为什么 AI 代理能够击败行业巨头

如果你想造飞机，就得跟成千上万的供应商打交道。

If you want to build an aircraft, you need to procure thousands of suppliers.

有人发来确认说："嘿，抱歉，这个零件要晚两周到。"

Someone sends a confirmation of, "Hey, sorry, this part is going to arrive two weeks later."

要是他们漏看了这封邮件，那损失可是数以亿计的。

And if they miss this email, hundreds of millions of them.

以前采购可能比较封闭，现在不一样了，它得跟法务、财务还有各种软件系统和人员对接。

Procurement historically may have been more in a box, and now it's like, okay, it's touching legal, it's touching finance, it's touching a bunch of different software systems and people.

AI原生初创公司的机会在于，我们可以说："我们要掌控整个端到端的流程。"

The opportunity for the AI-native startup is to say, "We're going to own that entire end-to-end arc."

没有哪家公司或企业会从第一天起就完全依赖自主谈判代理。

No company and no enterprise starts with fully autonomous negotiation agents from day one.

为什么？

Why?

因为他们不信任我们，也不信任这项技术。

Because they don't trust us, and they don't trust the technology from day one.

通过这种人在回路中的方法，我们把所有的反馈和学习都喂给我们的智能体，然后他们突然就信任我们了，能处理一万次、两万次、十万次的谈判。

And by having this human-in-the-loop approach, we are feeding our agent with all the feedback and all the learnings, and then they suddenly trust us for 10K negotiations, 20K negotiations, 100K negotiations.

当你考虑一个持久的垂直AI公司应该是什么样子时，你会看重哪些特质？

When you think about what a durable vertical AI company looks like, what are the qualities that you look for?

要预测未来的护城河真的非常困难。

It's really, really hard to forecast your moat going forward.

如果你回顾所有最优秀的企业，在早期阶段，它们都是

If you look back at all the best businesses, at the early stages, they were

如果现有企业已经拥有客户、数据和记录系统，那么AI原生的初创公司在哪里有优势？

If the incumbent already owns the customer, the data, and the system of record, where does an AI-native startup have an advantage?

在本期节目中，Elena Berger与a16z合伙人Seema Amble以及Lio联合创始人兼CEO Vlad Keil坐下来，通过企业中最复杂的部分之一——采购——来回答这个问题。

In this episode, Elena Berger sits down with a16z partner, Seema Amble, and Lio co-founder and CEO, Vlad Keil, to answer that question through one of the most complex parts of the enterprise, procurement.

他们深入探讨了为什么这么多实际工作发生在记录系统之外，跨越电子邮件、电子表格、合同、工程数据以及与供应商的对话。

They get into why so much of the actual work happens outside the system of record, across emails, spreadsheets, contracts, engineering data, and conversations with suppliers.

Vlad解释了Lio如何构建能够协调这些上下文的智能体，并越来越多地端到端地承担这项工作。

And Vlad explains how Lio is building agents that can coordinate that context and increasingly take on the job end to end.

他们还讨论了如何赢得足够的信任，让代理替你谈判，为什么构建内部AI工具在遇到例外情况之前可能看似简单，以及当代理最终在交易双方都出现时会发生什么变化。

They also discuss how you earn enough trust to let an agent negotiate on your behalf, why building an internal AI tool can be deceptively easy until you hit the exceptions, and what changes when agents eventually sit on both sides of a transaction.

欢迎回到a16z播客。

Welcome back to the "a16z" podcast.

我是Elena Berger，今天和我一起的是a16z的合伙人Seema Amble，以及Lio的联合创始人兼CEO Vlad Keil，Lio为企业的采购流程构建AI代理。

I'm Elena Berger, and today I'm joined by Seema Amble, a partner at a16z, and Vlad Keil, co-founder and CEO of Lio, which builds AI agents for enterprise procurement.

Seema，你最近写了一篇文章叫《现有巨头正在到来》，它提出了一个几乎每个AI应用公司都面临的问题。

Seema, you recently wrote a piece called "The Incumbents Are Coming," and it asks a question facing almost every AI application company.

如果一家成熟的软件供应商已经拥有客户和数据，而且模型可以在其工具中运行，那么初创公司的优势在哪里？

If an established software vendor already has the customer and the data, and the model can work across its tools, where does a startup have an advantage?

今天我们请到了Vlad，他可以帮助我们真正理解初创公司的优势在哪里。

And we have Vlad here, and Vlad can really help us understand where a startup has an advantage.

所以我认为一个好的起点是讨论支持现有巨头的理由和反对他们的理由。

So I think a good place to start is the cases for and against the incumbents.

那么，如果一家公司可以将一个能干的AI代理连接到它已经使用的软件上，那么另一个应用在哪里有实际的使用场景和优势呢？

So if a company can connect a capable AI agent to the software it already uses, where does another application actually have a use case and an advantage there?

好，让我先退一步，把背景讲清楚。

Yeah, let me back up and frame it up a little bit.

从历史上看，我们一直认为，一边是现有的大公司，另一边是初创公司，这就像分销和创新之间的较量，借用我合伙人亚历克斯·兰佩尔的话来说。

So historically, we thought about there is the incumbent and then there's the startup, and it's a fight between distribution and innovation, which to take my partner Alex Rampell's phrase.

但现在出现了第三个因素，也就是你提到的，现有大公司可以在自己的产品上叠加一个模型，这样一来，市场上就会出现一个更强大的竞争对手。

However, now there's this third piece, which you're pointing at, which is the incumbent can layer on one of the models on top, and then there's a much more formidable competitor in the market.

所以，如果你已经有了Claude Force，也就是把Claude和Salesforce结合起来，而且你所有的数据都已经在里面，员工也都习惯了用这个产品，那为什么还需要一个AI原生的初创公司呢？

So why do you need an AI-native startup if you've got Claude Force, which is taking Claude plus Salesforce and putting the two together, and then you've already got all your data, and your employees are all used to using the product.

为什么还需要另一个产品？

So why another product?

我仍然坚信，AI原生初创公司绝对还有存在的理由，关键在于，传统大公司受限于自己的记录系统，以及他们拥有的那些数据，他们并没有完成端到端的任务。

I absolutely still think there's obviously still a case for the AI-native startup, and it really centers around the fact that the legacy incumbent is limited to their system of record and that record that they have, and they're not completing the end-to-end job.

让我用一个具体的例子来说明。

So let me put that more concretely in an example.

比如说，你是一个客户，打电话来说，取消服务后还是被扣了钱。

So say you're a customer and the customer calls and says they got charged after they got canceled.

解决这个取消历史的问题，不只是客户在聊天里说一句“嘿，我被多收费了”那么简单。

Resolving that cancellation history isn't just the customer going into the chat and saying, "Hey, I got overcharged."

而那边的回应，必须得涉及到账单系统。

And the response there, it has to hit billing.

还得查看所有的聊天记录。

It has to look at all the chat history.

还得看合同。

It has to look at the contract.

这可不是单一的系统记录能搞定的。

That's not one system of record.

这涉及到围绕那个客户的所有知识，以及它接触过的一切，这是单一系统记录无法触及的。

That's the knowledge around that customer and everything it touched, and that's something that one system of record wouldn't touch.

不过，对于AI原生的初创公司来说，机会就在于说：“我们要掌控整个端到端的流程。”

However, the opportunity for the AI-native startup is to say, "We're going to own that entire end-to-end arc."

比如在法律领域，从起草文件一直到庭审，全都包揽下来。

So that could be in legal, so owning everything from brief all the way through trial.

Vlad可以多讲讲采购方面，但核心就是这种端到端的工作流程。

Vlad can talk more about procurement, but it's really this concept of owning the end-to-end work.

Vlad，你想聊聊采购中哪些地方体现这一点吗？比如现有的老牌厂商加模型的方式为什么不够用，以及你和你合作的公司都看到了哪些情况？

Vlad, do you want to talk about where that does show up in procurement, like where existing incumbent plus a model just is insufficient, and what have you seen just with the companies that you work with?

好的。

Sure.

说到采购，我猜大多数人想到的都是价格，对吧？

So when we think about procurement, I would assume that most people think about prices, right?

就是我们最终谈下来的价格是多少。

So what is the end price that we negotiated on?

但令人意外的是，记录看起来总是非常简单、非常容易。

And surprisingly, the record looks always very, very simple and very easy.

就只是8K。

It's just 8K.

比如，这就是最终结果。

That's the result, for example.

销售也是一样的，对吧？

And it's the same for sales, right?

所以即使我去找西玛，告诉她：“嘿，我们现在和另一家企业合作了。

So even if I come to Seema and tell her, "Hey, we now partnered up with another enterprise.

看看这份合同上的签名。”

Look at the signature here for this contract."

看起来很简单，但西玛看不到背后的所有工作，对吧？

It looks very easy, but Seema doesn't see all the work behind it, right?

所以可能开了30次利益相关者会议，发了500封邮件，做了20个Excel表格，而对方的采购也是一样的情况，对吧？

So there are probably 30 stakeholder meetings happened, 500 emails, 20 Excel sheets, and that's the same for the counterpart procurement, right?

所以你在ERP系统里看到铝材的价格是8K，但你看不到供应商可能反对过，要求10K。

So you see in your ERP system 8K for aluminum, but you don't see that maybe the supplier did a pushback and asked for 10K.

你看不到成本工程师花了三周时间做Excel表格和3D建模，来确定零件的价格和其他一切。

You don't see that a cost engineer had run three weeks of Excel sheets and 3D modeling to find out the prices of the part and everything else.

这就是我们看到的情况。

So that's what we see.

实际上，采购中的大部分工作都发生在ERP或任何记录系统之外。

So most of the work in procurement actually happens outside of this ERP or any system of record.

我相信在你描述的工作流程中，一个智能体可以完成很多事情，不同类型的智能体也能做很多事情。

And I'm sure in the workflow that you described, an agent can do a huge number of things, and different kinds of agents can do a large number of things, too.

我觉得这正好可以引出下一个问题。一年前，很多老牌公司都在发布聊天机器人，那基本上就是你能看到的所有东西了。

And I actually think that's a good bridge into the next question, which is a year ago, I think a lot of the incumbents were- ... releasing chatbots, and that was kind of the extent of what you would see.

但是Seema，在你写的这篇文章里，你列出了四种不同类型的智能体。

But Seema, in this piece that you wrote, you lay out four different kinds of agents.

检索智能体，就是我们熟悉的聊天机器人，还有流程智能体、政策智能体和主智能体。

Retrieval agents, which are kind of the chatbots that we're familiar with, process agents, policy agents, and principal agents.

你能不能给我们逐一介绍一下，并解释一下在这些智能体中，所需的判断力有什么变化？

So can you just walk us through all of them and explain what changes in the kind of judgment that's necessary across all of them?

一年前，我做了个表情包，就是拍打聊天机器人策略，基本上所有老牌公司都有一个聊天机器人，它放在记录系统上面，你可以通过它来检索信息，也许还能做一些简单的分析。

So a year ago, I made this meme, which was the slap on a chatbot strategy, which is essentially all the incumbents effectively had a chatbot that sat on top of the system record, which you could chat with to retrieve information, maybe do some light analytics.

那其实就是第一类，也就是检索智能体。

And that really was in that first bucket of what the retrieval agent is.

那我用客户支持的例子来分别说明一下检索型、流程型、政策型和主理型这几种代理，这样比较好理解。

And maybe let me walk you through an example of what each of the retrieval, the process, the policy, and the principle back to the customer support example, just because it's easier to understand.

想象一下，你是个客户，遇到服务中断，你打电话说：“嘿，我想为这次服务中断获得赔偿。”

So imagine if you're a customer and there's a service outage, and you're calling in to say, "Hey, I want to get compensated for this service outage."

现有的检索型助手，也就是那些传统厂商可能会有的，它能调出信息说：“是的，合同条款是这样写的”，“是的，确实发生了中断”，然后核实这些信息。

The retrieval assistant, which the incumbent may have, is going to be able to pull up, "Yes, this was what the contract term said," and, "Yes, there was an outage," and just verify that information.

它从数据库里调出客户的信息，然后反馈回去，可能还会做一些总结。

It's pulling up information about the customer, it's in the database, and it's just sharing it back and maybe synthesizing.

代理序列中的第二步是流程型代理。

The second step in the agent sequence is the process agent.

这个流程型代理可能能够完成信用审批，并说：“好的，根据我们的政策手册，服务中断是从这些日期开始的。”

So that process agent may be able to pull through an approval of credit and say, "Okay, based on our policy handbook, it was out from these dates.

因此，我们认为你有权获得赔偿，然后它就可以直接处理账单，完成整个流程。

Therefore, we say you are entitled to money" and it can just apply the bill and just do process.

这里面不涉及任何判断。

There's no judgment involved.

然后继续到政策代理，政策代理不只是执行流程，而是会说：“好的，在这种情况下，中断了20分钟。”

Then as you keep going to the policy agent, so the policy agent isn't just going to apply the process, but it's going to say, "Okay, in this situation, it was out for 20 minutes.

“这足以被视为一次重大中断。”

That's enough to be considered a significant outage."

他们在运用这种判断，因为对此并没有严格的定义。

And they're applying that judgment because there isn't really a strict definition around it.

然后最后一个阶段，也就是原则代理，你实际上在权衡，好的，我们是否应该提供更多补偿？

And then the last stage, which is the principal agent, you're actually weighing, okay, should we offer more compensation?

因为那是一次相当糟糕的中断，我们想要维护关系，值得做得比流程或政策规定的更多。

Because that was a pretty terrible outage, and we want to preserve the relationship, and it's worth doing more beyond even what a process or policy says about it.

但这四件事的重要性在于，大多数现有企业，如果有所涉足的话，都是从第一类开始的。

But the importance of these four things is that most incumbents started out in, if at all, in bucket one.

现在，至少他们在宣传自己正在向流程和政策迈进，这意味着他们能够运用更多的判断。

Now, at least they're marketing that they are moving towards process and policy, meaning they would be able to apply more judgment.

如果你看看他们推出的东西，这些是工作流代理，能帮你签署文件、从转录中提取信息之类的。

And if you look at what they've launched, these are workflow agents that will help you get a document signed or input information from a transcription or things like that.

我觉得它们目前主要还是局限在检索和一点流程处理上。

They're very much still limited to, I would say, retrieval and a little bit of process.

还没涉及到需要更多判断力的环节，我可以解释一下原因。

They've not gotten into more judgment, and I can get into why.

有很多因素在阻碍它们往那个方向发展。

There are all sorts of incentives that are preventing them from that.

不过传统厂商也在努力尝试，有些人的思路是跟OpenAI或者Anthropic这些实验室合作，把他们的模型能力拿过来，跟自己现有的东西结合起来，实现更强的赋能。

But the incumbents are trying, and I think what they're not able to do, they're like, some of them are saying, "Okay, let me partner with OpenAI, Anthropic, one of the labs, and try to build out, take that model capability and complement what they have and superpower it."

或者收费，我应该这么说。

Or charge it, I should say.

那么，你觉得为什么有些老牌企业还在犹豫不决呢？

Well, do you want to say why sort of some of the incumbents are holding back?

我觉得他们确实有所保留，但也是被形势所迫。

Okay, so I would say they're holding back, but they're held back.

明白了。

Okay.

我相信他们想全力以赴，但可能有两个方面。

I'm sure they want to be full force, but there's probably two pieces.

第一，他们有分销优势，对吧？

One, they have this advantage of distribution, right?

他们拥有客户的信任，这让他们能向客户销售更多产品。

They have the customer trust, which enables them to then sell more products with the customer.

就拿Salesforce来说。

So take Salesforce.

当Agentforce推出时，客户很容易就说：“好的，我要注册Salesforce的代理”，尤其是如果几乎不额外收费的话。

When Agentforce launched, it was very easy for customers to say, "Yeah, I'm going to sign up for the Salesforce agent" especially if it was offered at almost no extra cost.

所以他们有这种信任，有分销渠道。

And so they have this trust, they have the distribution.

启用产品通常非常顺畅。

It's often pretty seamless to turn on the product.

但另一方面，存在所有这些内部激励问题，对吧？

The flip side is there's all these internal incentive issues, right?

也就是说，如果你开始处理更复杂的代理，就会和现有的提供工作流程的产品产生内部冲突，而不是去解决问题本身。

Which is if you start getting into more complicated agents, it's an internal conflict with an existing product that's offering workflow versus you're resolving the work.

这两个产品是不同的，如果你要端到端地解决客户支持问题，而不是为支持代理（人工代理）提供工作流程，那么买家是不同的。

And those are two products that are, if you're solving a customer support issue end to end versus providing a workflow for a support agent, a human agent, those are different buyers.

这两个团队是相互冲突的。

Those two teams are in conflict.

而且在此基础上，从销售的角度来看，你卖给客户的是什么？

And then on top of that, I think from a sales perspective, what are you selling to the customer?

然后，在经典的现有企业问题中，通常会有两个副总裁，对吧？

And then oftentimes in a classic incumbent issue, right, is there's two VPs, right?

他们属于不同的组织，销售不同的产品，他们永远无法确定正确的激励措施和正确的销售对象。

And they're different orgs, and they're selling different products, and they will never be able to figure out what the right set of incentives and the right person to sell to.

但不管怎样，我认为现有企业也会遇到所有这些经典的现有企业问题。

But anyway, so there's all these sort of classic incumbent issues I think that the incumbent also runs into.

有道理。

Makes sense.

Vlad，在检索代理、流程代理、策略代理、主代理这个谱系里，Lio 处于哪个位置？

Vlad, where across this kind of spectrum of retrieval agent, process agent, policy agent, principal agent, where does Lio sit?

我们其实覆盖了所有这些类别，具体取决于我们代理所承担任务的复杂度和风险，对吧？

So we span across, I would say, all of those categories, and it really depends on the complexity and the risk our agents take, right?

有时候，我们已经有一些完全自主运行的用例。

So sometimes we kind of we already have use cases where we run fully autonomously.

有时候则需要人工介入。

Sometimes you have the human in the loop.

这真的取决于预算审批，以及任务的复杂度和风险程度。

It really depends on the budget approval, how complex and how risky this is.

但也许回到 Seema 提到的关于现有企业信任的问题。

But maybe coming back to what Seema said about trust for the incumbents.

你谈到了内部信任。

You talked about internal trust.

我觉得还有外部信任的问题，对吧？

I think there's also an external trust thing, right?

那么，你怎么说服别人从那种只是检索信息、可能跑跑流程的代理，过渡到完全自主运行的程度呢？

So how do you convince someone to go through from, hey, just like an agent that retrieves some information and maybe runs processes to do something completely autonomously.

这显然有产品层面的因素，但主要还是人的因素，所以他们需要信任你。

It's like, obviously, there's a product component to it, but mainly it's a people component, so they need to trust you.

而我们作为初创公司、成长型公司，必须赢得这份信任。

And we as a startup, scale-up, we have to earn the trust.

老牌企业已经拥有信任，但这也意味着如果他们过早推出产品，而产品可能不工作、很糟糕，或者做出糟糕的决策，他们就会毁掉这份信任。

Incumbents already have the trust, but this also means they can destroy the trust if they ship something too early and the product maybe doesn't work, or it's bad, or it makes worse decisions in a bad way.

所以这就是我们能做的。

So that's what we can do.

然后有趣的是，或者说让我们很惊讶的是，这些流程代理实际上成了我们的一个意外收获，因为当我们看发票流程，或者说发票代理作为一个流程时。

And then interestingly, or what was really surprising for us, those process agents actually became kind of like a side quest for us because When we look at the invoice process, or invoice agents as a process.

我们检索一些信息，你把它和其他文档进行匹配，然后把它推回SAP Oracle。

So we retrieve some information, you match this across other documents, and then you push this back to SAP Oracle.

一个非常清晰的流程。

A very clear process.

然后我们发现，这其实覆盖了100%的软件市场。

And then we figured out, okay, that's 100% of the software market.

这就是发票软件。

So that's invoice software.

现在大家就是这么做的。

That's how you build it today.

但实际上，这只是整个工作、整个任务、整个问题的20%。

But it's actually only 20% of the work, of the job to be done, of the problem.

因为80%的问题在于，如果发票是伪造的怎么办？

Because 80% of the problem is what if the invoice is fraudulent?

如果信息不匹配怎么办？

What if there's a mismatch?

如果我们不走常规路径怎么办？

What is if we don't take the happy path?

所以我们很快就转向了下一步，让代理来处理这些异常情况。

And so we very fast shifted to the next step of agents doing those exception handlings.

我们说服了客户，我觉得我们运气好，总是能稍微领先一步，所以能向他们推销下一代智能代理。

And we convinced the customers by, I think we got lucky being always slightly ahead of the curve, so we were able to pitch the next generation of agents.

比如，三年前我们刚开始的时候，只是能检索文档。

So for example, when we started out three years ago, it was just retrieving a document.

现在看起来没什么了不起的，但三年前，这已经非常令人惊叹了。

Not impressive at all today, but three years ago, this was crazy impressive.

所以我们把这个推销给了客户。

So we pitched this to customers.

我们发现，这确实是个问题，是个他们愿意付费的真实用例。

We find out, okay, that's a real problem, that's a real use case they would pay for.

然后几周后我们就能交付这个功能了。

And then we were able to ship this some weeks later.

同样地，我们现在也在为下一步做这件事，比如流程代理、完全自主的代理，以及长期运行的代理。

And in the same way, we are doing this now for the next step, for process agents and for fully autonomous agents and for long-running agents.

我们向他们推销这些，确认这是个问题，然后就能很快交付。

We are pitching this to them, finding out it's a problem, and then we're able to ship this very fast.

我觉得关于信任，有个很有意思的点，就是内部信任和外部信任的区别。

I think an interesting point on trust is this internal versus external trust.

另一个角度看，是的，你需要客户购买采购软件，并信任它用于他们的内部流程。

The other lens on that is, yeah, so you need your customer to buy the procurement software and trust to use it for their internal processes.

但我们第一次见Vlad时，最有趣的一点是，他们还在做谈判。

But one of the really interesting things when we first met Vlad was that they're also doing the negotiation.

所以你得信任Lio代理会去和第三方对接，而且——

So you have to trust that the Lio agent is going to then interface with a third party, and-

……这里面就涉及信任的问题。

... and there is that piece of trust.

当然，我觉得很多人对现有供应商感到失望，他们功能有限，过去宣传的很多功能都没实现。

And of course, I think a lot of people feel burned by the incumbents and they're pretty limited and haven't been able to do what they've marketed they've had in the past.

但抛开这些不谈，Vlad，我很好奇，你是怎么说服客户信任AI代理去处理谈判的？

But putting that aside, I don't know, maybe Vlad, I'd love to hear a little bit how you convinced the customers to trust an AI agent to now take on negotiations.

对，而且你讲的时候，能不能也描述一下采购具体涉及什么，你们的客户是谁，他们习惯用什么样的传统系统？

Yeah, and when you do that, can you also just paint the picture of what's involved in procurement, and who are your customers, and what kind of legacy systems are they used to?

说到采购，你可能只会想到买东西，或者看看B2C的世界，就是买点东西。

So when you think about procurement, you maybe just think about purchasing or when you look at B2C world, just you buy something.

但实际上，这是一个非常复杂的过程，它支撑着整个经济运转。

But actually, it's a very intense process which runs the economy.

而且它涉及多个利益相关方，很多部门都要参与进来。

And it includes multiple stakeholders and a lot of stakeholders and a lot of departments.

法务、成本工程、当然还有采购、财务，所有这些部门都得在这些决策上协作。

Legal, cost engineering, obviously procurement, finance, all of them have to work on these decisions.

基本上，我觉得我们说服他们的原因，就是我之前提到的那些。

And essentially, I think the reason is how we convince them is what I already said earlier on.

我们总是稍微领先一步。

So we're always a little bit ahead of the curve.

我们知道这项技术会到来。

So we knew the technology is coming.

所以在ChatGPT发布之前，我们就在它突破前几周开始了。

So even before ChatGPT came out, we just started a few weeks before this ChatGPT breakthrough.

所以我们听到了所有的问题，然后听到了炒作和所谓的技术泡沫，我们能够向这些企业推销，并且很快发现，好的，这可能是一个有趣的用例，比如聊天机器人应用或检索代理、文档处理。

So we already heard all of the problems, then we heard the hype and the, let's say, tech bubble, and we were able to pitch this enterprises, and we quickly figured out, okay, this could be an interesting use case, like just chatbot applications or retrieval agents, document processing.

然后我们能够找出问题，并迅速交付给他们。

And then we are able to find out the problem and then ship this quickly to them.

然后显然，这是关于信任的人的因素，我们告诉他们一些事情，然后我们真的能够把产品投入生产。

And then obviously, this is the people factor of trusting, so we are telling them something about, and then we're able to really ship something in production.

但还有产品视角，我们有很多评估机制。

But then there's also this product perspective where we have a lot of evals in place.

没有公司或企业一开始就使用完全自主的谈判代理。

So no company and no enterprise starts with fully autonomous negotiation agents from day one.

没有人会这么做。

No one does that.

为什么？

Why?

因为他们不信任我们，也不信任这项技术从一开始就可靠。

Because they don't trust us, and they don't trust the technology from day one.

所以我们有一个非常简单的做法，就是让人参与其中。

So we have a very easy approach of having a human in the loop.

这对我们来说非常有用，因为我们有一个，比如说，完美的谈判代理，也就是一个完美的采购谈判专家。

And this is extremely helpful for us because we have, let's say, the perfect negotiation agent, which is overall the perfect procurement negotiator.

但我们并不完全了解这家特定的财富10强企业是如何运作的。

But we don't know exactly how this specific Fortune 10 enterprise operates.

通过这种让人参与其中的方式，我们把所有的反馈和学习都喂给我们的代理。

And by having this human in the loop approach, we are feeding our agent with all the feedback and all the learnings.

然后他们突然就信任我们进行1万次谈判、2万次谈判、10万次谈判。

And then they suddenly trust us for 10K negotiations, 20K negotiations, 100K negotiations.

然后你还有其他一些运行时间更长的代理。

And then you also have other agents that are more long-running.

所以当我们谈到数百万美元的谈判，需要分析复杂的3D模型和技术图纸时，我们特意总是让专家参与其中。

So when we talk about multimillion-dollar negotiations where you analyze complex 3D models and technical drawings, there we on purpose have always experts in the loop.

所以有一个代理运行了好几个小时，然后我们向成本工程师征求反馈，然后它再做下一步工作，就这样继续下去。

So there's an agent running for multiple hours, and then we ask for feedback of the cost engineer, and then it does the next work, and so on.

也许我们可以深入聊一下，在谈判方面，你们具体是在哪个环节引入人工介入的？

Maybe just to double-click on that, where do you put the human in the loop on the negotiation side?

你提到了成本工程师，但如果是在来回谈判的过程中，主要是围绕成本工程这类数据，还是说还有其他方面也会有人工介入？

You mentioned the cost engineer, but if you were going back and forth on a deal, is it mostly around the data for something like cost engineering, or is there anything else where you have humans in the loop there?

这还是要看谈判的级别。

So, again, depends on the level of negotiation.

我们需要区分两种情况：有些谈判是企业以前根本不会去做的，因为他们没有这个能力。

So we have to distinguish between negotiations where enterprises, they never did those because they didn't have the capacity.

但通过部署智能体，他们就能抓住以前没意识到的节省机会。

But by deploying agents, they can just capture savings that they weren't aware of.

在智能体或Lio出现之前，他们根本不会去管5万以下的事情。

So before agents or before Lio, they just didn't care about everything which happened below 50K.

所以你可以——这可能是对其他初创公司的一个小窍门。

So you can just This is maybe a hack for other startups.

你可以直接给企业发一张4万的发票。

You can just send an enterprise an invoice for 40K.

他们可能不会谈判，因为他们没有这个能力。

They will probably not negotiate because they don't have the capacity to do so.

除非他们有Lio代理，那我们就会跟你谈判。

Except they have Lio agents, then we are going to negotiate against you.

但除此之外，他们就直接付钱了。

But other than that, they are just paying that.

而且显然，如果你完全不谈判，风险是什么？现在有个不好的谈判代理，风险又是什么？

And obviously, what is the risk of you didn't negotiate at all, so what is the risk now of having a bad negotiation agent?

几乎为零。

Nearly zero.

所以也许我们会错过一些谈判，但总比没有好。

So maybe we miss out on some negotiations, but it's better than nothing.

但话说回来，我们谈的是商业关系，而商业关系并不总是关于成本和金钱的。

But still, we're talking about business relationships, and business relationships are not always about the cost and the money.

所以也许你在供应商上花得不多，但你可能需要这种商业关系。

So maybe you're not spending a lot on the vendor, but maybe you need this business relationship.

比如播客或者营销服务就是一个很好的例子。

So a good example might be podcasts or marketing services.

好的，那可能只是支出的一小部分，但你跟某些人比如搭建录音室的人有明确的业务关系，你不想让随便什么人来干这活，因为他们已经知道a16z是怎么运作的，也知道你想怎么录制这些东西。

Okay, that's probably a friction of the spend, but you have a clear business relationship with someone like setting up the studio, and you don't want some random people doing that because they already know how a16z operates and how you want to record all this stuff.

所以在这种情况下，我们采用人机协同的方式，采购人员在乎的是关系，所以他们关心语气和运作方式。

So there we have human-in-the-loop approaches, where procurement people care about the relationship, so they care about the voice of tone and how it works.

但大部分是自动化的。

But mostly autonomously.

然后还有另一类代理，我们总是采用人机协同的方式，而且谈判中有多个步骤。

And then we have the other set of agents where we always have a human-in-the-loop approach, and there are multiple steps in negotiating.

不仅仅是价格问题。

It's not only the price.

还有合同是怎么设计的。

It's also how is the contract designed.

所以我们说的是法律方面的问题。

So we're talking about legal.

收款设计怎么样？

How is collection design?

所以我们谈到了财务。

So we talk about finance.

显然，成本结构，所以我们谈的是真正的成本工程。

Obviously, cost structure, so we're talking about really cost engineering.

然后我们谈商业条款。

Then we talk about commercials.

那就是采购。

That's procurement.

那些人不是后台人员。

And those are not back office people.

他们是受过高度训练的人，对特定公司的特定流程和特定行业有非常具体的知识，他们把这些见解提供给Lio的长期运行的代理。

Those are highly trained people, where they have very specific knowledge of a very specific process of a very specific company and very specific industry, and they feed those long-running agents of Lio with those insights.

这是另一个例子，说明采购过去可能更局限在一个框框里，现在呢，它涉及法律，涉及财务——

That's another example of how procurement historically may have been more in a box, and now it's like, okay, it's touching legal, it's touching finance-

……而且它涉及很多不同的软件系统和人员，既有专家也有通才。

... and it's touching a bunch of different software systems and people, and both specialists and more generalists.

我们能不能把这个映射到一个具体的行业，而不是具体的客户？

Can we map this onto a specific, not a specific customer, but a specific vertical?

比如我是一家无人机制造商。

Like I'm a drone manufacturer.

我制造人形机器人或者类似的东西。

I manufacture humanoid robotics or something like that.

我需要订购和采购多少个零件？

How many parts do I have to order and procure?

我涉及多少家工厂？

How many factories am I touching?

我涉及多少家供应商？

How many suppliers am I touching?

就是所有这些方面。

Just all of those things.

如果你愿意选一个垂直领域，Vlad，比如用Lio来管理所有这些复杂性的领域，然后带我们看看他们的体验是怎样的。

If you want to pick maybe, Vlad, a vertical that is just managing all of this complexity with Lio and just take us through what their experience is.

我觉得这真的能很好地说明你们所涉及的一切。

I think that would really help just illustrate exactly just everything that you touch.

再说一次，当我们Lio谈论采购或购买时，我们说的不是笔记本电脑和铅笔。

Again, when we as Lio, when we talk about procurement or purchases, we don't talk about laptops and pencils.

对吧？

Okay?

我们认为那些问题Lio代理也能解决，但那太简单了。

We think that's solved also by Lio agents, but that's easy.

我们三年前就解决了。

We solved this three years ago.

我们说的是当你想造飞机、机器人、无人机，甚至我们现在在做的这个关于AI炒作的播客，但AI本身也需要被建造。

We talk about when you want to build an aircraft, or robots, or drones, or even we're now doing a podcast about, again, AI hype, but even AI needs to be built.

所以你需要数据中心。

So you need data centers.

而建设就意味着采购。

And building means procuring.

如果你要造一架飞机，你需要采购数千家供应商的产品。

If you build an aircraft, you need to procure thousands of suppliers.

你需要建一个工厂来制造这架飞机。

You need to build a factory to build this airplane.

而且很小的摩擦就能产生巨大的影响。

And really small frictions can have a crazy impact.

所以，如果你在运营一个非常大的项目，比如建数据中心或造飞机，如果有一个特定零件晚到两周，这可能会造成数亿美元的损失，并且推迟项目。

So if you're running a very large project of building a data center or building an aircraft, if there's one specific part which arrives two weeks later, this can have a damage of hundreds of millions of dollars, and postpone the project.

所以这就是为什么所有这些决策都必须协调一致。

So that's why all of those decisions have to be coordinated.

其中一部分是，你需要弄清楚你需要什么，和哪些供应商合作，哪个供应商最适合提供这个零件。

And one part is you need to figure out what you need, with what suppliers you work, what is the best supplier to get this part.

但一旦你决定了所有这些事情，背后还有所有这些运营后台工作，这些听起来可能不必要且无聊，但再说一次，运营意味着有人发送确认说：“嘿，抱歉，这个零件会晚到两周。”

But then once you decided all of the stuff, there's all this operational back office stuff behind it, which might sound unnecessary and boring, but again, operational means someone sends a confirmation of, "Hey, sorry, this part is going to arrive two weeks later."

而这只是采购经理在Outlook或Gmail里收到的500封邮件中的一封。

And this is one of 500 emails in the Outlook or Gmail of a procurement manager.

如果他们错过了这封邮件，损失可能高达数亿美元。

And if they miss this email, hundreds of millions of damage.

就这样。

Done.

这种情况经常发生，因为他们在系统里记录的只有日期。

And this happens regularly because the only thing they store in their system of record is then just the date.

所以不是这个周三，而是下个周三。

So not this Wednesday, next Wednesday.

你在系统里看到的就是这个。

That's what you see in the system.

但你不会看到，也许这没问题，但也许这是1亿美元的损失，必须有人来做决定，这也是我们的代理每天都在做的事情。

But you don't see maybe that's okay, but maybe that's a $100 million damage, and someone has to decide that, and that's also what our agents are doing day to day.

他们不仅仅是获取信息，还要做出决策。

They're not only retrieving the information, they are then making the decisions.

比如，这会产生影响吗？

Like, does this have an impact?

具体是哪种影响，我们该怎么解决呢？

What kind of impact, and how can we resolve this?

当你如此深入地融入物理世界时，你能介入并解决哪些现实中的问题？

When you are so deeply embedded in the physical world, what kinds of physical world problems can you intervene with?

我觉得有些问题是根本无解的。

Some things I would think are just unsolvable.

假设有一批货要运进来，结果东西从船上掉下来了，或者道路封闭了之类的。

Let's say you have a shipment coming in and a bunch of stuff falls off the ship, or the street is closed or whatever.

有些事儿是你没法控制的，当然也有些事儿你是可以处理的。

There are things that you can't do, and obviously, there are things that you can do.

所以关键就在于，你能在哪些环节介入，并且真正起到作用？

So where can you intervene, and where does that really make a difference?

其实这关乎的是概率问题。

But actually, it's about probability.

所以很明显，如果船上有损坏，你是改变不了的。

So obviously, you can't change if there's damage on the ship.

你处理的每个例子，你都改变不了，但如果你掌握了所有背景信息，你可以预测到这一点。

Every example that you manage, you can't change that, but you can If you have all of the context, you can predict that.

因为你可以预测供应商有多可靠。

Because you can predict how reliable is the supplier.

对吧？

Okay?

所以有办法保护你运输的货物。

So there are ways on protect the goods that you're shipping.

如果你有所有背景信息，你有一个供应商，20%的货物会丢失，另一个只有1%的货物丢失。

And if you have all the context, you have one supplier where 20% of the goods are missing, and then 1% of the good is missing.

也许那个20%丢失率的供应商便宜10倍。

And maybe this one with 20% is 10x cheaper.

但对于这个用例，你多付10倍的钱是值得的，因为货物实际到达的概率更高。

But for this use case, it's fine for you to pay 10x the amount, because you have a higher probability that this thing actually arrives.

所以，这就是它的强大之处，因为你不仅在企业内部有上下文，我们之前也谈到了多个利益相关者。

So, and this is the powerful thing, because you have context not only within one enterprise, so we talked about multiple stakeholders.

但外部世界也有上下文，对吧？

But there's also context on the outside world, right?

所以，智能体应该掌握所有外界的新闻动态。

So just the agent should have context of all the news out there.

甚至可能包括像Polymarket上的某些预测，比如，好的，那些中断将会发生。

Maybe even having context of some bet, say, on Polymarket, of like, okay, those disruptions are going to happen.

还有关于供应商方面、卖方方面、需求方面的信息。

Then information about on the supplier side, on the seller side, on the demand side.

通过整合所有这些上下文，我认为从长远来看，不会有局限性。

And by combining all of those contexts, I wouldn't say that there is a limitation in the long run.

当然，今天我们面对的是不同的概率组合，但我们可以在整个过程中提供帮助，这就是我们在Lio正在做的事情，对吧？

Obviously, today we have different sets of probability, but we can help throughout the process, and this is what we are building at Lio, right?

所以，这远不止是公司内部的采购事务。

So it's much bigger than just procurement intercompany.

这更多是公司内部的，以及企业之间如何做生意，也就是买方和卖方之间的互动。

It's more intra-company, and how enterprises are doing business with each other, so buyer and supplier side.

你把Lio描述成一个多智能体系统。

You describe Lio as a multi-agent system.

那你能说说这些不同的智能体各自都在做什么吗？

So can you describe what the different agents are doing?

一个层面是，Lio的智能体覆盖了Seema在文章里提到的所有四类。

One level is that Lio agents span across all those four categories that Seema mentioned in her article.

而且这又取决于风险和复杂程度。

And it, again, depends on the risk and the complexity.

所以我们全都用上了，也就是多个智能体。

So we use all of them, so multiple agents.

但另一方面，为了端到端地完成一项工作，这些智能体需要相互共享信息。

But the other thing is that in order to do a job end to end, those agents need to share information with each other.

而且它们必须按照非常特定的顺序来做这件事。

They need to do this in a very specific order.

当我们谈到多智能体系统时，这基本上就是我们在做的事情。

And when we talk about a multi-agent system, this is essentially what we are doing.

我们是在端到端地解决任务。

We are solving the task end to end.

而且因为这种人类级别的任务涉及八个人、八个利益相关者，可能还有三个部门和五种不同的软件工具，我们需要覆盖所有这些才能端到端地完成工作，而这些智能体之间也需要相互沟通。

And because the also human-level task involves eight people, eight stakeholders, and maybe three departments and five different software tools, we need to cover all of those to do the job end to end, and those agents need to then communicate with each other.

只有多智能体系统才能端到端地完成一项工作。

And only with a multi-agent system you can do a job end to end.

我们刚开始的时候，三年前是从检索开始的，更像是一个副驾驶。

When we started out, we started off with a retrieval, more like a copilot, obviously, three years ago.

但接下来的步骤是单一智能体。

But then the next step was a single agent.

但我们很快发现，即使是在谈判中，如果没有合同智能体，没有可能查看新闻和之前描述的所有内容的智能体，你也无法解决问题。

But then we very quickly discovered, okay, you can't solve in negotiation even, without having a contract agent, without maybe having an agent looking at the news and everything that I described before.

所以这就是我们定义的多智能体系统。

So that's what we define as a multi-agent system.

比如说，航空公司要采购一个螺栓，波音公司需要采购一个螺栓。

So if you have a bolt, like an airline company needs to procure a bolt for, let's say, Boeing needs to procure a bolt.

那采购螺栓的具体流程是什么？Lio 在这个流程中扮演什么角色？

What exactly is that process for procuring the bolt, and where does Lio step in on that process?

这是我们能够完全自主运行的采购之一，之所以能做到，是因为这个多智能体系统。

So this is one of the purchases that we can run fully autonomously, and we can do this because of this multi-agent system.

首先，有人有需求，对吧？

So first of all, someone has a demand, right?

他们需要以某种方式把这个需求传达出来。

So they need to somehow communicate it.

就连这一步都非常复杂。

And even this part is extremely complicated.

你得打电话找人。

So you need to call someone.

也许你是个建筑工人，得打开笔记本电脑。

Maybe you open up your laptop because you're a construction worker.

你每两周才打开一次笔记本电脑，现在却要你用SAP或者其他ERP系统工作。

You open up your laptop only every second week, and now you are required to work with SAP, or any other ERP system.

所以你连需求都提不出来。

So you can't even issue the demand.

这就是我们让事情变得非常简单的方式。

So this is how we make it very easy.

你拍张照片，上传报价单、Excel表格，这就是关于采购你需要知道的一切。

So you take a photo, you upload a quote, an Excel sheet, and that's actually everything that you should know about procurement.

采购部门之外，没人关心品类、总账科目、框架合同这些。

No one cares outside of the procurement department about categories, GL accounts, framework contracts.

没人关心。

No one cares.

我们采购圈的人关心，但圈外的人不关心。

We in the procurement world care, but no one outside that cares.

然后我们的智能体就接手了，去检查库存。

And then our agents take off, and they check the inventory.

他们查了一下，发现可以问另一个工厂。

They find out, okay, they ask another plant.

“好，我们能不能内部采购这些螺栓？”

"Okay, can we source those bolts internally?"

不行。好，那我就跟采购代理沟通，看看我们有没有内部供应商。

No. Okay, then I'm talking to the sourcing agent, finding out, do we have internal suppliers?

有没有外部供应商？

Do we have external suppliers?

然后另一个代理要起草询价单，通过邮件发出去。

Then another agent has to draft the RFQ, send out the RFQ over email.

然后一堆邮件就来了。

Then a bunch of emails arrive.

有些完全是废话，有些就写在邮件里，有些是PDF，有些是Excel表格。

Some of them are completely nonsense, some of them are just in the email, some of them are PDFs, some of them are Excel sheets.

我们把这些信息提取出来。

We retrieve those informations.

然后我们进行下一步，或者根据我们的价格基准比较，可能有谈判的机会。

Then we do the next step, or maybe based on our price benchmarking, there's an opportunity to negotiate.

然后我们有代理来决定下一步该怎么做。

And then we have agents that essentially decide on the next step.

所以谈判可能意味着与人类参与的战略性谈判。

So negotiation could mean strategic negotiation with a human in the loop.

这可能意味着自主谈判。

This could mean autonomous negotiation.

这可能意味着拍卖和电子拍卖，然后调用特定的代理来进行谈判。

This could mean auctions and e-auctions, calling then the specific agent, doing the negotiation.

然后从头到尾完成这个过程，包括确认订单、跟踪发货、处理发票。

And then doing this end to end, think about confirming the order, shipment tracking, invoices.

我们能够完全自主地运行这一切，并捕获所有上下文信息。

And we are able to run this fully autonomously, capture all the context.

然后显然，更强大的下一步是将其应用于更复杂的零部件，涉及直接采购，这也是我们运营的领域。

And then obviously the next powerful thing is do this for more complex parts where we talk about direct procurement, where we also operate.

什么是直接采购？

What's direct procurement?

基本上我刚才描述的所有内容，核心目标就是自动化，对吧？

So essentially everything I just described, the main goal is here automation, right?

所以你可以完全自主地运行这个流程，而且在过程中还能产生更多的节省，对吧？

So you can run this process fully autonomously, and then throughout the process, you can generate even more savings, right?

所以我们是这样看的：这个任务要完成什么，它看起来是什么样的？

So we look at it like, okay, what is this and job to be done, how does it look like?

他们每天要做一千次的事情，但实际上他们希望一天做零次？

So what are they doing thousand times a day, but actually they want to do it zero times a day?

我们已经运行了完全自主的代理，但也有机会：他们一天做零次的事情，如果企业一天做一千次，那会对损益产生巨大的影响。

We've run fully autonomous agents, but there's also opportunities of what are they doing zero times a day, but if a business would do this a thousand times a day, that would have a crazy P&L impact.

对从未谈判过的支出进行自主谈判。

Autonomous negotiations on spend they never negotiated before.

这属于间接采购部分，比如维修零件、建工厂，就像我们做的那个大胆的例子，还有笔记本电脑和铅笔、营销服务，或者有人需要搭建这个播客工作室。

And this is in the indirect procurement part, think about MRO parts, building a factory, the bold example that we did, but also laptops and pencils, marketing services, someone who needs to build up this podcast studio.

这些都是间接采购。

Those are all indirect.

然后我们还有直接采购的部分。

And then we have direct parts.

就像造飞机的时候。

This is like when you build an airplane.

那些都是你真正需要用来造飞机、造无人机或者造机器人的供应商。

Those are all these suppliers that you actually need to build the airplane, or to build the drone, or to build the robot.

这种情况下，我们说的不是五万家供应商，而是最多一百家或者两千家。

And there we don't talk about 50,000 suppliers, we talk about 100 suppliers or 2,000 suppliers maximum.

而且这些供应商在战略上极其重要。

And those are extremely strategically important.

可能一个供应商的采购额就高达十亿。

And you have maybe on one supplier, one billion of spend.

所以你不会想用自动化谈判来处理这种采购。

So you don't want to run an autonomous negotiation.