旧软件护城河已死——如何识别真正胜出的企业AI公司

这就像给了别人一台最好的相机，就称他为摄影师。

It's like handing someone the best camera and calling them a photographer.

我们绝不会这么做的。

We would never do that.

所以，把这么强大的技术交给人，然后突然就说他们是AI专家，或者认为这套系统能一夜之间彻底改变我们的商业模式，我觉得这完全不对。

So handing someone this amazingly powerful technology and all of a sudden saying they are an AI expert or that this system is going to go, you know, automatically overnight change how we do business, I think is is absolutely incorrect.

这是钨自动化公司首席人工智能官亚当·菲尔德对为什么即使给了企业世界上最强的AI工具也不够的解释。

That was Adam Field, Chief AI Officer at Tungsten Automation, on why handing companies the most powerful AI tools in the world still isn't enough.

钨公司服务着超过两万五千家机构，其中包括百分之四十的财富一百强企业。

Tungsten serves over 25,000 organizations, including 40% of the Fortune one hundred.

亚当几十年来一直在观察企业如何成功或失败于这种转型。

And Adam has spent decades watching enterprises succeed and fail at exactly this kind of transformation.

我是富达分析师沃伦。

I'm Motley Fool analyst, Warren.

在这次对话中，亚当和我深入探讨了真正区分那些从AI中获得回报的公司与那些浪费预算的公司的关键点，包括一个出人意料的简单信号——你只要观察他们的招聘方式，就能分辨出来。

In this conversation, Adam and I dig into what's really separating the companies getting ROI from AI from the ones burning the budget on it, including a surprisingly simple signal you can use to tell them apart just by watching their hiring.

希望你们喜欢。

We hope you enjoy.

当我们谈论人工智能背后的巨额资本投入时，大家几乎总是第一时间想到芯片和基础模型。

When we talk about the massive capital expenditures surrounding artificial intelligence, The conversation almost always defaults to microchips and raw foundational models.

但对于那些支撑全球经济的大型企业来说，真正的竞争优势并不仅仅是租用一个模型。

But for the massive enterprises that power the global economy, the true competitive advantage isn't just about renting a model.

真正的关键在于自动化那些支撑企业运转的数十亿复杂流程、交易和文档。

It's about automating the billions of complex workflows and transactions and documents that keep those businesses running.

今天加入我们的是来自Tungsten Automation的首席人工智能官亚当·菲尔德。

Joining us today is Adam Field, Chief AI Officer at Tungsten Automation.

亚当在软件自动化和人工智能领域拥有数十年的深厚经验，长期主导这家公司在数字工作流转型领域长达四十年的全球产品愿景与企业级AI战略。

Adam brings decades of deep expertise in software automation and AI, leading global product vision and enterprise wide AI strategy for a company that has been a giant in digital workflow transformation for four decades.

Tungsten Automation服务着超过两万五千家全球企业，其中包括百分之四十的《财富》百强公司。

Tungsten Automation serves over 25,000 global organizations, including 40% of the Fortune 100.

现在，亚当将为我们揭开：全球顶尖品牌如何将沉睡的数据转化为实际收益，为什么那些不起眼的底层基础设施才是企业软件业务真正的利润引擎，以及这一切对你的投资组合、或者你正在关注的股票意味着什么。

Now Adam is here to break down how the world's biggest brands are turning dark data into actionable revenue, why the unflashy layer of infrastructure is the real cash cow for a lot of enterprise software businesses, and what all of this means, for the stocks in your portfolio or the ones you might be watching.

亚当，欢迎来到节目。

Adam, welcome to the show.

哦，蕾切尔，谢谢你邀请我。

Oh, Rachel, thank you for having me.

能来这儿真的很开心。

It's great to be here.

我想先为今天的对话打个基础。

I really wanna lay the foundation for our conversation today.

你多年来一直主导产品愿景和软件自动化，这一点大家都知道。

You know, you've spent decades leading product vision and software automation.

对于想了解这个领域的投资者来说，传统数字自动化和如今AI驱动的工作流引擎，核心区别到底在哪里？

For investors that might be trying to understand this space, what is kind of the fundamental difference between the old school modes of digital automation, what an AI driven workflow engine can actually do today?

如果回看过去三十多年软件开发的发展，我们通常说它是非常确定性的——也就是说，你开发的是软件自动化、流程自动化。

So when you think about thirty years plus of software development, you know, we like to talk about it as being very deterministic, meaning you built software automations, process automation.

可能有人听说过RPA，也就是机器人流程自动化工具。

Some may have heard of RPA, robotic process automation bots.

你直接告诉它们该做什么，它们就照着重复一遍。

You basically told them what to do, and they went and repeated that.

人工智能和机器学习其实早就有了，但通常输入相同，输出也一模一样。

There was some there's been artificial intelligence and machine learning for a very long time, but generally the same inputs got the same outputs.

现在有了智能代理，你可以给它一个任务、一个目标，它就会自己调动手头的工具和信息去完成它。

Now what we're able to do with agents, you know, everything's called an agent nowadays is give it a task, give it an output, give it a goal, and it will use the tools and information at its disposal to go and get something done.

这才是根本性的转变。

And that's really the fundamental shift.

当然，背后还有很多细节，但核心就是这个变化。

There's obviously a lot of nuance under all that, but that's the basic shift.

我觉得现在市场特别关注芯片公司，还有那些在打造大型语言模型的公司。

I think we're in a a time where the market's very focused on, obviously, the the chip companies, the companies building the massive LLMs.

这些领域确实很吸引人，但很多人容易忽略下面那层基础设施。

And certainly, that's an exciting area, but I think there's sort of a tendency to ignore a lot of that infrastructure beneath.

所以我想问问你，从你的角度看，为什么说自动化这一层才是真正创造价值的地方？

So I'm wondering from where you sit, why is that automation layer where some of that real value is is being created?

其实你可以分成三层。

Well, you've got kinda three layers.

一层是芯片制造商，比如英伟达、AMD这些公司。

You've got the chip manufacturers, the Nvidias, the AMD's of the world.

一层是大家熟知的模型公司，像Anthropic、OpenAI，还有不少开源模型；再往上就是应用层。

You've got the model companies that we know, Anthropic OpenAI, and a bunch of open source models, and then you have the the application layer.

像英伟达这样的公司，其实两个领域都在做。

Some of those like NVIDIA are playing in in both spaces.

但说实话，对我们大多数人来说，软件自动化和模型本身其实没那么重要——就拿我公司在Tungsten的员工来说，他们每天95%的工作，用哪个模型根本无所谓。

But, you know, really where software automation the models themselves have generally, for many, I would say like 95% of what my employees at Tungsten do every single day, the model really doesn't matter.

当然，如果你在写代码，做高强度的编程，让多个智能体循环运行、修复代码，那模型就很重要了。

Now, if you're coding and you're doing heavy duty coding and you have lots of agents running in loops, fixing code, the model matters.

有些模型在这些任务上确实明显更出色。

And there's some that are just far better than others at those tasks.

但像我们在Tungsten做的很多工作，都是文档密集型的，比如处理发票，根本用不着Fable 5，也不用消耗那么多token。

But for a lot of things we do, which is document heavy workloads at Tungsten, processing an invoice doesn't require Fable five, doesn't require eating all of those tokens.

所以模型本身其实已经有点像大宗商品了。

So the the model itself becomes a bit of a commodity.

真正关键的，是你在上面搭建的东西、你融入的行业基础、还有你给它的数据——这才是真正的区别所在。

So I think it's what you build on top of it, the industry foundations that you build into it, the data that you give it, that's what the difference is, the data that you give it.

否则的话，谁都能够用到这些相同的模型。

Otherwise, everyone has access to these same models.

我觉得另一个值得看看的，是人工智能的发展历程。

I think another sort of interesting thing to look at is also the evolution of AI.

我的意思是，这几年我们已经听太多关于它的讨论了。

I mean, obviously, we've been hearing about it so much the last few years.

但就像你之前提到的，人工智能其实并不新鲜。

But as you noted earlier, AI is not new.

它现在的这个版本当然已经发生了很大变化。

Its current iteration obviously has changed a lot.

你能给我们听众简单讲讲这个演变过程吗？

Maybe you could walk our audience through that journey a bit.

拉切尔，我老觉得挺有意思，别人——不管是分析师、客户还是投资者——总会问我：你们的产品里用AI了吗？

Rachel, I always find it funny when people ask me analysts, customers, you know, investors, and they ask, are you doing AI in your product?

我就说，当然用了，我们四十多年来一直在用AI。

I say, yeah, for forty years, we've been doing AI in our product.

所以，这到底是什么意思呢？

So, you know, what does what does that really mean?

机器学习，但说实话，有点可惜。

Well, machine learning and here's what's unfortunate.

我觉得很多传统的AI技术现在都被边缘化了，大家都只盯着生成式AI最新的那些玩意儿。

Unfortunately, I think a lot of that traditional AI has sort of gotten kicked aside and everyone's focused on the latest and greatest in generative AI.

但像钨金这样的公司，我们特别引以为豪的是，给不同的任务匹配最合适的人工智能方案。

But companies like Tungsten, you know, we really pride ourselves in providing the right AI for the right job.

我觉得我们真该多想想这个问题。

And I think we need to be thinking about that a lot more.

机器学习技术依然有巨大的价值。

There's still a tremendous amount of value in machine learning technologies.

速度更快。

It's faster.

通常也更省钱。

It's often more cost effective.

更环保，因为不需要用大型显卡来运行，靠它来回答问题。

It's more environmentally friendly because it doesn't you know, you take these large GPUs to power it to answer questions.

我们认为，把传统的机器学习方法和最新的生成式AI、大语言模型结合起来，用合适的AI解决合适的问题，这才是组织能获得最大价值的地方。

So we think bringing together, you know, traditional machine learning methods and the latest in generative AI and large language models to use the right AI to solve the right problem is where organizations are gonna get the most value.

能够实时判断该用哪种技术来应对具体问题的平台，我觉得这才是最能带来突破的地方。

So platforms that can provide that and on the fly figure out what technology to use to solve the question that's asked, the problem that's in front of it, I think that's where you get your greatest win.

到2026年，我一直想改善自己的健康状况，但我发现了一件事。

In 2026, I've been trying to improve my health, but here's the thing I found.

如果你不跟踪自己的体检数据，那就等于在瞎子摸象。

If you're not tracking your blood work, you're basically flying blind.

所以我特别高兴能和Rhythm合作。

That's why I'm excited to partner with Rhythm.

Rhythm 是全球最容易使用的血液检测服务，帮你了解身体内部的变化。

Rhythm is the world's easiest blood test to help you learn what's happening inside your body.

在家操作只需两分钟，不用扎针，只需贴在手臂上一个设计简约、无痛的采集装置。

Takes about two minutes at home, no needles, just a sleek and painless collection device that sticks on your arm.

血液采集完成后，把样本放进 Rhythm 提供的包装盒里，他们还会安排上门取件。

Once your blood is collected, you put the sample in the package that Rhythm provides, and they will even arrange to have it picked up from your front door.

我一个月前做了第一次检测。

I did my first test about a month ago.

Rhythm 的检测套件寄到了我家。

The Rhythm Kit came in the mail.

我花两分钟就采好了血。

I drew my blood in about two minutes.

再简单不过了。

Couldn't be easier.

两天内结果就出来了，他们的在线平台让你轻松看懂各项指标。

The results were live in about two days, and their portal makes it easy to understand what's going on.

他们还配备了一个AI健康教练，帮你解读结果并给出建议。

They've even got an AI health coach to help explain your results and recommend actions.

自从做了检测，我的饮食习惯就变了。

Since I got my labs, I've changed the way I've eaten.

体重开始下降了，而且很容易判断像GLP-1这样的疗法是否适合你。

Pounds are starting to come off, and it's easy to determine if something like a GLP one is right for you.

我平时不太会动不动就跑医院。

I'm not one to run to the doctor.

正因如此，能在家完成检测、快速拿到结果，我才这么上瘾，第二套工具已经寄来了。

So the fact that I can do this at home and get fast results is what has me hooked, and my second kit is already on the way.

Rhythm每月只要79美元，远比传统实验室检测便宜。

Rhythm is only $79 per month, a fraction of traditional lab testing.

直接送货上门，随时可以取消订阅。

Ships right to your door, and you can cancel anytime.

现在通过 rhythm.com/fool 订购，首月立减15%，还包邮。

Right now, Rhythm is offering our listeners 15% off for your first month and free shipping at rhythm.com/fool.

这是 Rhythm，访问 1hrythmhealth.com/fool，首月享八五折并免运费。

That's Rhythm with 1hrythmhealth.com/fool for 15% off your first month and free shipping.

别再猜了。

Stop guessing.

开始检测吧，访问 rhythmhealth.com/fool。

Start testing rhythmhealth.com/fool.

我见过你用这个词来形容企业系统里那些被埋没的数据——暗数据。

I've seen you use this phrase to describe the information trapped inside enterprise systems, dark data.

对于想评估一家老牌巨头或某家公司的投资者来说，企业能否释放那些藏在合同、发票里的数据，这究竟意味着什么？

And maybe, you know, for an investor that's trying to evaluate a legacy giant or a company, how do how does a business' ability to unlock data, whether it's trapped in, you contracts, invoices, and so forth.

这种能力又如何直接影响到公司增长的走向？

How does that translate directly to, the trajectory of their growth story?

当然。

Sure.

首先，要定义暗数据，大多数分析人士认为，企业内部超过百分之八十的信息都属于所谓的暗数据。

Well, I mean, first to define dark data, most analysts say that it's 80 or more percent of information inside of an organization is, quote, unquote, dark data.

那这到底意味着什么？

So what does that mean?

在我们看来，这80%里的大部分，也就是整体的80%，都困在各种文档里。

In our world, it means most of that, 80% of that 80% is trapped inside of documents.

比如合同和年度报告这类东西。

Things like contracts and annual reports.

但近几年，'文档'这个词的含义已经扩展了，现在连邮件也算在内。

But the term document itself has even evolved over the last few years to mean things like an email.

邮件也是一种文档。

An email is a document.

它们都是非结构化的。

It's unstructured.

合同也是非结构化的。

Contracts are unstructured.

没有哪两份合同是一模一样的。

No two contracts look the same.

没有两封邮件是一模一样的。

No two emails look the same.

想想你们客服中心接进来的电话录音，这些都是未结构化的暗数据。

You think about the transcripts from phone calls coming into your contact center, that's all unstructured dark data.

所谓的暗数据，就是这些录音、合同、年报、邮件、发票都躺在你们公司里，大部分现在是数字格式，但还有一些是纸质的，或者扫描后存着备用，可既没有机器也没有人去用它们。

And what that means is the reason they call it dark data is because those transcripts are sitting there, those contracts, those annual reports, those emails, those invoices are inside of your organization in documents, mostly now digital, but some still paper stored somewhere or scanned and stored somewhere for reference later, but no machine or human are making use of it.

这正是现在不一样的地方。

And that's the difference now.

过去，机器学习主要擅长处理结构化数据。

So machine learning traditionally had made really good use of structured information.

你的客户在买什么？

What are your customers buying?

你的供应商在给你提供什么？

What are your suppliers supplying you?

他们收多少钱？

How much are they charging?

你的发票里具体有哪些项目？

What's the line items in your invoice?

我们早就开始利用这些信息了。

We've been able to take advantage of that information for a very long time.

直到最近，我们才真正能利用起这些非结构化数据，比如那些动辄几百页的合同、年度报告、数以百万计的邮件，现在能把它们拆解、整理出来。当你在服务客户、银行做投资决策，或者保险公司评估风险时，就能把过去那些结构化数据，和这些原本被忽略的非结构化信息结合起来，做出更明智的判断。

Just recently now, we've been able to take advantage of all of this dark data because we can read these highly unstructured, multi 100 page contracts, annual reports, millions of emails, and begin to break them up and put them in a place that now when you're serving your customer or you're making an investment decision as a bank or you're trying to evaluate risk as an insurance carrier, you can now look at all of that unstructured information in these documents alongside all of the information you used to look at and make better decisions.

这些模型能做到的，就是这些。

And that's what these models are capable of doing.

这才是真正让人兴奋的新突破。

So that's what's really exciting and new.

我觉得这特别有意思，因为很多人听到AI能做什么，往往只是一些模糊的概念，但其实这些应用都非常实际，我们现在就能看到实实在在的价值。

I think it's fascinating too because I think sometimes it's easy for a lot of people to hear vague conceptions of what AI can do, and I think those are really practical, you know, real world applications where we're seeing the value now.

说起来挺有意思的。

You know, it's interesting.

当然，你可以看一些指标，比如公司的现金转换周期，以及其他相关数据，来真正看清企业的增长势头到底有多快。

Obviously, you can look at metrics like a company's cash conversion cycle, as well as others to really see, you know, how how much that growth story is accelerating.

从你们在Tungsten看到的情况来看，你们服务的公司很多都是上市公司，那些大众耳熟能详的大企业，自动化文档流程能把现金周转周期缩短多少？

From what you see at Tungsten, obviously, company you serve, many publicly traded companies, many of the the large household names that a lot of people know, how drastically does something like automating document workflows shrink that cash conversion cycle?

这对这些大型企业来说，能带来多大的资本效率提升？

You know, what kind of capital efficiency does it bring about for these big corporate entities?

这么说吧。

Well, look.

比如说在银行业这样的领域。

I think in in areas let's say banking, for instance.

大家都知道，哪怕一点点流程上的卡顿，都可能让银行失去客户。

I mean, we all know that any little bit of friction will increase the odds that this bank loses a customer.

不管是个人贷款还是企业贷款，任何一点障碍都会影响体验。

Let's talk about even personal loans, commercial loans, any little bit of friction.

他们对客户了解得越深，越能实时做出决策——这一点已经被多次验证，能显著提升转化率，从而更好地吸引新客户、留住老客户。

So the more they can understand about their customer and make decisions often in real time will, I think, very greatly, I don't think, we've seen it proven multiple times over, that it increases conversion rates, you know, leading to obviously better customer acquisition and better customer retention.

但很多时候，这些机构根本做不出这些决策，因为他们还得花大量时间去审阅那些还躺在文档里的信息。

And quite often, these organizations can't make these decisions because they spend time reviewing things quite often that are still in documents.

拉切尔，我得花点时间说一句，到目前为止，我已经说了上千次‘文档’这个词了，这场对话结束前，我还会再说上千次。

And Rachel, I should spend a second to say, I've said the word document probably a thousand times so far, and I'll say it a thousand more times before the end of this conversation.

没错，文档在我们这个世界里依然无处不在，尽管我们都觉得都2026年了，文档早该被淘汰了，可它们还在。

Yes, documents are still very prevalent in our world, even though we think it's 2026 and documents should be gone, they're not.

所以，回到你提到的银行业例子，拿一笔大额商业贷款来说，背后涉及的文件多得惊人，比如房产评估、收入证明等等。

And so, you know, back to the banking example in your question, you think about a large commercial loan, there are so many documents involved in, let's say it's a property being assessed, income verification.

银行越快处理完这些材料，提取关键信息，评估风险，发现异常，甚至察觉到那些藏在段落里的重复风险，就能越快做出决定。

The quicker that a bank can process all of these things, pull the relevant information out, look at the risk, see if there's any anomalies, see if they have any overlapping risk in places because it might be buried in a paragraph somewhere.

他们决策越快，客户转化就越快，业务自然也做得更好。

The quicker they can make that decision, the quicker they can convert the customer and write better business, of course, too.

我想聊聊企业级AI实施为什么常常卡住，又在哪些地方能真正成功。

I wanna talk a bit about sort of the mechanics of of why enterprise AI implementations tend to to stall and and where they succeed.

在企业界，成功运行一个AI试点项目和把AI真正大规模落地生产之间，差距正越拉越大。

You know, there there's sort of this widening gap in the corporate world between running a successful AI pilot and deploying AI in actual scaled production.

我想听听你的看法：为什么这么多企业AI项目都跨不过试点阶段，没法真正推广开来？

I wonder, you know, if I could hear your thoughts, why do so many enterprise AI projects fail to make that leap past the pilot phase?

因为很多时候，这些试点项目都会聚焦在某个特定场景，我们经常用八二法则来说明。

Because I think quite often, these pilots will focus on a specific scenario, and we use the eighty twenty rule a lot.

对吧？

Right?

他们只关注那80%的部分。

They'll focus on the 80%.

在我们这儿，你看我从文档里提取信息有多快。

In our world, look how quickly I can extract information from a document.

没错。

Mhmm.

这种情况我们经常遇到。

And we get this a lot.

我们跟很多客户交流过，他们都有非常聪明、资金充足的AI卓越中心。

We talk to, you know, many of our customers, and they have very intelligent, highly funded AI centers of excellence.

他们可能会用Claude、ChatGPT或者大厂提供的服务，说：我丢进去一份文档，结果提取准确率达到了100%。

And they might use Claude or ChatGPT or services from the hyperscalers and say, look, I threw a document at it and I get a 100% extraction accuracy.

但当你退一步，看看那另外20%的真实情况时，现实中的文档往往是什么样呢？我们有些客户，人们在光线昏暗的地下室里，把文件和身份证歪着拍下来。

Now when you step back though and look at the other 20% and what the real world is like, how often do those documents come we have customers where, you know, people are taking photos of their documents and IDs in a darkly lit basement, you know, that are sideways.

这种情况下你怎么应对？

How do you handle that scenario?

你怎么解决这个问题？

How do you fix that?

你怎么处理这些例外情况？

How do you handle the exceptions?

就算你有完美的质量，又该怎么处理这些例外？

Even if you have perfect quality, how do you handle the exceptions?

你必须能相应地处理这些情况。

Well, you have to be able to process that accordingly.

当那些大型基础模型公司突然决定弃用某个模型，把硬件全挪去跑最新版本时，你怎么办？

How do you handle when the large foundational model companies decide on a whim that they're going to deprecate a model so that they can put the hardware to their latest and greatest model?

这时候你就得让团队重新测试所有东西。

Now you have to have that team go and start retesting everything.

这正是像Tungsten这样的公司所做的事情。

Well, that's what companies like Tungsten do.

所以在试点阶段，很多事情看起来都会进行得特别顺利。

So I think in a pilot scenario, quite often, things look like they're gonna work really well.

成本似乎也控制得很好。

Costs seem contained.

但一旦你算上维持系统所需的人力、所有测试工作，还有那些意外情况，我就有一张自己做的图表。

But when you start to factor in the human capital that's necessary to maintain them, all the testing that goes into it, the exceptions, I have a chart that I built.

我们这个行业叫智能文档处理。

You know, the industry that we're in is known as intelligent document processing.

在这个系统里，大概有125个不同的功能模块，都是成熟IDP平台能为客户提供的一整套服务。

And there's probably a 125 different little squares on that that are bits of functionality that a well formed IDP platform provides our customers.

但大多数试点项目都没考虑到这些全部的环节。

Most pilots aren't considering all of those pieces.

最后一点，我之前简单提过，就是成本问题。

And then the the last bit, as I mentioned it very quickly, is cost.

看起来算成本很简单，觉得处理一份文档就几美分的事儿。

It's easy, I think, a vacuum to look at cost and say, you know, this document to extract, it's gonna cost this many pennies.

但一旦你开始算上支持成本、处理异常情况的令牌消耗——那些不是常规情况的部分，费用立马就会失控。

But when you start, again, factoring in the support, the token usage for the the exceptions, not the rule, it really can go off the rails fast.

所以我们看到越来越多的企业不再纠结于到底是自己开发还是从像Tungsten这样的公司购买，而是先用这些平台做他们擅长的部分，再在上面加自己的功能。

So that's why we see more organizations not necessarily trying to decide if they're gonna build it themselves or buy it from someone like Tungsten, but using these platforms to do what organizations like Tungsten do well and then build on top of it.

然后用自己的AI专家团队做差异化，把数据用出新花样，让自家业务脱颖而出。

And, you know, then use their clever AI experts to go differentiate, take that data and go differentiate your organization with it.

每周二我们都会在《Security Now》聊安全话题。

Every Tuesday, we talk security on security now.

大家好。

Hi.

我是利奥·拉波特，邀请你和史蒂夫·吉布森一起聊聊AI、新型的智能对手，以及如何真正修复漏洞。

This is Leo Laporte inviting you to join me and Steve Gibson as we talk about AI, the new agentic adversary, actually fixing bugs.

我们还会谈到新版Chrome修复了数百个漏洞，以及欧盟可能要推出新的聊天控制功能。

We'll also talk about hundreds of bugs that were fixed in the newest version of Chrome, and the new chat control may come to the EU.

这真让人害怕。

That's scary.

本周所有内容尽在《Security Now》。

All of that this week on Security Now.

你可以在 twitter.tv/sn 或你常用的播客平台收听。

You'll find it at twitter.tv/sn or wherever you get your podcasts.

每周三，我们会聊人工智能、机器人，还有你家里那些智能小玩意儿。

Every Wednesday, we talk AI, robotics, and all of those smart doodads in your house.

这是智能机器。

It's intelligent machines.

你好。

Hi.

我是TWIT播客网络的主持人Leo Laporte，诚邀您本周收听我们的节目。

This is Leo Laporte, host of the TWIT Podcast Network inviting you to join us this week.

本期我们邀请了巴黎·马丁诺、杰夫·贾维斯，还有拉菲·克里科里安一起讨论。

Paris Martineau, Jeff Jarvis, and I'll talk to, Raffi Krikorian.

他是 Mozilla.org 的首席技术官，正在为开放权重的 AI 模型发声。

He's the CTO of mozilla.org, and he's making his case for open weight AI models.

我们为什么需要它们？又为什么应该支持它们？

Why do we need them, and why should we all support them?

本周《智能机器》就聊这个话题。

That's this week on intelligent machines.

你可以在 twit.tv/iam 和各大播客平台收听。

You'll find it at twit.tv/iam and wherever you get your podcasts.

我还想谈一点，AI 正在深刻改变软件行业的整体面貌。

One of the things as well I wanna talk about, obviously, AI is profoundly changing the software as a surface industry.

没错。

Yeah.

我们肯定能看见私营领域也是如此。

We we've I'm sure that's true in the private markets.

在公开市场，我们已经看到 2026 年软件行业的大滑坡。

Certainly in the public markets, we have seen the software meltdown in 2026.

我相信这种趋势还会继续下去。

I'm sure we will continue to see that.

当然，这些公司并不都一样，但现在确实是软件行业发生巨变的时刻。

Obviously not all of these businesses are created equal, but it is very much a time where software is.

在某些情况下，开发起来更容易、成本也更低。

It's easier to, cheaper to build in some cases.

我担心的是，代码本身正在变得商品化，一些通用的软件功能可能几天就能搭建起来，而不是像以前那样要花好几个季度。

And I think there's been this concern that code itself is becoming commoditized and some of the the generic software functionality, you know, could be spun up in days rather than quarters.

那这些传统的SaaS企业原有的竞争优势还剩多少？

So what happens to the competitive moats of a lot of these traditional SaaS businesses?

你对这个问题怎么看？

What are your thoughts on that?

你是在说SaaS的末日吧？

Well, you're referring to the SaaS pocalypse.

对吧？

Right?

我查过谷歌，也看过谷歌趋势的数据，大概从十八到二十四个月前开始，这个词的搜索量就一直在上升。

And I looked at Google I looked at Google trends reports, you know, starting about eighteen to twenty four months ago, that term started to increasingly get googled.

然后每次OpenAI发布大更新，搜索量就猛涨，过一阵又回落；等OpenCLaw一出来，又是一波高峰，接着再回落。

And then, you know, OpenAI will make a big release and it spikes and then it goes back down and then OpenCLaw comes out and it spikes again and then comes back down.

我对这个问题想了很多。

I've thought a lot about this.

我读了很多相关的内容。

I've read a lot.

我特意静下心来仔细思考过，虽然我们是家私营公司，但确实是一家SaaS企业。

I've sat back and, you know, thought about it because even though we're a privately held company, we're a SaaS company.

我的客户经常会很坦率地问我：你们公司能撑到我们这份合同到期吗？

And one of the questions I get very candidly from my customers is, will you guys be around by the end of the term of the contract that we're signing right now?

对吧？

Right?

这确实是个很实在的问题。

I mean, it's a honest question.

他们看到那些大型上市公司，估值一路下滑。

They see the big behemoths, publicly traded companies, their valuations going down.

这就是我的看法。

Here's what I think.

过去那种‘我是你的核心系统，你换不了我’的思维，已经过时了。

I think the old mote of I'm your system of record, therefore, you can't replace me.

那种想法早就不存在了。

That mode is gone.

现在迁移起来太简单了。

It's easy to point.

我刚订阅 Cursor 的时候，三天内就重写了我一个用了二十年的个人效率工具。

I when I first got a Cursor subscription, I rewrote a twenty year old personal productivity app of mine in three days.

最省事的部分就是告诉它：看看旧数据库，再看看新数据库，把数据迁移过去就行。

And the easiest part of it was pointing it at the database saying, look at the old database, the new database, and move my data over.

我去泡了杯茶，回来一看，已经搞定了。

I went to go make a tea, and I came back and it was done.

所以搬数据，其实挺简单的。

So moving the data, like, that's easy.

这已经算不上什么护城河了。

That's not a moat anymore.

但你觉得这些大型 SaaS 公司会因为估值下降就消失吗？

But do I think, like, these big SaaS companies are gonna go away in these reduced valuations?

你觉得背后真有什么原因，还是只是大家心里慌？

Do I think that there's, you know, a reason for it versus just, you know, people are worried?

我觉得更后者的情况更多。

I think it's more the the latter.

是这样的。

Here's the thing.

如果你能做出真正解决实际问题的优秀技术，积累了多年的经验，最重要的是能帮企业降低风险，那这些 SaaS 公司还是能挺过来的。

I think if you create great technology that solves a real problem, I think if you have many years of know how and you most importantly take risk away from an organization, then I think those SaaS companies will survive.

三十年、四十年的经验，是没法被替代的。

You can't replace thirty, forty years of know how.

举个例子，我说的转移风险，指的是我们做的是发票处理和应付账款合规业务，覆盖140个国家。

One example when I say transferring risk, we're in a business that processes invoices, accounts payable compliantly in a 140 countries.

要在140个国家都拿到运营许可，得花上几百万美元，还要好几年时间。

To get the license to do that in a 140 countries would cost millions of dollars and take years to do.

所以我们帮企业把这部分风险承担下来。

So we take that risk away from organizations.

我觉得这类SaaS应用和公司是能挺过来的。

I think those SaaS applications and those companies will survive.

我觉得另一个有意思的问题是，

I think another, kinda interesting question that comes to mind as well.

现在好多公司都说自己在利用AI、整合AI，

You know, we're at a time where a lot of companies will say that they are leveraging the power of AI, they're integrating AI.

但他们往往说得含糊其辞，

Sometimes they're very vague in what those claims are.

当然，确实有不少优质企业真正在这么做，但对投资人在公开市场上来说，有时候很难分清哪些是真价值，哪些是炒作。

Obviously, there are a lot of quality businesses that are in fact doing that, but it can be sometimes difficult, I think, for investors looking at the public markets to discern the value from the hype.

从你的角度看，你接触过很多大型企业，它们确实把这些工作流程融入了日常运营。

And from your vantage point, you work with a lot of these major enterprises that are actually implementing these workflows into their daily operations.

所以我很好奇，有哪些明显的特征或迹象，能让人判断一家公司或机构是在真正利用AI推动业务增长，而不是仅仅让AI成为财务报表上的负担？

So I'm just kind of curious, what are the hallmarks, the signs that one can look for to actually see whether a company, an institution is truly leveraging AI in a way that's actually gonna bring meaningful growth to the business rather than just, you know, weighing on the balance sheet?

是的。

Yeah.

这个话题现在特别有意思，也非常重要。

It's It's so interesting and such an important conversation right now.

为了好玩，几个月前我上了互联网档案馆，翻了看一些公司的网站，再跟几个月后的版本做对比。

For fun, a few months ago, I went in the Internet way back machine and looked at some organizations' websites and then compared it to their website just a few months later.

之前还叫机器人或者压根没名字的东西，六个月后就改叫智能代理了，但我很确定，他们的产品或实际做法根本没变。

And the same things that were called bots or nothing, you know, six months prior were called agents six months later, even though I'm very convinced nothing at all had changed in their product or, you know, what they were doing.

所以，要拨开这些炒作的迷雾真的非常困难。

So it is really, really difficult to cut through that hype.

不过，我几个月前自己造了个词，叫‘无聊的AI’。

But I I coined a term, you know, some months ago that I call boring AI.

我的意思是，你得深入那些真正打基础的公司内部去看。

And what I meant by that was I think what you gotta look under the covers in the organizations that are doing the foundational work.

所谓‘无聊的AI’，就是你不能为了用而用，以为随便部署就能出成果。

And what boring AI means is you can't just go roll out this technology for the sake of rolling it out and think you're gonna get some results.

要真正做好，得花大量时间搭建数据基础，构建智能体编排的体系。

It takes a lot of time to build that data foundation, to build that agentic orchestration foundation in order to do things right.

举个例子，就在钨星公司内部，我今年一月接手了首席AI官这个新职位，组建了一个团队。

And the example that I'll give you is even internally at Tungsten, I took on this new role as Chief AI Officer in January, and I pulled together a team.

直到现在，六个月后，我们才开始向全体员工推广AI和智能体工具。

We're just now six months later rolling out some AI and agentic tools to our entire employee base.

这六个月我们一直在打磨数据基础，打通所有系统权限，处理安全和脱敏，确保完全合规。

We spent six months getting that data foundation right, access to all of our systems, all the security, the redaction to make sure that we're in compliance.

这些事都急不得。

So those things take time.

所以你真正该关注的，不是那些营销口号和花哨术语，而是实际带来了什么成果。

So I think what you really need to look for is not the hype and the flash and all of the the words in the marketing term, but ask about the outcomes.

他们到底达成了哪些实际成果？

Like, what outcomes are they actually achieving?

如果他们的说法里没提到‘我们把某项指标降低了X%，提升了Y%，或者在某方面实现了增长’，那就要小心了。对了，你之前问过自动化有没有上限，这个问题也值得讨论。

If the end of their statement doesn't say, and we reduced this by x, we increased this by y, we grew here's the other thing too is you asked earlier about, is there a ceiling to automation?

在很多流程中，能压榨出来的节省空间其实很有限，无非就是几美分、几次点击，或者几个人力。

There may with many of these processes only be so many pennies or clicks or people that you can squeeze out of it.

但在当前这个圈子里，我们根本很少谈到这一点。

Well, we don't talk enough about at all in this world.

我们总是大谈效率。

We do talk a lot about efficiency.

却很少谈熟练度。

We don't talk about proficiency.

对于这些技术如何推动收入增长，我们讨论得也不够。

We don't talk enough about revenue generation with these technologies.

我们总在说能替代多少人，但这种思路对经营企业来说简直太糟糕了。

We talk about how many people we can replace, which is just an absolutely horrible way to go about running a business.

我们总在谈能减少多少点击量，但这些数字终究是有限的。

We talk about how many clicks we can reduce, which, you know, those are finite.

但我觉得，那些我愿意投资的公司，往往是那些真正做出更好产品、反应更快的公司。

But what I think you'll find in companies that I want to invest in are the ones that are actually creating better product and you see they're reacting faster.

他们能随着世界的变化调整策略，这在今天其实并不是坏事。

They're changing their strategy as the world changes, and that's not necessarily a bad thing in today's day and age.

他们在推动更多的收入增长。

They're driving more revenue.

有些研究甚至显示，他们反而在增加招聘，而不是裁员。

They're actually some studies have shown that they're hiring more people, not reducing their headcount.

如果你看到一家公司大力投资人工智能，同时还在招人，那就是我想搭上车的公司——我觉得他们才做对了，他们用技术来成长，而不是一味地从石头里榨出最后一滴油。

If you see a company that's making huge AI investments and hiring people, those are the companies that I wanna, you know, hook my wagon to because I think they're doing it right, and they're using the technology to grow, not just to, you know, squeeze more out of that that stone.

还有一个最后的问题。

Well and one final question.

展望未来几年，是什么让你对人工智能及其各种形态感到最担忧？

You know, looking ahead in the coming years, what makes you the most cautious about AI and its various iterations?

那你觉得最令人兴奋的是什么？

And what do you find the most exciting?

你对哪些领域特别看好？

What what areas are you most excited about?

我先说说可能不太好的一面，也就是风险，然后再讲点积极的。

So the the I'll with the maybe the negative side, the collation first, and then we'll end on something positive.

所谓的顾虑主要集中在安全问题上。

So the caution side comes to come comes around to security.

你知道吗，这些基础模型公司到底在拿我的数据做什么？

You know, what are these foundational model companies doing with my data?

现在我们看到，甚至开源社区也越来越多地被恶意行为者渗透，他们往我们每天都在用的工具里注入有害代码，速度比以往快得多。

What happens when someone we're starting to see a lot more, even the open source world is suffering from bad actors putting, you know, injecting nefarious code into tools that all of us use every day at a much more rapid rate than we've ever seen before.

我就不必多说了，你们应该都知道美国政府曾叫停了Mythos、Fable和Thropic这些项目，因为它们被怀疑可能属于网络武器。

I don't need to tell probably all the viewers about what happened with Mythos and Fable and Thropic in the US government and that being put on hold because it was deemed maybe a cyber weapon.

这些才是真正让人担忧的地方。

So those are the things you worry about.

你知道吗，各国都在怎么投入资源？这对国家安全的未来意味着什么？

Like, you know, how are countries investing and what does this mean for the future of national security?

这些才是真正让人担忧的大问题。

You know, those are the things that are obviously majorly concerning.

正面来看，我最近听了一位教育领域的专家讲话，他有几十年的经验。

On the positive side, I was listening to a speaker recently who's an expert in had decades of expertise in education.

他谈到世界上一些贫困地区的人，根本得不到优质的教育。

And he was talking about people in impoverished parts of the world who lack access to really good education.

现在有了Cachypti、Claude这些工具，情况就不一样了。

Now being able to get it with the power of cachypti, Claude, and some of these other tools.

你可以用这些工具把一位老师的能力成倍放大，帮助那些缺乏教育机会的人学习。

Because what you could do is you could take a teacher and multiply him or her exponentially with these tools to allow them to educate people who lack proper education.

这真的让我特别受鼓舞。

That's totally inspiring to me.

看到人们找到新方法去学习、去成长。

Learning new ways to do things and learn.

每天用这些工具，我都能学到新东西，要是靠以前，可能得花十倍的时间去翻书或者找网站才行。

I learn something new every single day with these tools that may have taken 10 times as long, and I would have had to go find the right book or website or something.

正因如此，我才会一直坚持下去，我觉得这种潜力真的让人振奋。

So that that's what keeps me going, and I think that that potential is inspiring.

是的。

Yeah.

太有意思了。

Fascinating.

随着人工智能的发展，确实有很多值得期待的地方。

Well, I I think there is a lot to be excited about, as the world of AI progresses.

非常感谢你今天抽出时间，Adam，能来参加这个对话。

And thank you so much, Adam, for your time today and, for joining me.

Rachel，能和你聊我很开心，也感谢所有在线收听的观众。

Oh, Rachel, it's been my pleasure, and, thank you to all the viewers for for signing in and have a listen.

一如既往，节目中提到的股票，嘉宾可能有个人持仓，而The Motley Fool也可能已发布正式的买入或卖出建议。

As always, people on the program may have interests in the stocks they talk about, and The Motley Fool may have formal recommendations for or against.

所以别光听别人说就去买或卖股票。

So don't buy or sell stocks based solely on what you hear.

所有个人理财内容都遵循富达的编辑标准，且未经广告商审核。

All personal finance content follows Motley Fool editorial standards and is not approved by advertisers.

广告属于赞助内容，仅作信息参考之用。

Advertisements are sponsored content and provided for informational purposes only.

如需查看我们的完整广告披露，请参阅节目说明。

To see our full advertising disclosure, please check out our show notes.

说明。

Notes.

我是拉切尔·沃伦，代表富达隐藏宝石投资团队。

For the Motley Fool hidden gems investing team, I'm Rachel Warren.

感谢收听。

Thanks for listening.

下次再见。

We'll see you next time.