, there's also the fact that for many people, the browsing and discovery is part of the value of a shopping experience.

当然，我们不必把购物看成铁板一块。如果有人对我说，即使是最爱购物的消费者，也会有一些他们完全不在乎、基本上觉得在浪费时间的购物体验，我大概会同意。

Now of course we don't have to view shopping as a monolith, and if someone argued to me that even for the most excited shoppers, there are going to be types of shopping experiences that they don't care at all about and are basically time-wasting for them, I would probably agree.

但我肯定不是唯一一个对 agentic 购物有所怀疑的人。

But I'm certainly not the only one to have some amount of skepticism around agentic shopping.

Ron Johnson 是 Apple 的零售专家，他最近在采访中说，AI 是一种会改善在线购物体验的新技术，但我不确定它是否会改变我们的购物方式。

Ron Johnson, Apple's retail guru, recently commented in an interview, AI is a new technology that will improve the online shopping experience, but I don't know that it's going to change which way we shop.

Upstart 的创始人 Jay Girard 写道，我怀疑 Muse 这类产品能否在消费者中成为主流。

Upstart founder Jay Girard wrote, I'm skeptical Muse and the like will go mainstream with consumers.

我在网上买的东西，99% 都是通过 Amazon、Shopify、Instacart 和 DoorDash 买的。

99% Of what I buy online is via Amazon, Shopify, Instacart, and DoorDash.

我不太确定一个智能代理能让这些购物变得更简单、更愉快、更自动化，或者价格明显更低。

I'm not sure an agent will make those purchases any simpler, more pleasing, more automated, or meaningfully less expensive.

餐厅和旅行也一样。

Same with restaurants and travel.

我的看法是，面向消费者的智能代理的赢家，会是那些跟主流设备绑定的公司，也就是 Apple 和 Google，它们能让你日常生活中的几百个小瞬间变得更轻松。

My view is that the winners in consumer agents will be those tied to the dominant devices, i.e., Apple and Google, who can make hundreds of small moments in your day easier.

AppLovin 的 CEO Adam Foroughi 说，现实情况是，一部分人会用像智能代理这样的东西来优化某些购物行为。

AppLovin CEO Adam Foroughi says, the reality is, part of the world will start using things like agents to optimize certain shopper behavior.

比如说，我可能会把我的保健品订阅放到一个智能代理里，让它每个月都帮我优化，并且准时送到。

For instance, I might put my supplement subscription into an agent and have it optimized every single month and delivered on time.

但这些发现平台不是这样的，而且普通消费者也不是那些深度玩智能代理、整天刷 Twitter、第一时间采用最新技术的人。

But these discovery platforms aren't that, and the typical shopper is not the person who's deep into agents and sitting on Twitter and adopting the latest technology.

每天仍然有大量的人在使用 Yahoo 的产品。

There's still a ton of people using Yahoo properties every single day.

普通消费者想要的是找到一个商品，然后真正去经历那个购物行为——他们想橱窗购物，想体验整个交易过程，想跟踪物流。

The typical shopper wants to find a product and wants to actually go through that shopper behavior— they want to window shop, they want to go through the transaction experience, they want to track it.

如果你事后告诉他们，嘿，一个智能代理本来可以帮你搞定这些，还能省下20%的钱，我觉得在一笔50美元的交易上，这并不重要，因为他们享受的是整个购物过程带来的快感。

And if you told them after the fact, hey, an agent could have done this for you and saved you 20%, I don't think that matters on a $50 transaction, because the dopamine hit from going through it is what they enjoy.

不过，就像 AI 领域的很多方面一样，我觉得在这个问题上我们非常需要保持认知上的谦逊。

Still, like so much of AI, this is another area where I think epistemic humility is greatly required.

尽管我对智能代理购物作为用户采用 AI 代理的转化体验有一定的怀疑，但我对这个判断的信心并不高。

Although I have some amount of skepticism around the power of agentic shopping as a conversion experience for users to adopt AI agents, my confidence in that assessment is not high.

而且说到底，无论个人代理体验是通过处理邮件、取消订阅还是购物来驱动，Muse 的成功对其他 AI 实验室来说都会是相当不容忽视的。

And push comes to shove, whatever combination of dealing with your email or unsubscribing or buying things the personal agent experience is being driven by, the success of Muse is going to be fairly unignorable for the other AI labs.

确实，The Information 报道说 OpenAI 正在开发一个跟 Grok 竞争的个人代理，并且还讨论了一个跟 Muse 竞争的产品。

Indeed, The Information reports that OpenAI is working on a personal agent to compete with Grok, and have discussed a Muse competitor as well.

AI 爆料者 Tibor Blaho 暗示我们可能很快就能看到 OpenAI 的代理，他泄露了一些跟名为 Aeon 的代理相关的平台代码。

AI leaker Tibor Blaho suggested we could be getting OpenAI's agent pretty soon, leaking some platform code associated with an agent called Aeon.

AI 新闻聚合者 Andrew Curran 认为 Aeon 可能会在本周四发布，作为 OpenAI 开发者日的重要发布之一。

AI news aggregator Andrew Curran believes Aeon could be launched this Thursday as one of the big releases for OpenAI's DevDay.

现在，OpenAI 推出专属个人代理这件事有一个有趣的维度，那就是 Codex 其实已经能完成用户想要的所有事情了。

Now, one of the interesting dimensions of OpenAI launching a dedicated personal agent is that Codex already could do everything a user would want.

它可以处理邮件、管理日程，甚至帮你购物，但正如 The Information 所说，很多 OpenAI 用户并不一定知道这些功能，而且可以说 xAI 和 Meta 在最近几周用它们各自的代理产品抢了一些风头。

It can triage emails, handle a calendar, or even shop for you, but as The Information puts it, many OpenAI users aren't necessarily aware of these features, and it's safe to say xAI and Meta have stolen some thunder with their respective agent products in recent weeks.

换句话说，对于 AI 的某些用例来说，处于绝对顶尖水平才是最重要的，但说到代理管理，你必须提供一种用户能理解的体验，而且这种体验不能只是一张白纸。

In other words, there are certain use cases for AI where being at the absolute state of the art is all that matters, but when it comes to agent management, you gotta have a user experience that people understand and that leaves them with something more than a blank page.

我觉得我们很可能很快就会听到更多关于这个话题的消息。

I think it is likely that we get a lot more on this topic very soon.

不过现在，故事就讲到这里。

For now though, that is the story.

代理大战已经打响了，今天的 AI Daily Brief 就到这里。

Agent wars have begun, and that's going to do it for today's AI Daily Brief.

感谢你的收听或收看，下次再见，peace。

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

已剔除 3 处广告（点击展开查看）

M10:44广告 · 已剔除

First of all, thank you to today's sponsors KPMG, Blitzy, Harbor, and HyperAgent.

To get an ad-free version of the show, go to patreon.com/AIDailyBrief, or you can subscribe on Apple Podcasts. And to learn more about sponsoring the show, send us a note at [email protected].

M110:19广告 · 已剔除

Learn more about what separates AI Amplifiers from everyone else at kpmg.com/us/aiamplifiers.

Blitzy deeply understands your codebase before it writes code.

Here's the first place that pays off: security in the age of AI.

Vulnerabilities don't live in isolation.

They live buried inside millions of lines of interconnected code, where patching one thing quietly breaks 3 others.

That's why surface-level scans fail.

Blitzy starts from its knowledge graph of your entire application, identifies and surfaces CVEs across the full estate, proactively recommends patches, and can execute the PR.

Each fix is grounded in how your systems connect and validate so nothing new breaks, and the knowledge graph dynamically updates, keeping you ahead of an ever-accelerating threat landscape.

One Blitzy customer resolved 21 active CVEs across 6 core microservices in 4 days.

Zero compile errors, every validation scanned clean, months of planned work fixed in less than a week.

Security remediation grounded in real architectural context at the speed of compute.

Harden your codebase at blitzy.com.

That's B-L-I-T-Z-Y dot com.

M111:29广告 · 已剔除

Maybe you think OpenAI and Anthropic can stay ahead, or perhaps Meta's open-source strategy can win out.

Whatever your view, every AI lab creates a different investment opportunity.

Harbor Capital Advisors AI Lab Ecosystem ETF Suite lets you invest in the ecosystem behind the AI lab you believe in.

Search Harbor AI Lab Ecosystem ETFs wherever you invest or follow @HarborCapital on X to learn more.

Visit harborcapital.com for a prospectus containing investment objectives, risks, fees, expenses, and other important information.

Read and consider it carefully before investing.

Risks include principal loss and artificial intelligence-related risks.

Harbor ETFs are distributed by Foresight Fund Services, LLC.

Harbor is not affiliated with AI Daily Brief, and the funds are not affiliated with, sponsored by, or endorsed by any AI lab.

This is a paid advertisement and not personalized investment advice.

Investing involves risk, including possible loss of principal.

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

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

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

Marketing agents turn competitor moves into landing pages.

Sales agents enrich leads, draft emails, and update the CRM.

Ops agent chases the paperwork and tracks the budget.

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

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

Hire yours at HyperAgent.

Get $100 in credits at hyperagent.com/AIDailyBrief.