41 Stats That Tell the Story of AI Right Now — 文稿

M10:00

今天的 AI Daily Brief，我们来聊大概 41 个数据，看看 AI 现在到底讲了个什么故事。

Today on the AI Daily Brief, 41-ish stats that tell the story of AI right now.

AI Daily Brief 是一个关于 AI 里最重要新闻和讨论的日更播客和视频节目。

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

好了，朋友们，在我们开始之前，先说几条简短说明。

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

M10:34

现在，在 23 年、24 年、25 年这个 AI Daily Brief 的时代里，有一种很常见的内容，就是拿某家大型专业服务公司发布的一份新大报告，分析里面那些关于 adoption、企业使用之类的有趣数据。

Now, one very frequent type of content in the '23, '24, '25 era of the AI Daily Brief was to take some big new report from one of the big professional services firms and analyze all the interesting stats therein about adoption and corporate usage and things like that.

今年我做这类内容少多了，当然不是因为研究报告或者别的什么变少了。

I've been doing a lot less of that this year, and it's certainly not because the number of studies or anything have slowed down.

我觉得原因是，在很多情况下，调查出来的这些数据和 AI 能力的现实感受脱节得太厉害了，几乎让人觉得没什么用。

I think the reason is that in so many cases, the stats that come out of surveys just feel so disconnected from the reality of AI capability that they feel almost not useful.

这当然也是今年年初那次变化带来的一个结果：新一代模型出现了，大家开始理解 harnesses 的重要性，agentic AI 也真正上线了，于是这个世界基本上分成了两类人：一类人意识到了这种全新的工作方式，并且正努力适应它；另一类人还在沿用某种旧模式辛苦摸索。

Now, this is of course another byproduct of the shift that happened around the beginning of this year as we got a new set of models, people started to understand the importance of harnesses, and agentic AI truly came online in that the world has pretty much separated itself into those who recognize and are trying to adapt to this totally new way of working and, on the other hand, those who are still laboring under some old model.

当然，每个人、每家公司在 AI 上都有自己的路径，而且很多人以后终究会采用 agents 和 advanced AI，现在还没用也有很多充分的理由。

Now, obviously, everyone and every company has their own journey in AI, and there are plenty of good reasons why a lot of people who will eventually adopt agents and advanced AI haven't yet.

但与此同时，站在我的角度，考虑我想支持的受众，我已经越来越没精力去说服人们使用 AI 了，反而更想把这份精力花在帮助那些已经做出这个决定的人，搞清楚怎么把它用好。

But at the same time, from my standpoint, thinking about the audience that I want to support, I've had less and less energy for trying to convince people to use AI and wanted to instead spend more of that energy on helping people who had already made that decision figure out how to do it well.

于是就有了 AI Summer Adventure、Claude Camp、AgentOS，还有所有这些东西。

Thus leading to the AI Summer Adventure and Claude Camp and AgentOS and all those things.

不过，如果我们想讲完整的 AI adoption 故事，事实是，我们其实还非常非常早。

And yet, if we are trying to tell a complete story about AI adoption, the truth is that we are still very, very early.

而且在很多方面，前沿和先锋那一边的人，跟落后那一边的人之间的差距，正在变得更大，而不是更小。

And in many ways, the gap between the people on the frontier and the vanguard and those who are behind is getting wider, not smaller.

所以考虑到这些，这一期特别的 Long Read Sunday，也就是 Weekend Big Think，我想最好还是去看看那些我没怎么花时间关注的不同调查，从里面挑出一些更有意思的数据。

So with all that in mind, I thought for this particular Long Read Sunday slash Weekend Big Think episode, it would be good to go out and check in on all those different surveys that I hadn't spent as much time with to pull out some of the more interesting numbers.

所以不分先后，经过我在 Fable 和 GPT-5.6 的帮助下做的研究，这里有一些数据，讲的是 AI 现在的故事，坦白说，也是 AI Daily Brief 里我们并不总是能看到的那个故事。

So in no particular order, as researched by me with the assistance of both Fable and GPT-5.6, here are some stats telling the story of AI right now, and frankly, the story that we don't always get on the AI Daily Brief.

先说第一条，5 月中旬的一项 Gallup 民调发现，现在有 52% 的美国员工在工作中使用 AI。

First up, a Gallup poll from mid-May found that 52% of US workers now use AI on the job.

这是我们第一次正式超过半数，而且考虑到 Gallup 的调查覆盖了各种不同类型的员工，我觉得这个 52% 的数字分量很重。

It was the first time that we had officially passed the halfway mark, and given that Gallup is going so broadly across all different types of workers, I think that 52% number is carrying a lot of weight.

在我看来，这基本上说明了一点：到了这个阶段，大家在工作中就是会用 AI，问题不再是有多少 adoption，而是用得好不好。

This is basically confirming in my mind that at this point you just use AI on the job, and the questions shift not to how much adoption is there, but how well is it being used?

关于 ROI 的数据就要复杂一些了。

Numbers around ROI are a bit more complicated.

Domino Data Lab 的一项研究发现，虽然 93% 的企业报告生产能力有所提升，但其中 57% 的企业又表示，AI 的 ROI 仍然跑不赢投入。

A Domino Data Lab study found that while 93% of enterprises reported improved production capability, 57% of those enterprises said that AI's ROI still fails to outpace spend.

PwC 的年中 CEO 快照调查是在 5 月中旬到 6 月中旬之间进行的，结果显示，目前有 39% 的 CEO 表示 AI 已经带来了正面结果。

PwC's mid-year CEO snapshot, which was fielded between mid-May and mid-June, saw 39% of CEOs reporting positive outcomes from AI so far.

而且我觉得这里说的应该是那种可以衡量、看得见摸得着的结果。

And I think this is meant to be specifically measurable and tangible outcomes.

这种情况在这些调查里你会经常看到。

And this is something you see come up a lot in these surveys.

很多时候，我们正处在一个很奇怪的中间地带：公司看得到 AI 的价值，尤其是在个人层面和小团队层面；但这种价值还没有完全传导到组织层面，也没有转化成可衡量的底线收益影响。

In many cases, we're in this weird hinterland where companies know the value of AI when they see it and know the value of AI on individual and small team levels, but it hasn't translated fully to the organizational level and into measurable bottom line impact.

这个故事的另一个版本来自 KPMG Global AI Pulse Survey，这项调查是在 5 月初做的。结果显示，只有 7% 的全球领导者表示 AI 已经实现了明确的 ROI；不过，认为 AI 早已带来有意义商业价值的人，占比在一个季度里却上升了 12 个百分点，到了 76%。

Another version of this story comes from the KPMG Global AI Pulse Survey, this one taken at the beginning of May, where only 7% of global leaders reported established ROI from AI, even though the percentage who said that AI was already delivering meaningful business value had jumped 12 points in a quarter up to 76%.

所以再说一遍，不是 AI 没有价值，而是这种价值要一路传导到底线影响，仍然处在非常早期的阶段。

So again, it's not that AI isn't valuable, it's that the translation all the way to bottom line impact is still very nascent.

我们在这里经常聊到的一个故事，现在也开始出现在调查里了，那就是对成本的担忧。

One story that we talk about a lot here that is now starting to show up in surveys is concerns around cost.

早在 5 月初，EY 的一项 AI Pulse 调查里，就有 98% 的高管表示，token 成本正在迫使他们重新考虑自己的 AI 计划。

Even back at the beginning of May, in an EY AI Pulse survey, 98% of C-suite leaders indicated that token costs were forcing them to reconsider their AI plans.

这当然也是向 agentic AI 转变带来的结果。你不再是按给多少个员工开账号、每月每人 20 美元或 30 美元来想 AI，而是开始思考 intelligence 的总成本，也就是按使用的 token 来算，这样的成本会高得多，多得多。

This is, of course, a byproduct of the shift to agentic AI, where you are no longer thinking about AI in terms of number of seats that you are giving people times $20 or $30 a month, and instead thinking about the total cost of intelligence as expressed by used tokens, which can get much, much higher.

有意思的是，在那 98% 说 token 成本迫使他们重新考虑计划的人当中，真正对使用情况做计量的只有 64%，还不到三分之二。

Now, interestingly, alongside those 98% who say that token costs have forced them to reconsider their plans, only 64% of them, less than two-thirds, actually meter their usage.

M14:54

现在，再说说我们经常谈到的另一个点，也就是前沿公司的情况和落后公司的情况之间存在差距。Ramp 使用它所服务的 7 万多家企业的刷卡数据，来描绘其客户群所在企业里 AI 的实际状况。

Now, speaking to this idea of a gap between the firms on the vanguard and the firms that are behind, Ramp uses card data from 70,000+ businesses that work with it to provide a portrait of where AI is in businesses that make up their customer base.

普通 AI 使用水平和最高 AI 使用水平之间的差距，简直大得惊人。

The gap between the average AI usage and the top AI usage is phenomenal.

Ramp 发现，在处于中位数水平的购买 AI 的企业里，公司每位员工每月花费 11.38 美元。

At the median AI-buying business, Ramp found companies spending $11.38 per employee per month.

再拿它和前 1% 的企业相比，后者每个月大约花 7500 美元。

Compare that to the top 1%, which are spending about $7,500 a month.

而且考虑到 Ramp 本身已经算是一家比较先进的公司了，你看到的这些企业，本来就更可能比一般企业更偏科技前沿，这也说明实际差距很可能比这些数字显示的还要大得多。

And given that Ramp is kind of an advanced company already, you're talking about businesses that are more likely to be tech forward than just your average business, showing that this gap is probably even more enormous than this suggests.

这个 Ramp Index 也证实，在他们的企业用户采用率里，Anthropic 已经反超 OpenAI，成为领先者。

That Ramp Index also confirmed that Anthropic had flipped OpenAI to be the leader among their businesses in adoption.

在他们 7 月的支出数据里，39.5% 的用户为 OpenAI 订阅付费，而 42.4% 的用户为 Anthropic 付费。

In their July spend data, 39.5% of their users paid for OpenAI subscriptions, while 42.4% paid for Anthropic.

我觉得值得注意的是，这些数据也显示出，正如大家一直感觉到的那样，现在这确实已经是一场两家公司之间的竞争。

I think it's worth noting that this data also showed that this is, as everyone has been feeling, really a 2-company race at this point.

不过，在这个新的、更注重成本和 token 效率的时代里，这种格局会发生多大变化，会非常值得观察。

Although how much that changes in the context of this new cost-conscious token efficiency era will be really interesting to see.

这会更难追踪，但我会持续关注的一件事是，是否有任何数字能显示，有多少比例的公司正在尝试 open weights models、fine-tuning 策略，或者利用 routers 这类工具的 multi-model 策略。

It's gonna be harder to track, but one thing that I'll be keeping an eye on is any sort of numbers that indicate what percentage of companies are experimenting with things like open weights models and fine-tuning strategies or multi-model strategies that take advantage of tools like routers.

不过我们来聊聊个人用户。

Let's talk about individual people though.

现在，有一些非常明显的两极分化，再次显示出 Vanguard 和 Frontier 用户，与普通用户或者抗拒型用户之间的差距很大，而且可能还在变大。

Now, there are some pretty strong dichotomies that show again that the gap between Vanguard and Frontier users as opposed to average users or opposed to resistant users is large and potentially getting larger.

最近一份 KPMG Quarterly Pulse Survey 里有一个比较有意思的研究，他们发现，员工对 AI agents 的抗拒在一个季度内翻了四倍，从 5% 涨到了 20%。

One of the more interesting studies from a recent KPMG Quarterly Pulse Survey was that they found that employee resistance to AI agents had quadrupled in a single quarter from 5% to 20%.

这确实是那种很大的跃升。

Now that's the sort of big jump.

这听起来像是数据里的噪音，但考虑到 KPMG 每个季度都会发布 Pulse survey，这绝对是一个值得继续关注的指标。

That feels like it could be noise in the data, but given that KPMG releases their Pulse survey every quarter, that's certainly one to keep an eye on.

更重要的是，AI agents 能力更强，因此也可能更有威胁感，所以它们当然有可能在那些还没有充分利用它们的员工中，引发更多反感。

What's more, AI agents which are more capable, and thus perhaps more threatening, could certainly plausibly generate more ire among employees who aren't taking full advantage of them yet.

而在那些已经在利用它们的员工当中，agentic 时代当然已经发生了巨大变化。

And certainly among the employees who are taking advantage, the agentic era has shifted dramatically.

根据 OpenAI 在 6 月底分享的数据，超过 25% 的 Codex 用户曾经把一项任务交给 agent，而这项任务估计需要人类工作超过 8 个小时。

According to OpenAI stats that they shared at the end of June, over 25% of Codex users have handed the agent a task that would be estimated to take more than 8 hours of human work.

换句话说，被委托给 AI 的工作类型正在显著增加。

The types of work, in other words, being delegated to AI are increasing significantly.

但其中很多活动仍然是在悄悄进行。

And yet a lot of that activity is still happening in quiet.

在 Atlassian 的一项受控实验中，他们发现，公开自己使用 AI 的员工，被评价为比那些保持沉默、其他条件完全相同的同事懒 10 倍。这当然会形成很大的同辈压力，让人不太愿意谈论自己使用 AI，而这可能会显著拖慢采用率。

In a controlled experiment from Atlassian, they found that workers who disclosed their AI use were rated 10 times lazier than identical peers who stayed quiet, which is certainly a big peer pressure reason to not really talk about your AI use, which can create dramatic drag on rates of adoption.

以我的经验来看，AI 在一个组织里传播最有效的方式，是让重度用户掌握这些能力，把它们调整到公司的具体背景、需求和独特情况中，然后在这些特定使用模式的范围内，向下一批用户布道；下一批用户再把它适配到自己的用途里，然后继续向下一批人布道，如此循环。

In my experience, the most effective way for AI to disseminate across an organization is for power users to take the capabilities, adapt them to the context and needs and unique situation of the firm, and then evangelize within the bounds of those particular usage patterns to the next set of users who then adapt it to their purpose and then evangelize to the next set and so on and so forth.

但如果这些人因为使用 AI 就被不断批评、被人斜眼看待、被认为没有真正干活，那他们当然不会谈论这件事。

But if those people are all being hammered and side-eyed and seen as not really doing the work because they're using AI, of course they're not going to talk about it.

现在，还有另一个原因让一些人选择悄悄使用 AI：根据 PagerDuty 的数据，66% 的办公室专业人士使用过他们认为违反公司政策的 AI 工具。

Now, the other reason that some people are keeping their AI usage quiet is that according to PagerDuty, 66% of office professionals have used AI tools that they believed violated company policy.

当然，我的看法是，这并不是恶意使用的问题。

Now, of course, it is my contention that this is not about malicious use.

问题在于，很多情况下，人们在工作之外能用到的工具，远远好过他们在公司内部能用到的工具。

This is about the tools that people have access to outside of work being in many cases dramatically better to the tools that they have inside work.

现在，希望随着 Codex 和 Claude Code slash Cowork 在企业内部的采用率上升，这个差距会开始缩小，人们从事这种影子行为的动机也会少一些。

Now, hopefully, as Codex and Claude Code slash Cowork adoption goes up inside the enterprise, this gap starts to close and there is a little bit less of an incentive for this sort of shadow behavior.

但同样，这也是接下来几个季度需要观察的事情。

But again, that'll be something to watch over the next couple of quarters.

我们也开始看到 agentic 时代里，人们如何使用 AI 的第一批数据。

We're also starting to see the first numbers in the agentic era of how people are using AI.

有一个我觉得特别有意思的数据，来自 OpenAI 在 Frontier Report 中的研究，他们分析了超过 80 万条工作消息。

One stat that I found super interesting came from OpenAI's work at the Frontier Report, which examined more than 800,000 work messages.

他们发现，在工作场景中，与具体职业相关的 ChatGPT 使用里，有 43.5% 是用于完成不属于用户本人职业、而是属于另一个职业的任务。

They found that 43.5% of occupation-specific ChatGPT use at work is for tasks belonging to a different occupation than the user's own.

比如说，做营销的人自己去更新营销网站，而不是等软件工程师来做。

This is, for example, the marketing person making updates to the marketing site instead of waiting for the software engineers to do it.

这种边界变得模糊的现象，会在未来几年里，对工作的变化产生一些最大的影响。我觉得这也确实很符合大多数人对 AI 作为一种增强能力的技术的体验。

This sort of blurriness is going to have some of the biggest impact on how work changes over the next couple of years, and I think certainly resonates with most people's experience of AI as a capacity-augmenting technology.

不过与此同时，这种适应过程也不是没有磕磕绊绊。

At the same time, this adaptation is not without bumps.

我们几周前做过一期关于 bot sitting 的节目。

We did an episode a few weeks ago about bot sitting.

根据 BCG 在六月的一项 AI at Work 研究，他们发现，47% 的员工表示，自己花在管理和监督 AI 上的时间，比真正干活还要多。

And according to a BCG AI at Work study from June, they found 47% of workers reporting spending more time managing and supervising AI than doing actual work.

我的观点是，尤其是在向 agents 过渡的这个阶段，这种 bot sitting 的挑战，以及这种 agent 管理类型的挑战，会特别严重，直到我们把一些核心模式和最佳实践再进一步固定下来。

My contention is that especially in this transition period to agents, this bot sitting challenge and just agent management type of challenge is going to be particularly acute until we get a little bit more locked in on core patterns and best practices.

不过，这里面有一部分不只是因为这种过渡本身，而是因为实际的工作会越来越不是去亲自干活，而是去管理那些干活的 agents。

Still, part of it isn't just based on the transition, but based on the fact that the actual job increasingly will be not to do the work, but to manage the agents that do the work.

所以很讽刺的是，当 47% 的员工说自己花在管理和监督 AI 上的时间，比，原话说的，真正干活还多时，我的看法是，到了未来，在很多不同的岗位和职位里，管理和监督 AI 本身就会成为真正的工作。

So ironically, when 47% of workers are reporting spending more time managing and supervising AI than, quote, doing actual work, my contention would be that in the future, for a lot of different roles and positions, managing and supervising AI will be the actual work.

这正是我们正在走向的那个世界。

That's exactly the world that we're headed into.

M113:22

现在，我们在 2026 年之前版本的 AI Daily Brief 里，还会密切追踪另一个数字，那就是各组织会披露，他们有多少代码是 AI 写的，而不是人类写的。

Now, another number that we used to track closely in the pre-2026 version of the AI Daily Brief was organizations sharing what percentage of their code was written by AI as opposed to written by humans.

到了现在，像 OpenAI 和 Anthropic 这样的公司基本上已经接近 100% 了，所以这些数字就开始变得没那么有意思了。

At this point, given that companies like OpenAI and Anthropic are basically at 100%, these numbers kind of started to get a little less interesting.

不过，既然我们现在试着把视野从只看早期采用者，扩展到更广泛的行业，那么 DX 发现就很值得注意：在今年第二季度，500 多家工程组织中，超过 50% 的代码是 AI 生成的，而这比前一个季度的 34% 又高了不少，这意味着，软件工程实践的这种大范围转变，已经不再局限于早期采用者组织了，而是基本上已经到处都在发生。

But given that we are trying to broaden from our understanding just of early adopters to a much wider sector, it is notable that DX found that across more than 500 engineering organizations in Q2 of this year, more than 50% of code was AI generated, and that was up from 34% just a quarter earlier, meaning that the broad shift to software engineering practice is no longer confined to the early adopter organizations, but is pretty much getting everywhere at this point.

现在，说到这些人们工作方式的变化，会怎样影响组织对劳动力的看法，我们又正处在一个非常混乱、非常迷茫的阶段。

Now, when it comes to how all these changes to how people work impact how the organization thinks about labor, again, we are in a very messy and confused period.

ZipRecruiter 在六月的一项研究发现，38% 的雇主已经把基础的数据录入和处理工作，从入门级员工身上转移给了 AI。

A ZipRecruiter study from June found that 38% of employers have already shifted basic data entry and processing work away from entry-level workers onto AI.

31% 的雇主提高了入门级岗位的经验要求。

31% have raised experience requirements for entry-level roles.

关于入门级工作的潜在影响，已经有很多人忧心忡忡。

There has been a lot of hand-wringing around the potential impact of entry-level jobs.

当然，这也是 Anthropic 的 Dario 特别爱谈、也很担心的事情。

Certainly that's something that Anthropic's Dario has loved talking about his concerns around.

不过值得注意的是，同一批受访者里，也有 35% 的人预计 AI 会让总员工数增长。

And yet it's worth noting that 35% of the same sample also expect AI to grow total headcount.

还有一个也许更让人担心的点，尤其是对早期员工来说。

Another on the perhaps concerning side, especially for early employees.

根据 Resume Templates 对 1,000 名招聘经理做的一项研究，48% 的招聘经理表示，他们公司更愿意投资 AI，而不是雇用并培训一名应届毕业生。

According to a study of 1,000 hiring managers from Resume Templates, 48% of those hiring managers said that their company would rather invest in AI than hire and train a new graduate.

不过，再一次，根据 Ramp 和 Revelio Labs 结合对超过 21,000 家公司的工资和支出数据的分析，那些大量采用 AI 的公司，在采用后的 2 年里，入门级招聘增长增加了 12%。

Then again, once again, according to Ramp in combination with Revelio Labs, by looking at payroll and spend data on more than 21,000 firms, heavy AI adopters saw an increase of 12% in entry-level hiring growth in the 2 years after adoption.

换句话说，大规模 AI 支出其实和更多初级员工相关，而不是更少。

In other words, serious AI spend correlates with more junior-level employees rather than fewer.

那么，有没有可能，不想说得太直白，那些把 AI 视为裁掉初级员工手段的公司，其实只是方向搞错了，最后会发现别的公司已经掌握了正确做法，然后又会重新增加那些之前因为 AI 而理论上被他们忽视掉的岗位招聘？

Is it possible then, not to put it too bluntly, that the companies that are thinking about AI as a way to get rid of junior employees are simply doing it wrong and will eventually figure out what these other companies have and re-increase their hiring in those areas which they had now been neglecting theoretically because of AI?

和这个相关的是，虽然美国整体职位发布量下降了 7%，但 Indeed Hiring Lab 发现，自 2025 年 2 月以来，软件开发岗位发布量增长了 15%。这些数字当然并不说明开发者突然就从地球上消失了。

Related to that, while overall US job postings fell 7%, the Indeed Hiring Lab found 15% growth in software developer postings since February 2025, numbers that certainly don't indicate that all of a sudden developers are being wiped off the face of the planet.

当然，为了说完整一点，值得注意的是，这种增长里有 71% 来自高级岗位。

Now, for the sake of completeness, it is worth noting that 71% of that growth happened in senior roles.

所以，这可能还是没有真正回答初级员工的问题，但我们显然已经看到越来越多的证据表明，AI 正在毁掉软件开发的这种叙事，不只是夸大了，甚至根本就是错的。

So this still might not be addressing the question of junior-level employees, but certainly we are starting to see more and more evidence that the narrative of AI destroying software development is not only overblown, but outright wrong.

我们现在所面对的一个巨大反差是，根据 Challenger, Gray, and Christmas 的说法，AI 已经连续 5 个月成为美国裁员最常被提到的理由；但截至今年 8 月，Yale Budget Lab 在美国整体职业数据里却发现，完全没有明确的 AI 痕迹。尽管 ChatGPT 发布到现在已经快 3 年了。

One of the big contrasts that we're living with right now is that while AI has been the number one stated reason for US job cuts, according to Challenger, Gray, and Christmas, for 5 months now, as of August of this year, the Yale Budget Lab has found exactly zero evidence of clear AI fingerprints in aggregate US occupation data, despite us coming up on 3 years since ChatGPT was released.

确实，我觉得越来越多的人开始意识到，很多被归因于 AI 的裁员，其实之所以归因给 AI，只是因为把裁员甩锅给 AI 在政治上比较说得过去。

Indeed, I think that there is a growing sense that a lot of the layoffs that are attributed to AI are attributed to AI because it is a politically palatable thing to attribute layoffs to.

正如我最近说过的，我觉得接下来几个月，这个借口的说服力会越来越弱。

As I argued recently, I think that that excuse is going to carry less water in the months to come.

所以我们只能看看，接下来的数据会不会证明我是对的，还是说 AI 会继续被拿来背各种不同类型裁员的锅。

So we'll have to see if the numbers back me up on that or if AI will continue to be blamed for all manner of different types of layoffs.

说到人们对 AI 的担忧，情况其实有点五花八门。

When it comes to people's concerns about AI, the story is kind of all over the place.

一方面，Inside Higher Ed 的一项研究发现，百分之五十五的大学生预计 AI 会损害他们的职业前景，只有百分之七的人说自己完全拥抱 AI。

On the one hand, an Inside Higher Ed study found that 55% of college students expected AI to hurt their career prospects, with only 7% calling themselves all in on AI.

但在今年七月 KPMG 的 summer intern Pulse 研究里，只有百分之五的人担心工作被取代，而百分之四十三的人说，他们最担心的 AI 问题是失去批判性思维能力。

But in a KPMG summer intern Pulse study from July of this year, only 5% feared job displacement, and 43% said that their top AI worry was losing critical thinking skills.

这或许再一次反映出一种差距：一边是人们想象 AI 将来会是什么样，另一边是那些真正已经在使用 AI、并且理解它如何和真实工作世界交叉的人。

This is perhaps a gap once again between people imagining what AI is going to be like versus people who are actually using AI and understanding how it intersects with the real world of real work.

同样在那份 KPMG summer intern Pulse 里，基本上有三分之二的受访者表示，AI 已经在协助他们完成超过四分之一的任务。

Also in that KPMG summer intern pulse, basically 2/3 of those surveyed had AI assisting with over a quarter of their assignments.

从更广泛的社会议题和公众看法来看，AI 行业仍然面临明显的信任赤字。

In terms of broad societal issues and opinions, the AI industry continues to face a significant trust deficit.

根据 Anthropic 的一项研究，当然那是去年年底的数据，不过我也很难想象现在已经改善了多少，只有百分之十五的美国人信任 AI 公司来决定 AI 应该如何发展。

And according to an Anthropic study, admittedly from late last year, although I can't imagine that this has improved very much, only 15% of Americans trust AI companies to decide how AI is developed.

有意思的是，Pew Research Center 今年六月的一项研究也发现，人们的看法是中国在 AI 方面领先于美国。

Interestingly, a Pew Research Center study from June of this year also found that the perception is that China is ahead of the US in AI.

事实上，在那项 Pew 研究里，美国人以三比一的比例表示，在 AI 上更先进的是中国，而不是美国。

In fact, by 3 to 1 in that Pew study, Americans said that China, not the US, was more advanced in AI.

毫不意外，围绕 AI 建设扩张的更广泛担忧，也开始体现在数字里。

Unsurprisingly, the broader concerns around the AI buildout are also showing up in the numbers.

根据 Reuters/Ipsos 的一项民调，百分之七十七的美国人担心 AI 会让电价变得更贵，顺便说一句，Republicans 和 Democrats 的比例基本差不多；另外，百分之五十七的人会反对在自己社区里建数据中心。

According to a Reuters/Ipsos poll, 77% of Americans, basically an equal share of Republicans and Democrats, by the way, worry that AI will make electricity more expensive, and 57% would oppose a data center in their own community.

那就像我之前很多次站在我的小讲台上吐槽过的那样，担心 AI 会让电价上涨，这本来就应该是那些建设数据中心的公司必须解决的基本问题。随着反弹声音越来越大，希望这个信息已经开始被听进去了。

Now, as I've ranted about on my soapbox many times before, the worry that AI will make electricity more expensive should be a table stakes concern to address for the companies that are building data centers, and hopefully that memo is starting to resonate as this backlash gets louder.

很明显，我们还处在非常早期，还有很多工作要做。

Clearly, we're very early and there's a lot of work to be done.

在我看来，企业 AI 里最大的缺口之一，仍然是赋能和培训。

In my estimation, one of the biggest gaps in enterprise AI continues to be enablement.

有意思的是，在 European Central Bank 的一项官方调查中，ECB 发现，大约百分之五十的公司计划投资培训现有员工使用 AI，相比之下，只有百分之十二的公司计划招聘 AI 专家。

Interestingly, in an official European Central Bank survey, the ECB found that just about 50% of firms planned to invest in training their current staff for AI versus only 12% that were planning to hire AI specialists.

不过，我也不太清楚他们打算怎么培训这些人，因为市场在提供好解决方案这件事上可以说完全失败了。

Now, I don't exactly know how they're going to train those folks given the market's utter failure to provide good solutions for that.

如果他们和美国同行差不多，那他们就会做高度定制化的解决方案，因为目前市面上基本也就只有这个选择。

If they're anything like their American counterparts, they will build highly bespoke solutions because that's pretty much all that's currently available.

而在这些数字里，你开始能看到一些更激进变化即将到来的苗头。

And in some of these numbers, you start to see the inklings of more radical changes to come.

在 AI 购物方面，Adobe 发现，在今年的 Prime Day，被 AI 引导过来的购物者，转化率比不是由 AI 引导来的购物者高出百分之四十。

On AI shopping, Adobe found that on this year's Prime Day, AI-referred shoppers had a 40% better conversion than non-AI-referred shoppers.

而且很显然，电子商务感觉会是一个迟早被新的消费者行为模式彻底颠覆的领域。

And certainly it feels like e-commerce is an area that is going to be completely upended by new consumer patterns sooner rather than later.

公司内部完成的工作和第三方提供的工作之间的关系，也很可能会发生一些剧烈变化。

The relationship between work that happens in-house versus from third parties is also likely to undergo some drastic changes.

比如 Axiom 在一项针对五百二十八位企业内部法律负责人进行的调查中发现，其中百分之九十二的人要么预计会、要么已经在和外部律师谈判与 AI 相关的费用削减。

Axiom, for example, in a survey of 528 in-house legal leaders, found that 92% of them either expect or are already negotiating AI-related cuts from outside counsel.

这并不是只会出现在法律服务里的现象，它也会成为专业服务行业故事的一部分。

Now, that is a phenomenon that is not just going to be unique to legal services, but will be a part of the professional services story as well.

我不认为这件事会简单到就是专业服务被内部替代方案黑掉或者取代，但我确实认为，内部团队和外部机构彼此合作的方式会发生巨大变化。

I don't think it's as clear-cut as professional services being hacked for internal alternatives, but the way that internal and external firms work with one another is going to shift, I think, dramatically.

从社会层面来看，接下来会变得更怪。Elon University 捕捉到了一些奇怪的时代氛围，也就是人们与 AI 的个人体验正在发生变化。

On the social side of it, it gets weirder from here, Elon University captured some of the weird zeitgeist of things that are shifting around people's personal experiences with AI.

在五月的一项研究中，他们发现，美国成年互联网用户中，现在有百分之二十七的人和 AI 有过某种社交或情感互动，其中百分之三十一的人把 AI 称为朋友。

In a study from May, they found that 27% of US adult internet users now had some social or emotional interactions with AI, with 31% of them calling it a friend.

很有意思的是，同一批用户里有百分之七十四的人预测，AI 会加深社会的孤独感。要是你想找点乐观的东西，那至少这算是一种自我意识，可能会带来一些不同的前进路径。

Fascinatingly, 74% of those same users predict that AI will deepen society's loneliness, which, if you're looking for optimism, is at least a sort of self-awareness that could create some alternative paths forward.

最后，看在老天的份上，这件事我们一定要做好。

And finally, and for the love of God, let's get this one right.

根据 Common Sense Media 的数据，九到十七岁的孩子里，百分之八十六已经在使用 AI，而且其中超过百分之四十的人说，从来没有父母和他们讨论过 AI 安全。

According to Common Sense Media, 86% of kids ages 9 to 17 use AI already, and over 40% of them say that no parent has ever discussed AI safety with them.

我的看法是，我们现在正在处理的许多最严重的社会问题，都源自年轻人使用社交媒体和互联网所带来的负面影响，尤其是围绕 Gen Z 的问题。

It is my contention that many of the most serious societal issues that we are dealing with right now stem from negative aspects of social media and internet use among young people, particularly around Gen Z.

他们之所以会被互联网和社交媒体带来的那些负面后果狠狠冲击，部分原因就在于，他们的父母大多在互联网真正普及之前，甚至在社交媒体真正普及之前，就已经是成年人、已经在工作了，所以他们其实根本没有一个足够的理解基础，去真正和孩子好好沟通，帮助他们应对某种最终会对社交关系、心理健康，以及很多别的方面都产生巨大影响的东西。

Part of the reason that they got hammered with all these negative consequences of the internet and social media is that their parents, who were mostly full-grown and in jobs before the internet fully came online, and certainly before social media came online, just didn't really have a basis of understanding to be able to properly engage with their kids and help them navigate something which would ultimately have huge impacts on social interactions and mental health and so much else.

我不认为我们有能力在 AI 这件事上再犯同样的错误，我觉得现在就是父母和孩子一起坐下来，共同去学习、去理解它的时候，而不是父母只是默认这该由孩子自己去弄明白。

I do not believe that we can afford to make that same mistake with AI, and I think that now is the time for parents and kids to get together and try to learn and understand this together rather than parents just assuming it's something for kids to figure out.

顺便说一句，这也是让我特别沮丧的一类事情：那些反 AI 的人试图装作这只是一个能被重新关回瓶子里的精灵。

This, by the way, is also the type of thing that makes me so frustrated when anti-AI advocates try to pretend like this is a genie that can be put back in the bottle.

他们没有认真面对 AI 已经到来、而且它会一直在这里这一现实，反而把所有精力都花在了把牙膏塞回牙膏管里，而不是去帮助身边的人一边适应这个新世界，一边参与塑造它。

By not engaging seriously with the reality that AI is here and it is here to stay, they're spending all their energy trying to put toothpaste back in the tube rather than help the people around them both adapt to the new world and have a hand in shaping it.

你们大概也能看出来，在我看来，AI 相关的世界不同部分之间的差距，真的正处在一个很奇怪、很分化的状态。

As you can probably see, it really does just feel to me like the gap between different parts of the world when it comes to AI is in a strangely divergent place.

在这件事上，我想留给你们一个想法。

And on that front, I will leave you with this thought.

你们这些在收听像 AI Daily Brief 这样节目的人，我相信，正是弥合这种差距时最关键的参与者。

You all who are listeners of a show like the AI Daily Brief are, I believe, the most critical actors when it comes to this gap.

真到关键时刻，真正塑造大众舆论的，不会是像我这样的人——我们整天、天天都在用 AI、聊 AI、围绕 AI 创作内容、用 AI 创作内容。

When push comes to shove, it is not going to be people like me who spend all day, every day using AI, talking about AI, creating content around AI, creating content with AI, who will shape broad public opinion.

而会是那些在自己正常的、非 AI 的生活背景下，努力去采用 AI、适应 AI 的人，他们会成为其他所有人的翻译者。

It will be the folks who are doing the hard work of adopting and adapting to AI, but from within the context of their normal non-AI life, who will be the translators for everyone else.

换句话说，我是在和你们交流，但你们是在和其他所有人交流。

In other words, my conversation is with you all, but your conversations are with everyone else.

而且，虽然我很感激你们有这么多人，但其他所有人的数量，还是要多得多。

And as grateful as I am that there are as many of you all as there are, there are still a whole lot more of everyone else.

所以换句话说，你们肩上的任务非常重要，也非常重大。

So in other words, you have a huge and important job.

一点压力都没有。

No pressure at all.

总之，今天的 AI Daily Brief 就到这里。

In any case, that is gonna do it for today's AI Daily Brief.

一如既往，感谢你们的收听或观看。

Appreciate you listening or watching as always.

我们下次再见，保重。

And until next time, peace.

保重。

Peace.

已剔除 3 处广告（点击展开查看）

M10:18广告 · 已剔除

First of all, thank you to today's sponsors, Blitzy, Section, Robots and Pencils, and HyperAgent.

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

And if you want to learn more about sponsoring the show, send us a note at [email protected].

M14:42广告 · 已剔除

For anyone who is trying to figure out how to think about tokens and costs, I'll point you back to last week's Long Read Sunday episode with Nufar, "Everything You Need to Know About AI Tokens," which goes deep on all of this.

M110:23广告 · 已剔除

Every AI coding tool on the market does the same thing first.

It starts writing code.

Blitzy does the opposite.

Before writing a single line, Blitzy spends days reverse engineering your entire codebase.

Thousands of agents ingest millions of lines, mapping every dependency, every undocumented constraint, every architectural decision made over the last decade.

The result is a dynamic knowledge graph that understands your software the way a principal engineer would after 30 years in the building.

Other tools guess at context with grep searches and markdown files.

Blitzy never guesses.

It builds true understanding first, then delivers over 80% of entire software epics autonomously.

Validated, end-to-end tested, production-grade pull requests.

That's why Fortune 500 engineering teams trust Blitzy with the codebases that matter most.

See for yourself at blitzy.com.

That's blitzy.com.

Here's a harsh truth.

Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized.

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

Most employees are still using AI to summarize meeting notes.

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

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

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.

The result?

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

Your employees move from meeting summaries to solving actual business problems.

And you can prove the ROI.

Stop guessing if your AI investment is working.

Check out Section at sectionai.com.

That's S-E-C-T-I-O-N-A-I.com.

I cover the capability gap between AI potential and AI reality every day on this show.

Most companies are still figuring out how to start.

Robots and Pencils is already launching and scaling.

Agentic and generative AI in production at large enterprises in weeks.

AWS Advanced Tier Partner more than doubled in a year.

And they're hiring!

50 open roles.

If you're someone who knows this moment is different, who wants to be inside it, not watching it, this is worth a look.

At Robots and Pencils, the best ideas win, and the team is purposefully kept super high quality.

This is the kind of place you look back on as the best decision you ever made.

Take a look at robotsandpencils.com/careers.

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

New users get $1,000 in inference.

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's agent turns competitor moves into landing pages.

Sales' agent enriches leads, drafts emails, and updates 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 add agents that feel like teammates.

Build yours at HyperAgent, built by the team at Airtable.

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