访谈：西耶拉和OpenAI的布雷特·泰勒

《经济学人》。

The Economist.

你好。

Hi there.

我是Sonos人工智能客服。

I'm the Sonos AI support agent.

你可以像和别人交谈一样和我聊天。

You can talk to me just like you would with anyone else.

哦，你好。

Oh, hi.

你刚才说你是人工智能吗？

Did you say you were an AI?

是的。

Yes.

我是一个人工智能助手，专门为你提供与Sonos相关的一切帮助。

I'm an AI powered assistant here to help you with anything Sonos related.

我借了同事的Sonos无线音箱，以此为借口拨打客服热线，因为我知道接电话的不会是真人。

I've borrowed a colleague's Sonos wireless speaker as an excuse to call up its customer support line because I know it won't be a human that answers the phone.

是的。

Yeah.

不是。

No.

我需要寻求帮助。

I'm looking for for help.

所以我今天把音箱带到了办公室，想把它连接到办公室的Wi-Fi，但遇到了一些困难。

So I brought my speaker into the office today, and I'm I'd like to connect it to the office Wi Fi, and I'm having difficulties.

你能帮我吗？

So can you help?

Sonos应用里有显示任何具体的错误吗？

Are you seeing any specific errors in the Sonos app?

没有。

No.

好的。

Okay.

请稍等一下。

Give me a moment here.

像这样的AI客服正在迅速普及。

AI customer support agents like this one are spreading fast.

这并不完美。

It's not perfect.

我们经常打断对方，但这个客服听起来很自然，能够回应我提出的任何问题，无论多么无关。

We talk over each other a lot, but the agent sounds natural and is able to respond to any question I throw at it, however irrelevant.

你最喜欢的音乐是什么？

What's your favorite music?

哦，好问题。

Oh, good question.

我正在想一下。

I'm thinking about that for a sec.

我没有个人偏好，但一旦你的Sonos One设置好，我可以帮你播放任何你喜欢的音乐。

I don't have personal favorites, but I can help you play anything you like once your Sonos One is set up.

如果你还想继续排查问题，随时告诉我。

If you wanna keep troubleshooting, just let me know.

它显然经过训练，能够专注于任务。

It's clearly been trained to stay on task.

你被设定成具有什么样的性格？

What kind of personality have you been prompted to have?

我的目标是友好、直接且乐于助人，专注于通过清晰的分步指导帮你解决Sonos问题。

I'm here to be friendly, direct, and helpful, focused on getting your Sonos working with clear step by step support.

这个描述相当准确。

That's a pretty good description.

代理告诉我如何连接扬声器。

The agent tells me how to connect the speaker.

当我遇到办公室Wi-Fi问题时，它主动提出引导我进行工厂重置，或转接给真人客服。

And when I hit a wall with the office Wi Fi, offers to walk me through a factory reset or to hand me over to a human.

我不会说这是一次愉快的体验。

I wouldn't call it a delightful experience.

但如果我真的想要那样的体验，我就不会打电话给客服了。

But if I really wanted one of those, I wouldn't be on the phone to customer support.

我会去读《哈姆内特》。

I'd be reading Hamnet.

这个代理很有耐心，也很专业。

The agent is patient and professional.

也许最棒的一点是，它立刻接听了电话。

And perhaps the best thing about it is that it answered the phone immediately.

如果我正在为社会做些什么，那就是你再也不用等在线了，我觉得这是一项非常受欢迎的使命。

If there's anything I'm doing for society, it means you'll never have to wait on hold again, which I think that'll that's a very popular mission.

我必须说，作为一项使命，这比通用人工智能重要得多。

That's way more important than AGI, I have to say, as a mission.

这位有使命在身的人是布雷特·泰勒。

The man on a mission is Brett Taylor.

他是Sierra的联合创始人兼首席执行官，这是一家帮助Sonos等公司构建客户服务机器人的科技初创企业。

He's the cofounder and CEO of Sierra, a tech startup that helps companies like Sonos build customer service bots.

毫不意外，他非常相信AI代理。

Unsurprisingly, he's a big believer in AI agents.

它的价值主张非常简单。

And the value proposition is really simple.

你知道，如果这是1995年，我想我们不会在做播客，但如果你能想象一下，我会坐在这里告诉你为什么每家公司都需要一个网站，以及它将如何改变你的业务。

You know, I think if it were 1995, and I don't think we'd be podcast dude, but if you could sort of imagine for a second, I'd be sitting here and telling you why every company needs a website and how it will change your business.

我认为在2026年，每家公司都需要一个AI代理，你与客户进行的绝大多数数字互动都将通过你的代理完成，而这个代理能完成你网站能做的所有事情。

I think in 2026, every company needs an AI agent, and the vast majority of the digital interactions you'll have with your customers will be via your agent, and that agent's gonna do everything your website can do.

我是《经济学人》的管理专栏作家安德鲁·帕尔默。

I'm Andrew Palmer, management columnist at The Economist.

至于布雷特·泰勒，他远不止是一位普通的科技创业者。

And as for Brett Taylor, he's a lot more than your average tech entrepreneur.

他奠定了后来成为谷歌地图的基础。

He built the foundations for what would become Google Maps.

他曾是Facebook的首席技术官。

He was once chief technology officer of Facebook.

他曾经担任推特董事长和Salesforce的联席首席执行官。

He's been chairman of Twitter and co CEO of Salesforce.

而今天，当他不忙于经营Sierra时，他还担任着一个名为OpenAI的不太为人所知的机构的董事长。

And today, when he's not busy running Sierra, he's also the chairman of a little known outfit called OpenAI.

本集《Boss Class》特别节目是对Brett的访谈。

This bonus episode of Boss Class is an interview with Brett.

今年年初，我与他探讨了AI代理的兴起，以及这对客户服务、软件行业和人类工作的未来意味着什么，同时也探讨了这对当下意味着什么。

At the start of this year, I spoke to him about the rise of AI agents and what that means for the future of customer service, the software industry, and human jobs, but also what it means for the current moment.

为什么那些试图实施生成式AI的管理者常常感觉像在撞墙？

Why is it that the managers trying to implement Gen AI often feel like they're banging their heads against a brick wall?

当一项新技术出现时，无论是大型机、个人电脑，还是后来的互联网，都很少有现成的解决方案可以利用这项技术。

When a technology is new, whether it's the mainframe or the PC or eventually the advent of the Internet, there's not a lot of off the shelf solutions to leverage that technology.

所以，当大型机和个人电脑问世时，许多公司不得不从零开始构建一切。

So, you know, when mainframes and PCs came out, a lot of companies had to build everything from scratch.

然后，第一代软件公司基本上将构建这些软件的研发成本分摊到成千上万甚至数十万客户身上，这完全是合理的。

And then the first generation of software companies essentially amortize the research and development costs of building that software across thousands or hundreds of thousands of clients, which is just rational.

所以，你最终会看到，世界上每家公司都曾自己开发软件，然后过渡到从这些软件供应商那里授权使用软件，接着互联网出现了。

So you end up where, you know, every company in the world built their own and then you transition to, you know, licensing software from these software vendors, and then the Internet comes out.

在1997年或1998年左右，《连线》杂志上有一篇文章。

And there's an article in Wired in 1997, '98, that time period.

文章讲的是一些银行花费两千万到五千万美元来让它们的网站实现交易功能，但如果你读过这篇文章，就会发现，所谓‘交易功能’其实只是加了一个登录表单，让你能查看自己的账户信息，而不仅仅是浏览银行的介绍内容——这种功能，一个参加过编程训练营的人周末就能搞定，而现在有了Codex和Cloud Code这样的工具，你只需提个要求就能完成。

And it was about a set of banks that were spending between 20 and $50,000,000 to make their websites transactional, which if you read the article, basically, meant adding a login form so you could actually see your stuff and not just, like, information about the bank, which is something a you know, someone who goes to a coding boot camp could do in a weekend or now with tools like Codex and Cloud Code, you could do just by prompting.

但当时，仅仅为了让这个网站运行起来，就要花四千万美元的咨询费用。

But at the time, it was $40,000,000 of consulting fees to just make this website work.

整篇文章都在讨论，他们花了这么多钱，结果却并不满意。

And this whole article was about how they spent all this money and weren't happy with the outcomes.

我们现在大致正处于人工智能和AI代理的类似阶段，人人都知道代理将产生重大影响。

We're roughly in that era of AI and AI agents in particular where everyone knows that agents are gonna have a big impact.

如果你想想在供应链中引入一家新供应商，对于一家快消品公司来说，每年可能要重复这个过程几百次。

If you think about onboarding a new vendor to your supply chain, you do that, you know, hundreds of times a year if you're a consumer packaged goods company.

智能代理应该能够承担这一任务，降低成本、提高速度并增强可靠性。

An agent should be able to take that and make it lower cost, faster, more reliable.

如果你有一个免费客服电话，这是合乎逻辑的。

If you have a toll free number for customer support, it stands to reason.

你的客户能与AI代理对话，而不是等待接通或被转接到海外的外包呼叫中心，这会更好。

It's better your clients can talk to an AI agent rather than wait on hold and route to a BPO offshore somewhere.

但目前，对于许多应用场景来说，还没有现成的解决方案。

But right now, for a lot of use cases, there isn't an off the shelf solution.

因此，许多人不得不使用AI的原始组件——模型和代理开发工具包，试图将它们拼凑起来。

So you end up with a lot of people taking the raw component parts of AI, the models, and agent building tool kits, and are trying to string it together.

有些人能成功做到，有些人则不行。

Some people can do it with success, some people won't.

我们正逐步为一些最重要的应用场景和智能代理推出成熟的现成解决方案。

Slowly but surely, we're emerging with off the shelf solutions for some of the most important use cases and agents.

你知道，Sierra，我们在客户服务领域的AI代理方面是领导者。

You know, Sierra, we're the leader in AI agents for customer service.

这里在旧金山还有一家非常出色的公司，叫Harvey，专门为法律行业开发AI代理。

There's a really neat company based here in San Francisco as well called Harvey that makes AI agents for the legal profession.

所以，如果你展望四五年后，我希望对你的听众来说，针对每一个使用场景和关键部门——比如季度结账后的财务审计——都会有一个现成的代理，你可以直接购买。

So if you fast forward four or five years, I'm hopeful that for your listeners, for each of the use cases and each of the key departments, let's say, auditing your financials after your quarter close, there's gonna be an agent for that that you can just buy.

但与此同时，由于还没有这样的产品，你不得不自己构建，这就带来一个问题：你是否愿意承担这种复杂性和成本？

In the meantime, because there isn't, you'll have to build it yourself, and you'll have to go through this question of, do you want to incur that complexity and cost?

如果你选择这么做，顺便提醒一下，你最好做好准备，一旦有供应商推出产品，就立刻弃用它，因为大多数公司并不想成为软件公司。

And if you do, by the way, you should probably prep yourself here to throw it out when a vendor is available because most companies don't wanna be software companies.

对吧？

Right?

大多数公司只希望事情能顺利完成。

Most companies just want the job done.

所以我们现在还处于早期阶段，我期待五年后，这个领域会涌现出一批成熟的供应商，他们出售的是解决具体问题的代理方案，而不是只卖模型，说‘这是堆木头，自己盖房子吧’——目前很多公司正处在这样的阶段。

So we're just in the early innings, and I'm hopeful five years from now, it'll be a very mature landscape of vendors who sell agents as solutions to problems rather than people selling models and saying, here's a bunch of wood, build a house, which is kind of the case there for a lot of companies where they are today.

我想重点谈谈这个过渡期。

Just wanna zero in on this sort of interim period.

所以，论点是如果你等得足够久，生态系统就会成熟。

So the argument is if you wait long enough, the ecosystem will be there.

这里显然带有一些自身利益的成分。

There's obviously a bit of self interest here.

人们将能够从你们这样的公司购买服务。

People will be able to buy from firms like your own.

但与此同时，人们是否应该进行尝试呢？

In the meantime, though, should people be experimenting?

我的意思是，作为一种通用技术，实验、探索和培养对这项技术的直觉肯定是有好处的。

I mean, a lot of this is as a general purpose technology, there must be benefit to experimentation, to playing around, to developing an intuition of the technology.

确实如此。

There absolutely is.

这是一个非常微妙的决定。

And it's a very nuanced decision.

所以，我会从第一性原理的角度来梳理一下我的看法。

So I'll sort of walk through what I believe is sort of the first principles view.

因此，人工智能本质上具有通缩性。

So AI is inherently deflationary.

我的意思是，理想情况下，它能让你用更少的资源做更多的事。

I mean, ideally, it helps you do more with less.

所以它能带来成本节约。

So there's a cost savings aspect to it.

那么，你会怎么处理这些成本节约呢？

And what do you do with cost savings?

嗯，你可以把节省下来的钱分给股东，这很有价值。

Well, you can pass it on to your shareholders, which is valuable.

但我是个资本主义者。

But I'm a capitalist.

而且，你知道，如果你的竞争对手也能使用同样的技术，总会有人找到办法重新投资这些资金以获得竞争优势。

And, you know, if your competitors have access to the same technology, someone will find a way to reinvest that money to gain a competitive edge.

因此，等待太久是有风险的，因为如果你的竞争对手比你更善于采用这项技术，那么在你袖手旁观、他们积极行动的这段时间里，可能会累积哪些结构性优势呢？

So there's a risk to waiting too long because if you have a savvy competitor who's able to adopt this technology more than you can, What structural advantages might compound in the period where you're waiting on the sidelines and your competitor is not?

我认为这正是推动大量紧迫感的原因，许多董事会和首席执行官之所以自上而下积极推动人工智能的采用，是有充分理由的。

I think that's driving a lot of the urgency and and I think a lot of boards and CEOs are actually driving the adoption of AI top down for a good reason.

但另一方面，这还涉及消费者行为。

But the other part of it though is consumer behavior.

我认为ChatGPT正逐渐成为人工智能的入口，也就是消费者接触人工智能的主要门户。

And I think ChatGPT is really becoming the front door, you know, the sort of the consumer front door for AI.

如果你看看搜索引擎传统上所扮演的角色，很多方面正在被重新调整。

And if you look at the role that search engines have traditionally played, a lot of that is being reshuffled a bit.

对于品牌而言，如果你思考它们如何触达消费者，这一点也会随之改变。

And for brands, if you think about how they reach consumers, that changes as well.

目前人工智能领域还没有类似付费广告市场的机制。

There isn't the equivalent of a paid ads market right now with AI.

有很多人试图帮助你优化在ChatGPT中的定位，但现阶段这还远未形成一门科学。

There's a lot of people trying to help you optimize your positioning in ChatGPT, but it's not exactly a science at this point.

我认为还有一些非常有趣的问题，比如在这个新世界中，中介的角色——像经纪人这样的角色——究竟会如何变化？

And I think there's some really interesting questions, which is what are the role of middlemen in this new world, like brokers and things like that?

同样，我猜我并不是唯一一个这样做的。

Similarly, I'm probably not alone in this.

当我去看我的全科医生时，我会在见他之前把我的化验结果上传到ChatGPT，并问一些不同的问题，比如，专业知识的角色是什么？

When I see my primary care physician, I upload my lab results to chat GPT before I see him and I ask different questions like, what is the role of expertise?

你的所有病人都在这么做。

All of your patients are doing that.

比如，这会如何改变医疗、法律、金融服务？

Like, how does that change, you know, health care, the law, financial services?

所以这是一个冗长的回答，但我认为很难预测未来。

So that was a long winded answer, but I think it's very hard to predict the future.

因此，鉴于事物变化如此迅速，进行实验非常重要。

So I think it is very important to experiment, as a consequence of things shifting so quickly.

任何尝试过生成式AI的人都会知道，还有很大的改进空间。

Anybody who has experimented with generative AI will know that there's a lot of room for improvement.

幻觉仍然会发生。

Hallucinations still happen.

聊天机器人偶尔还是会失控。

Chatbots still occasionally go rogue.

安全是一个重大问题。

Security is a big concern.

尽管模型一直在不断改进，但布雷特警告说，一些问题本质上是技术固有的。

And although the models are getting better all the time, Brett warns that some problems are inherent in the technology itself.

是的。

Yeah.

模型并不完美。

The models are imperfect.

而且我认为，更具挑战性的是，模型是非确定性的。

And actually, even more challenging, I would argue, is that models are nondeterministic.

所以你连续两次输入相同的提示，可能会得到两个不同的答案。

So you can give it the same prompt two times in a row and get two different answers.

这使得测试和鲁棒性的概念变得非常困难，因为你希望能够确定模型是否永远不会这样做。

That makes testing and the concept of robustness very challenging because you wanna be able to say, will the model never do this?

我不确定是否能这么说。

I'm not sure it's possible to say that.

不过，我认为非常重要的一点是，人本身也是不完美的。

One thing I think is really important to remember though is people are imperfect as well.

所以如果你考虑金融服务领域，众所周知，理财顾问承诺回报是大忌。

So if you think in financial services, you know, one of the things that's a big no no is a financial adviser promising returns.

对吧？

Right?

这在大多数国家都是违法的，而且在所有国家都应该是违法的，因为根本无法承诺，而且显然不合适。

That's illegal in most countries and should be illegal in all countries because it's impossible to promise and obviously inappropriate.

但人们确实这么做过。

But people have done it.

这就是为什么会有这么多监管措施。

That's why there's all the regulations in place.

对吧？

Right?

事实上，对于许多银行来说，他们会录音电话通话并进行转录，同时设置控制机制来检查是否存在违规行为。

And in fact, for a lot of banks, people record phone calls that have transcription, have controls in place to inspect and say, did people do something wrong?

我认为，用这种方式来看待人工智能其实相当健康。

I think that's actually a fairly healthy way to think about AI as well.

如果你不再等待它完美无缺，而是承认它必然会有缺陷，那么我们是否已建立技术和流程控制机制，以识别其缺陷并及时纠正？

If you stop waiting for it to be perfect and say, it will be imperfect, do we have the technical and procedural controls in place to recognize when it is imperfect and remediate it?

我们是否已设置防护措施来降低这种风险？

Have we put in guardrails in place to mitigate that risk?

这可以成为一场非常有建设性的对话。

It can be a very constructive conversation.

我认为，使用场景越狭窄，设置这些防护措施就越容易。

And I think the more narrow the use case, the easier it is to put in those guardrails.

如果你是一家零售商，想让人工智能助手协助处理退货，那么这一标准操作流程是非常具体的。

So if you're a retailer and you want to have an AI agent help return an item, well, that standard operating procedure is very concrete.

我们有三十天的退货政策。

We have a thirty day return policy.

这件物品是否被穿过？

Has the item been worn?

你知道的，所有这些因素。

You know, all those things.

而且存在一些幻觉风险，等等，但你已经缩小了使用场景。

And there's some risk of hallucination, all that, but you've narrowed the use case.

所以我开玩笑说，这从一个科学问题变成了工程问题。

So I joke it goes from being a science problem to an engineering problem.

随着技术的成熟，你可以越来越广泛地进行推广。

And then as the technology matures, you can generalize more and more and more.

但对于像OpenAI这样研究通用人工智能的实验室来说，那是一个科学问题。

But, you know, for labs like OpenAI working on AGI, that's a science problem.

对于一家公司来说，如果你考虑为某个流程构建一个智能代理，能否缩小领域范围，使其变成一个工程问题？

For a company, if you think about building an agent for a process, can you narrow the domain so it becomes an engineering problem?

你可以设置控制机制，以便在出现错误时能够发现并加以纠正。

And you can put the controls in place so that when there are errors, you find them and remediate them.

所以，如果你思考一下你在Sierra客户群中遇到的边缘案例，目前的前沿在哪里？

So if you think about the edge cases in your own client base with Sierra, what is the frontier at this point?

我的意思是，如果我们和一家银行交谈，他们说，嗯，我们可以接受一些非常简单的表单填写操作。

I mean, you know, if we had a conversation with a bank which said, you know, we can get comfortable with a kind of pretty simple form filling activities.

如果你开始进入咨询领域，比如，这就会变得在声誉上过于复杂和棘手，让我们难以承受。

If you start to move into an the advisory space, for example, it just gets way too reputationally complex and problematic for us.

那么，我们在这一连续体上的位置在哪里？

So where are we on the continuum?

目前什么是可能实现的极限？

What's the art of the possible right now?

是的。

Yeah.

首先，我非常惊喜地发现，受监管行业到目前为止已经广泛采用了AI代理。

So first, I've been really pleasantly surprised how much regulated industries have adopted AI agents so far.

正如你所直觉到的，在医疗保健的背景下，有些对话比其他对话更复杂。

As you intuited, you know, there are some conversations that are more complex than others in the context of health care.

预约骨科专家的门诊相对风险较低，因为这只是安排一个预约。

Scheduling an appointment with, say, an orthopedic specialist is relatively low risk because it's scheduling an appointment.

让AI代理进行分诊诊断，决定你应该看哪位专家，则风险高得多，因为这涉及医疗决策。

Having an AI agent do a triage diagnosis of which specialist you should talk to is much higher risk because there's, you know, medical decision making there.

所以我认为，很多人正用美国棒球的比喻来说，先在一些低风险的使用场景中积累一些实战经验，这样随着技术的成熟，特别是监管环境的完善，他们就能拥有丰富的经验，逐步过渡到更敏感的使用场景。

And so I think a lot of people are saying, using the American baseball metaphor, getting some at bats with, I think, some of the lower risk use cases so that as the technology matures and in particularly the regulatory landscape matures, they have deep experience with this technology so they can evolve towards those more sensitive use cases.

我认为，实际发展速度会比人们预期的更快，因为AI虽然不完美，但比人类更稳定。

I think it will actually evolve more quickly than people expect because AI, while it is imperfect, it is actually more consistent than humans.

所以回到提供财务建议这个问题，其中存在巨大风险。

And so if you go back to the giving financial advice, there's a ton of risk.

AI可能会产生幻觉。

AI can hallucinate.

如果有人告诉你没有任何风险，那他们就是在卖假药。

There's if anyone tells you there's no risk, they're selling you snake oil.

然而，如果你观察一家大型金融服务公司与客户的所有沟通，其中有多少比例是不完美的？

However, if you look at a very large financial services firm and all of the communications they have with their clients, what percentage of those are imperfect?

这可能非常高。

It's probably very high.

我们对人类完美的期望要低得多，因此我们接受了这一点。

Our expectations of human perfection are just much lower, and so we accept that.

所以我认为AI实际上可以增强控制的稳健性。

And so I think AI can actually improve the robustness of control.

无论现在是否普遍持有这种观点，我相信在未来几年内它会成为共识。

Whether or not that's the commonly held belief now, I believe it will become that over the next few years.

我大脑的一部分还在琢磨什么是‘打席’，不过我们先放一放。

One part of my brain is still trying to work out what an at bat is, but we'll leave that.

就客户反应而言，你其实是在说，人们需要适应代理、AI以及人类会犯错这个事实。

Just in terms of customer reactions, because you're kind of describing, like, people need to adjust to the idea of agents, of AI, humans are fallible.

这样更好。

This is better.

但你的客户们知道他们在和AI对话吗？他们对此有何反应？

But to what extent do customers of your customers know that they're talking to an AI, and how do they respond to it?

它们都会。

All of them do.

所有基于Siro构建的代理都会表明自己是AI，并说：嗨。

All of the agents built on Siro identify as an AI, say, hey.

我是一个AI。

I'm an AI.

而且，实际上，大多数代理都会说，我偶尔也会犯错，因为这是一种建立信任的举措。

And, actually, most of them say, I occasionally make mistakes too because it's a trust building exercise.

对吧？

Right?

你知道的。

You know?

基于Sera构建的AI代理的客户满意度得分非常高，几乎普遍高于此前的人工交互。

And the customer satisfaction scores of the AI agents built on Sera are incredibly high and almost uniformly higher than the human interactions that preceded it.

我最喜欢的一段对话是我一位客户发给我的，来自一家电信公司：一位老年男子打电话来，因为他的电视接收器坏了，无法看电视，他和这个AI聊了三十多分钟，最后说：谢谢。

One of my the favorite conversations that one of our clients sent me was for a telecommunications company and an elderly man called because his receiver wasn't working so he couldn't watch television and spent more than thirty minutes on the phone talking to this AI and ended with, thank you.

你是个好机器人。

You've been a good robot.

这让我感到非常有趣，因为这个AI对一位很可能在其他情况下遇到不那么有耐心的人的用户表现出了极大的耐心，毕竟长时间保持通话对人类来说成本很高。

And it was fascinating to me just because the degree of patience this AI had with someone who probably in these interactions would have encountered a much less patient person on the other side just because of the inherent cost of staying on the phone with someone for that long.

而那种表达感谢的共情举动。

And just the empathetic gesture of thanking it.

他很清楚自己是在和一个机器人对话。

He was quite aware it was a bot.

我想我们会感到惊喜，因为AI代理可以真正地用你的语言交流。

I think we'll be pleasantly surprised just because AI agents can speak your language literally.

在这里加利福尼亚，有英语使用者、普通话使用者、他加禄语使用者、越南语使用者。

Here in California, we there's English speakers, there's Mandarin speakers, there's Tagalog speakers, there's Vietnamese speakers.

现在提供多语言服务实际上是免费的，而且无限耐心。

It's now free effectively to provide multilingual service, infinitely patient.

没有人会在你身后说：嘿，快挂电话吧。

There's no one behind you saying, hey, get off the phone.

你每天必须再打十个电话才能完成你的业绩指标，或者 whatever 它可能是啥。

You have to do 10 more phone calls a day to reach your quota or whatever it might be.

而且，你知道的，它还能适应一些非常独特的需求。

And, you know, can accommodate even idiosyncratic things.

这并不意味着它在所有方面都更好。

It doesn't mean that it's better for everything.

我不是这个意思，但我的联合创始人克莱·巴沃有一个很好的说法：在计算机的历史上，我们一直得学习如何使用计算机。

I don't mean to imply that, but my cofounder Clay Bavor has a great way of putting it which is, for the history of computers, we've had to learn how to use computers.

想想你第一次使用 Microsoft Excel 时，那感觉有多吓人。

Think of the first time you use Microsoft Excel and just how intimidating it was.

AI 代理会了解你，你只需要说话，它就能为你弄明白。

AI agents learn you and you just talk and it figures it out for you.

所以我认为，这是计算机历史上一次非常人性化、非常积极的演进——从打孔卡片，到键盘鼠标，到触摸屏，到现在只需说话。

So I think it's a really humane, really positive evolution in the history of computers going from punch cards to keyboards and mice to touch screens to now just speaking.

那这里的一些隐性成本呢？

What about some of the hidden costs here?

那么，当人们与你的AI代理互动时，是否有真人参与其中？

So, are there humans in the loop when people are interacting with with your AI agent?

在实际操作中，人工接管的情况有多频繁？看起来是怎样的？

And how often what does a hand off look like in practice?

在某些情况下，确实应该有真人参与。

There can be and there should be in some circumstances.

我举个房地产抵押行业的例子吧。

I'll just give you an example on, like, the mortgage industry.

如果你在再融资或购房，其中一部分是抵押贷款的实际呈现，这部分可能应该由银行职员来处理。

If you're refinancing a home or or buying a home, some of it is the actual presentation of a mortgage, which probably should be done by a banker.

但很大一部分工作是收集信息，比如你的收入、资产和信用记录，而这类任务正是AI代理可以辅助银行职员的地方，让银行职员不必花太多时间收集表格和PDF文件，而是专注于他们最擅长的工作。

But a big part of it is collecting information, like your income and assets and credit, and that's an example where an agent can just, sort of augment an experience that you have with a banker, and the banker can spend less time on, you know, collecting forms and PDFs and actually just doing what he or she does best.

我认为这应该是一个商业决策。

I would say it should just be a business decision.

你希望AI扮演一个协作者的角色，辅助专业的领域专家吗？

Do you want the AI to sort of be a copilot, if you will, to professional subject matter expert?

你希望它完全自主吗？

Do you want it to be autonomous?

同样，客户也应该拥有这种选择权。

Similarly, the client should have agency in that.

如果你想要和真人交谈，你就应该能够做到。

You know, if you wanna talk to a person, you should be able to.

这些AI代理的出色之处在于，它们表现良好，不像过去那些让人讨厌的老式机器人，大多数人都是自愿使用它们的。

The nice part about these AI agents, because they're good and they're not like the old bots that everyone hated, Most people are opting into using them.

因此，你实际上可能会获得更高品质、更昂贵的人工服务，因为你通过AI承担了大量简单的交易性工作，从而释放了大量预算。

So you could end up actually having, I'll say, higher quality, more expensive interactions with your people because you've essentially unlocked a bunch of budget by taking a lot of the simpler transactional stuff off with AI.

我们有一个客户，他们曾将客户服务外包到海外，但现在又重新转回国内，因为业务量发生了变化，他们现在负担得起，而且认为这样会提供更好的体验。

We have one client who had offshored their customer service who's now onshoring again as an example because the volumes have changed and they can afford it now and it they thought it would be a better experience.

因此，这里可能会出现一些非常有趣且反直觉的二级效应。

So there can be some really interesting counterintuitive second order effects here.

但既然模型仍有可能出错，你就必须建立某种监控机制。

But to the extent that models can still go wrong, you've got to have some method of monitoring.

你的观点是，这可以是另一个模型吗？

Are you is your contention that that can be another model?

你如何降低这种风险？

How do you mitigate that risk?

我喜欢把它看作是纵深防御，这是科技圈里常用于安全领域的术语。

I like to think of it as defense in-depth, which is a term in tech circles we usually use around security.

现在大多数公司都有首席信息安全官，而纵深防御的意思是，首先你尽力防止任何问题发生。

Most companies now have a chief information security officer, and what defense in-depth means is first you try to prevent anything from going wrong.

所以你会把所有门都锁上，但接着你会想，如果坏人真的进来了，我们能否快速发现并限制损害范围？

So you lock all the doors, but then you say, okay, if a bad actor does get in, can we detect it quickly and limit the blast radius?

因此，除了锁好所有门之外，你还要确保对每个人的笔记本电脑进行良好的监控。

So that's where you end up with, in addition to locking all the doors, you make sure there's good monitoring on everyone's laptop.

你会建立多层防护，这样即使其中一层出现了漏洞，你也能将其化解。

And you end up with all these layers so that even if one of the layers ended up with a vulnerability, you've mitigated it.

我认为AI也应当如此。

I think the same should be true of AI.

所以我认为第一层应该是AI监控AI。

So I think the first layer should be AI monitoring the AI.

在Sierra，我们使用一种称为监督模型的概念，它们实时监督底层模型的决策过程，判断这是否是幻觉。

At Sierra, we use a concept called supervisor models, and they essentially supervise the decision making of the underlying model in real time and say, you know, is this a hallucination?

它们是否遵循了标准操作流程？

Did they actually follow the standard operating procedures?

它们是否突破了任何安全护栏？

Did they break one of the guardrails?

这些都是实时进行的。

And that's in real time.

然后你可以使用一个更长期、更深入的模型，在对话结束后对其进行评估。

And then you can have a longer, more intense model, which is essentially after the conversation is done evaluating it.

这是一次低情绪的对话吗？

You know, was this a low sentiment conversation?

AI代理是否重复了太多次？

Did the AI agent repeat itself too much?

你可以结合整个对话的上下文来查看这些内容。

Things that you can sort of look at with the context of the full conversation.

然后你可以把这些对话放入队列中，供事后人工审查。

And then you can, you know, maybe put those conversations on a queue so people can review them after the fact.

这样做的好处是，不是只随机抽取一些对话样本，而是把问题对话放在最前面，也就是把针从干草堆里挑出来，让那些有问题的对话显而易见。

And what's really nice about that is rather than just looking at maybe a random sample of conversations, you put the needles at the top of the haystack, you know, so the problematic ones are there.

这里的核心理念是结合人工智能与人工干预，利用AI辅助人工，让他们不必浪费精力，而是把时间集中在最敏感、最棘手的对话上。

And the whole idea here is you use a combination of AI and humans in the loop, using AI to help those humans that are in the loop so that they're not just wasting their energy but actually spending their time on the most sensitive, the most problematic conversations.

我也想谈一谈评估的问题。

I also wanted just to talk about evaluation.

如果企业想为自己构建系统，并定义什么是高质量、什么是良好表现，这似乎让很多人感到困扰。

So if as an enterprise, you're trying to build things for yourself and trying to define what good quality is, what performance is, that seems to cause people quite a lot of trouble.

据我了解，你正在帮助企业客户应对这一过程。

You're inside organizations helping customers through that process as I understand it.

那么人们通常会遇到哪些主要问题？又该如何解决呢？

So what are the kind of big problems that people hit, and how should they resolve them?

这听起来可能过于简化了，但我认为作为企业领导者，你所能做的最重要的一件事是明确你想要实现的业务成果，而不是技术成果。

This is going to sound reductive, but I think one of the most important things you can do as a business leader is specify a business outcome you're trying to drive more than a technical outcome.

当你思考AI代理可以应用在哪些业务流程中时，这一点至关重要。

I think it's really important when you think about your business processes where AI agents can apply.

你真正关心的关键业务指标是什么？

What are the key business metrics you actually care about?

我举一个简单的例子：有多少比例的电话不需要用户与真人交流？

I'll just give you one that's simplistic, which is what percentage of calls don't people need to talk to a real person.

要达到100%其实很容易。

Well, it's easy to make it a 100%.

你只要不让人与真人交谈就行了。

You just don't let someone talk to a real person.

而你或任何消费者都知道，这种体验会让人极其沮丧。

And you or any consumer knows that can be an insanely frustrating experience.

所以这个指标很容易被操纵。

So that metric's just gameable.

单靠这个指标并不理想。

It's not a great metric on its own.

像客户满意度评分这样的指标，如果能和上述指标结合使用，会非常有效，因为这样可以兼顾消费者的情感反馈和业务成果，比如有多少客户能够自助解决问题而无需排队等待。

Things like customer satisfaction score coupled with a metric like that are really productive because you tend to get a mix of consumer sentiment plus the sort of business outcome of, you know, how many of your clients could help themselves without having to wait in a queue.

这会影响你的定价策略吗？

And that feeds into your own pricing approach?

没错。

That's right.

我们在西拉公司采用的是基于成果的定价模式，这意味着只有当AI代理真正替客户解决问题、无需升级到人工处理时，我们才会向客户收费。

We do what's called outcomes based pricing at Sierra, which means we only charge our clients when the AI agent actually solves the problem on behalf of the customer and we have to escalate to a person that's free.

我们的整个理念是，四年前，软件还只是人的生产力工具。

Our whole philosophy is that software ancient history, four years ago, software was, you know, a productivity tool for a person.

如果你问一个销售员，是你的CRM系统推动了这笔销售，还是你自己推动的？

And if you asked a salesperson, was it your CRM system or you that drove that sale?

当然，他们会把功劳归于自己。

Of course, they're gonna take credit for it.

现在，如果一个AI代理，比如SEER代理接听电话或回复WhatsApp聊天并处理客户服务请求，你就知道它是否解决了问题。

Well, now, an AI agency, if it's one of the SEER agents answering a phone call or answering a WhatsApp chat and handling a customer service inquiry, you know whether it solved the problem.

你还能知道客户满意度评分，为什么不为做得好的工作付费，而不是为使用软件的特权付费呢？

You also know the customer satisfaction score, and why not pay for a job well done rather than pay for the privilege of using the software.

所以我想最后再深入一点，谈谈你认为未来会如何发展。

So I wanna just end by opening this out a little bit more into kind of where you think things are going.

你已经描述了在未来五年左右，供应商的生态系统将会变得更加完善。

So you've described how in five years, say, the ecosystem for vendors will be that much more developed.

你处于一个非常有趣的位置，因为你同时涉足了两个领域。

You're in an interesting position where you've got feet in sort of two camps.

对吧？

Right?

一方面，作为Sierra，你是这些供应商之一，属于应用层。

With Sierra, you you are one of those vendors, the application layer.

另一方面，你还是OpenAI的董事长，属于基础模型层面。

You're also chair of OpenAI, one of the foundation models.

如果我们将时间推后五年，基础模型会包办一切吗？

If you take us forward five years, does the foundation model do everything?

像更强大的Claude或ChatGPT这样的模型，会对Sierra的业务构成威胁吗？

Is there a threat to your business at Sierra from a more competent Claude or ChatGPT?

我不这么认为。

I don't think so.

我不会因此创办Sierra，但我认为这是一个非常重要的问题。

I wouldn't have started Sierra, but I think it's a really important question.

当然，在旧金山的鸡尾酒会上，大家都在讨论这个话题。

And certainly in the cocktail parties here in San Francisco, it's what everyone's talking about.

这些基础模型擅长的一件事是生成代码来开发软件。

One of the main things that these foundation models can do well is generate code to produce software.

如果你还没试过，打开ChatGPT，问问它帮你写一个应用或者网站，它真的能做到，而且做得不错。

And if you haven't tried it, you know, open ChatGPT and, you know, ask it to write an app for you or a website, and it will, and it's pretty good.

所以当你看到这一点时，你会意识到，生产软件的边际成本正在急剧下降。

And so you look at that and you're like, wow, the marginal cost of producing software is going down dramatically.

这对软件行业意味着什么？

What does that mean for the software industry?

我为什么要许可一份软件，而不是直接去使用AI模型，让它帮我生成那份软件呢？

Why would I license a piece of software or can I just go to, you know, an AI model and say, generate that software for me?

在我整个职业生涯中，一直存在软件开发人员短缺的问题。

And for my entire career, there's been a shortage of software developers.

这是大多数公司最稀缺的资源。

It's the scarcest resource at most firms.

因此，这是一个对多种行业都非常重要的问题，不仅仅是软件行业，还包括咨询行业。

So it's a really important question to ask for a variety of industries, not this as a software industry, but the consulting industry.

比如，其影响是巨大的。

Like, the ramifications are dramatic.

就像任何如此根本且重大的变革一样，要准确预测第二和第三阶效应是很困难的。

And as with anything so fundamental and so big, it's hard to actually predict the second and third order effects correctly.

我个人认为，大多数公司并不想自己开发和维护软件。

My personal opinion is that most companies don't want to build and maintain software.

所以即使生成软件这一过程的成本大幅下降，你仍然需要构建并拥有它，这是众所周知的。

So even if the act of generating a piece of software goes down by a lot, you build it, you own it, you know.

我就以ERP系统为例吧。

And so I'll just take an ERP system as an example.

十年前，针对软件公司出台了一项新的会计准则。

There was a new accounting standard that came out for software companies called a decade ago.

它改变了收入确认的方式，特别是对订阅制软件企业而言。

And it just changed the way you recognize revenue, particularly for subscription software businesses.

因此，当时我在Salesforce工作，这对我们来说是一件大事。

And so being in, you know, at Salesforce at the time, I was it was something that was a big deal for us.

如果你想象一下，每一家公司都必须重新实现这些会计准则，这有多重要啊——你的审计师会关注这一点，将这些成本分摊到众多相似的公司身上，对我来说非常合理。

And if you imagine every single company has to go reimplement those accounting rules and just how significant it is, right, like your auditor will care about it, Amortizing that cost among lots of similar looking companies feels really rational to me.

我不确定你从ERP供应商那里购买的，真的只是编写软件的成本。

And I'm not sure just the cost of writing software is actually what you're purchasing from your ERP vendor.

你几乎是在购买他们所做的审计，以及其他客户发现并帮助修复的那些漏洞。

You're almost purchasing the, like, audit that they've done and the bugs that other clients found that they fixed.

这个生态系统本质上是集体强化这个对你至关重要的系统，它本身就具有内在价值。

That essentially ecosystem of, like, collectively hardening that system that is so mission critical for you has innate value.

因此，我认为公司仍然希望购买问题的解决方案。

So as a consequence, I think companies still wanna buy solutions to problems.

他们不想购买，你知道的，软件。

They don't wanna buy, you know, software.

我的假设是，你会购买能够执行特定任务的智能代理。

And my hypothesis is that you will buy agents that do purpose tasks.

你可能会购买一个每季度审计你财务状况的智能代理。

And, you know, you might buy an agent that audits your financials every quarter.

你可能会购买一个接听客户服务电话的智能代理。

You might buy an agent that answers your customer service calls.

你可能会购买一个为你的销售团队生成潜在客户的智能代理。

You might buy an agent that generates leads for your sales teams.

将会有众多公司竞争，致力于在这些领域生产出最高质量的智能代理。

And there'll be a ecosystem of companies that compete to produce the highest quality agents in those categories.

这是我的假设，但我以谦逊的态度提出这一点，因为我的行业——软件行业，正受到人工智能的冲击，其程度不亚于任何其他行业，甚至可能更甚。

That's my hypothesis, but I say this with the humility that it is my industry, the software industry, is being disrupted as much as any other, perhaps more than any other as a consequence of AI.

所以如果我们一年后再聊，我可能会有不同看法，但这是我目前的观点。

And so if we talk again in a year, I might have a different opinion, but that's my opinion right now.

所以你的愿景是无处不在的智能代理。

So your vision is agents everywhere.

Sierra的核心主张是，我认为我们能为你大幅节省人力成本。

Sierra's proposition is, I think, fundamentally, we can save you a lot of money on labor.

你对人类在哪些方面仍保有优势，以及这对整体就业意味着什么，有何假设？

What is your hypothesis on where humans continue to have an edge and what this means for jobs generally?

我需要稍作修正。

I'll do a small correction.

我不认为我们的主要价值主张是降低人力成本。

I I don't think our main value proposition is labor cost savings.

我认为我们的主要价值主张是提升你的销售业绩和客户关系。

I think our main value proposition is improving your sales and your relationships with your customers.

如果你想想移动运营商，看看推动他们业务的关键因素，那就是客户获取和客户流失，而流失实际上更隐蔽，因为你花了很多钱获取客户后，即使流失率降低几个基点，价值也极其巨大。

So if you think about, say, a mobile operator, if you look at what drives their business, it's customer acquisition and churn, you know, and churn is actually the more insidious because once you spent all that money acquiring that customer, even a handful of basis points of churn reduction is worth a ton.

而客户流失的主要驱动因素之一就是客户体验和客户服务。

And one of the main drivers of churn is customer experience, customer service.

所以，如果你考虑一下你用于与客户沟通的预算，假设你把每次互动的成本从10欧元降低到1欧元，那么你现在就有了十倍于以往的客户互动预算。

And so if you think about you have a budget for talking to your customers and you've made it so the cost per interaction goes down from €10 to €1 just for argument's sake, you now have 10 times the budget for customer interactions.

你会在成本节约上回收多少，又会把多少资金重新投入以降低流失率或推动更多销售？

How much will you recoup in cost savings and how much will you invest back in reducing your churn or driving more sales?

成本降低固然有趣，但大多数CEO的聘用和解雇都是基于增长表现。

Cost reductions are interesting, but I mean, most CEOs are hired and fired based on growth.

但我确实认为你关于工作的观点是对的，某些工作由AI代理完成会比人类更出色，这在过去也一直如此。

But I do think your point on jobs, certain jobs are going to be more capably done with an AI agent than a person, and that has been true in the past.

如果你看看自动取款机的出现，它从很早以前就改变了银行分支机构中人员的角色，甚至追溯到农业领域——当美国建国时（1776年），我们国家大多数人都从事农业，类似的情况已经发生过很多次。

If you look at the birth of the automated teller machine, it changed the role of people in bank branches all the way back to agriculture where when The US became a country in 7076, most of our country were farmers and so this happened many times.

工作已经发生了巨大变化。

Jobs have changed a lot.

我强烈不同意这个前提，而且我要强调，我是带着极大的谦逊来说的，那就是认为工作会消失、我们将无事可做的想法。

The premise I strongly disagree with, and again, I I come with this with lot of humility, is the idea of, like, jobs will go away and we'll have nothing to do.

我根本不认同这一点。

I just don't agree with that.

我认为实际上是我们缺乏想象力，无法想象围绕这项技术会诞生哪些新工作。

I think actually we just lack the imagination to think about what jobs will be formed around this technology.

但问题在于，技术演进的速度究竟有多快。

The question though is just how quickly the technology evolves.

我妈妈在一家石油公司工作了三十年，她不需要每五年就彻底重新学习技能。

You know, my mom worked for a oil company for thirty years, and she didn't have to completely reskill every five years.

对吧？

Right?

当时的变革步伐要慢得多。

It was a much slower pace of adoption.

美国和英国的电气化过程持续了几十年，你知道的。

The electrification of The US and The UK took decades, you know.

我们现在面对的是一种技术，就拿我自己的职业——软件工程来说，今天的最佳实践和十二个月前已经完全不同了。

And we're now have a technology where I'll just take my own profession, you know, software engineering where, like, what is the best practice today is completely different than it was twelve months ago.

要求公司里的员工如此快速地重新学习技能，这是一个极具挑战性的期望，我认为这对许多白领岗位来说都是如此，而这些岗位原本并不需要这样。

It is a challenging expectation for the individuals at your company to have to reskill that quickly, and I think it's happening for a lot of white collar jobs where that wasn't the expectation going in.

所以我觉得这真的很有挑战性。

So I think that's really challenging.

不过，我想给出一个非常乐观的看法：我们所有人都在同一条船上。

I'll give you what I think is a really optimistic take on it though, which is we're all in the same boat together.

在CERA办公室里，没有哪位软件工程师比其他人更懂编码代理，因为这项技术完全是全新的。

There's no software engineer here in the CERA offices that's an expert in coding agents anymore than any other software engineer because the technology is all new.

这就像个笑话，仿佛我们都是会计，而微软Excel是上周末才发明的。

It's joke it's like we're all accountants and Microsoft Excel was invented last weekend.

还没人会用数据透视表，但如果你是第一个学会的人，你就会成为最厉害的会计。

No one knows pivot tables yet, but if you're the one who learns it first, you're going to be, like, the best accountant on the block.

所以我觉得这令人畏惧，但如果你抱着初学者的心态，那些最能适应并灵活运用这项技术的员工，实际上反而能加速自己的职业发展。

And so I think it's intimidating, but it with a kind of a beginner's mindset, I think the individuals at your firms that actually adopt this technology the best and most fluidly can actually accelerate their careers.