为企业运营构建AI代理

在供应链这种真正推动世界运转的运作中，语音才是关键所在。

Voice was the unlock to many of the operations that are really needed to move the world if we talk about supply chain.

我们解决的并不是一个只属于供应链的问题。

This is not a supply chain specific problem that we are solving.

这本质上是一个企业协同的问题。

It's actually an enterprise coordination problem.

未来几年，语音AI最大的挑战其实是知道什么时候该说话，什么时候该沉默。

The bigger problem in the coming years for, like, voice AI is really knowing when to talk and when not to talk.

关键在于理解工作中这些微妙的细节，而不是降低延迟或让声音更逼真——我觉得这些现在根本不是瓶颈。

So it's understanding all these nuances in the work more than making the latency faster or making the voices more realistic, which I don't think that's the limiting factor today.

我觉得Happy Robot一直走在人性化的最前沿。

I feel like Happy Robot has always been at the forefront of kind of humanness.

你希望客户知道他们正在和AI对话吗？

Do you want the customers to know they're talking to an AI?

这个问题该怎么看？

Where does that go?

我觉得这一点至关重要，

I think it's super important that

大多数AI演示都发生在受控环境中。

Most AI demos happen in controlled environments.

真正的挑战在于，当AI需要在大型企业内部运作时，信息分散在各个系统、团队、邮件、电话和多年形成的流程中。

The real challenge begins when AI has to operate inside large organizations, where information is fragmented across systems, teams, emails, phone calls, and workflows that have evolved over years.

物流和供应链已经成为这些系统的早期试验场。

Logistics and supply chains have become an early proving ground for these systems.

成功不仅依赖于模型的智能，更取决于协调能力、上下文理解，以及在现实世界中可靠执行任务的能力。

Success depends not just on model intelligence, but on coordination, context, and the ability to execute work reliably in the real world.

奥利维亚·摩尔和阿尼什·阿查里亚与Happy Robot的帕布罗·帕拉福克斯和路易斯·帕罗普对话，探讨语音AI、企业智能体，以及在运营复杂的行业中部署AI所面临的挑战。

Anish Acharya and Olivia Moore speak with Pablo Palafox and Luis Paarup from Happy Robot about voice AI, enterprise agents, and the challenges of deploying AI in operationally complex industries.

我和奥利维亚今天邀请到了Happy Robot的两位杰出创始人，帕布罗和路易斯。

Olivia and I are here with the two incredibly talented founders of Happy Robot, Pablo and Luis.

欢迎二位。

Welcome, guys.

谢谢你们。

Thank you, guys.

特别期待。

Super excited.

很高兴你们能来。

Very excited to have you.

我们早就该聊聊这个了。

We're overdue to have this conversation.

今天我们主要是想聊聊你们的公司，还有你们走过的这段非凡历程。

Well, look, we're here to kinda talk about the company and the incredible journey that you've been on.

我记得我们第一次见面时，YC的创始人和其他人都在热议，说你们站在技术前沿，同时市场推广也势头强劲。

I know when we first met you, there had been a lot of buzz amongst YC founders and other folks about how you guys are sort of at the edge of the technology and then really getting a lot of pull from a go to market perspective.

能不能带我们回溯一下，当初在第二十街那个只有四五个人的小办公室，公司和产品最初是怎么起步的？

So maybe take us back in time to the little office that had four or five people on 20th Street, and what the origins of the company and the product were.

当然。

100%.

我和路易斯是在大学开学第二天认识的，从那以后我们就一直一起搞项目。

So Luis and I met on our second day of college, just to set the scene, ever since we've been building stuff together.

我们的另一位联合创始人塔维是我弟弟，所以我早就认识他了。

Our other co founder, Thavi, he happens to be my brother, so I've known him for a little while.

我们一直想一起做点东西，对吧？

We always wanted to build something together, right?

所以当我们加入Y Combinator时，就在找一些复杂的难题来解决。

So when we got into YC, we were looking for complex problems we could solve.

别忘了，路易斯和我以前为了参加机器人比赛，真地造过潜水艇，专门去水下找假人。

Keep in mind that Luis and I have been literally building submarines for robotics competitions to find mannequins underwater.

我们想找的就是这种级别的难题。

That is the sort of problems we were looking for.

于是，当我们决定攻克这类复杂问题时，就去看了塔维当时作为全球最大橄榄油分销商CFO在做什么。

So when we decided on solving for that complexity, we looked at what Javi was doing as CFO of the largest olive oil distributor in the world.

他每天都在跨国运输成吨的橄榄油，正是这种复杂性让我们对物流和供应链产生了兴趣。

He was literally moving tons of olive oil across the ocean, and that was that complexity that drew us into logistics and supply chain.

他不得不专门雇人打电话给司机，确认他们在哪里，货物到哪儿了，因为沃尔玛一直在问他：我的橄榄油货到底在哪儿？

He literally had to hire insurance to call drivers to see where they were, to see where the shipment was, because Walmart was asking him, where the hell is my shipment of olive oil?

这就是我们想解决的那种问题。

So that was the sort of problems that we wanted to tackle.

也许你可以讲讲，为什么我们当初选择了语音这个方向？

And maybe you can talk about why we actually started with voice there.

我想我们一开始是从技术角度出发的。

I guess we took it from a very tech driven approach.

那时候最大的瓶颈，就是能不能做出一个在电话里说话像真人一样的智能代理。

Really the limiting factor back then was having an agent that could speak on the phone realistically.

我们跑遍各种展会，海蒂到处奔波，问人：如果我们能开发一个语音AI，能打电话去谈货运、追踪货物，你们会买吗？

Like we were in conferences, Heidi was like traveling all around, like asking people, hey, if we were to create a voice AI that could pick up the phone and sell these loads and track these shipments, would you buy it?

人家就说：老兄，这还用说吗？简直太有必要了。

And he's like, Dude, of course, this is a no brainer.

我只是觉得，你们根本做不到。

I just don't think you can do it.

所以从一开始，这个想法和市场需求就是匹配的。

So it was more so like the idea market fit or product market fit made sense from the beginning.

真正的问题在于，我们能不能证明自己真的能做出这项技术。

It was more so like can we prove ourselves we can build this technology.

那时候大语言模型开始兴起，大概就是2023年底吧。

And you know LLMs were picking up, we're talking about late twenty twenty three probably.

大语言模型已经够用了，十一小时的语音合成也在进步，各种技术慢慢能凑到一块儿了，但我们还得把所有环节真正连起来，做出一个能实际运行的东西。

LLMs were like decent enough, eleven hours was picking up with the text to speech and everything was kind of working together but we had to build something that could actually connect all the dots and actually make something work no.

这也塑造了我们公司的方向——技术与创新是核心，我们一直想冲在最前沿，亲自解决这些问题。

That kinda shaped our company where we really had technology and innovation as a core of our company and always pushing this frontier and on solving it probably like firsthand.

这就是我们进入语音助手领域的由来。

So that's how we got started in the OA space.

太棒了。

Amazing.

我特别难忘的一个回忆，是咱们第一次在一家特别吵的咖啡店外见面，你当时拨通了一个真人语音代理，整个过程丝滑得不行，环境那么差，效果却出奇的好。

One of my favorite memories of working with you guys is actually when we first met outside a very crowded coffee shop and you called one of the live voice agents and it was seamless and it did an incredible job in a very non ideal environment.

很多人可能是因为你们那些出色的语音助手演示视频才知道Happy Robot的。

I feel like a lot of people might know Happy Robot from your amazing demo videos of the voice agent.

但这只是产品的一部分，虽然非常重要。

And that's definitely not all the product is, but it's an important part of it.

所以能不能跟我们说说，为什么一开始选择语音？语音又为你们打开了哪些更广泛的可能性？

So maybe walk us through, like, why voice to start and then what voice has maybe unlocked for you more broadly.

没错。

Yeah.

Luis刚才说的非常关键。

What Luis was saying is very important.

语音技术解锁了供应链等领域许多真正关键的运营环节。

Voice was the unlock to many of the operations that are really needed to move the world if we talk about supply chain.

那时候我们去参加各种展会，大家都觉得不可能，说你们怎么可能做出能打电话、谈运费的系统，但这就是我们重点突破的方向。

So when we were going to these conferences and people were like, no way you're gonna build these things that talk on the phone, negotiating rates on shipments was actually a big one.

所以我们当时就对大语言模型做了专门的微调。

So we actually fine tuned LLMs back then.

我们当时微调了Mistral和Llama，就是为了让这些语音助手运行得更快，因为那时候用GPT-4实在太慢了，而GPT-3.5在推理和谈判方面又差得离谱。

Like, fine tuned Mistral and Llama to actually make those voice agents faster because otherwise using some GPT-four at the time was like extremely slow, and GPT-three 0.5 at the time was like terrible at reasoning and actually negotiating.

所以我们不得不在背后做大量工作，搭建自己的智能代理系统，同时也开发了专属的语音交互能力，这样才能比竞争对手更快地迭代创新。

So we had to do a lot of tricks behind the scenes, build our own agent infrastructure, if you will, but also build our own voice agent capabilities so that we could innovate faster than competition.

这让我们在物流运输领域的早期阶段占据了明显的优势。

And that actually gave us a really good edge in logistics and transportation in the early beginnings.

我们先是和货运经纪公司合作，接着扩展到货运代理，然后是海运公司，再后来是卡车运输公司。

So we started working with these freight brokers, then we expanded to these freight forwarders, then ocean carriers, then trucking companies.

如今，我们服务的客户已经涵盖供应链领域许多最大的企业。

And today, we actually serve many of the largest companies in the space of supply chain.

之前我们聊到过，美国前十大货运经纪公司里有九家，十大卡车运输公司里有七家，还有一些规模最大的车队，它们负责把货物运遍全美，这简直不可思议。

We were discussing before, know, nine of the top 10 freight brokers in The US, seven of the top 10 trucking companies, like some of the largest fleets that actually move our goods everywhere in The US, which is crazy.

两家最大的海运公司，就是我们在港口看到的那些巨型货轮，它们也是我们重点服务的对象，而语音正是它们许多业务流程中的关键环节。

Two of the largest ocean carriers, those big boats we see in the bay, That is sort of customers that we needed to build for and where voice was the analog for many of the operations.

所以听起来，这并不仅仅是语音的问题。

So it sounds like it wasn't just voice.

那不仅仅是语音，还包含谈判。

It was also voice plus negotiation.

所以我们最初是从跟踪追踪开始的，也就是客户服务和销售，本质上就是谈判。

So perhaps track and trace, which is customer support and sales, which is sort of this negotiation is where we started.

这迫使我们开发出比原本可能更深入的技术体系。

And I think that forced us to build a deeper set of technology than we otherwise would have built.

也许，路易斯，你来给我们讲讲这段技术发展历程吧。

Maybe, Luis, take us on the technology journey a little bit.

对。

Yeah.

在回答这个问题之前，我想说，当我们站在技术前沿时，从一开始就很清楚：哪些东西必须重新发明，哪些已经存在了。

So before I tackle that, guess one of the things that we had very clear from the beginning when you're working on the frontier of technology is really what you have to reinvent versus what already exists no.

有些人可能会为了创新而全盘重来。

And I think people might take an approach where they just reinvent everything just for the sake of it.

也有人只是简单地把现有东西包装一下，当作市场推广的手段。

Some people would just wrap around anything else and be like more of a go to market thing.

我们一直从最关键的瓶颈入手。

We started tackling the limiting factor always.

那时候，Pablo提到GPT-3.5虽然速度还行，但效果并不理想。

And again, back then GPT, Pablo mentioned, now like 3.5 was relatively fast but not so good.

所以我们得去优化另一端。

So we had to fine tune the other end.

很快我们就发现，提示词技术加上那些优秀的模型已经足够好了，之前的方案可以扔了，就用这个。

Soon enough we realized that prompting and all these good models came out, prompting was good enough, scratch that, let's do that.

我们始终紧盯那些瓶颈，比如语音识别中的背景噪音，还有供应链这一块，情况特别复杂。

And always focusing on that limiting factor then voices, like the background noises, like supply chain is extremely messy.

你得跟开着卡车、放着收音机和背景音乐、口音各异的司机对话。

You're talking with drivers in their trucks with the radio on and background music and noises and accents.

所以一直聚焦在这些真正的难点上。

So always focusing on those limiting factors.

至于谈判部分，我们经常被问到：怎么防止机器人胡编乱造运价或者最高收购价？

On the negotiation part, something we got very often was how do you prevent the bot from hallucinating a rate or like max buy.

这就好比，哥们儿，我搞这个东西，结果它老是瞎编最高报价，根本不会谈价。

It's like, dude, I'm building this thing and it's just hallucinating max buy and it doesn't know how to negotiate.

你们是怎么做到的？

How are you guys able to do that?

我觉得原因在于，你根本不需要让AI看到它不该看的东西。

And I think it's because you don't need to show the AI what it doesn't need to see.

从一开始我们就很明确这一点：我们搭建了中间服务器，只给智能体提供它真正需要的信息。

And I think we're very opinionated about this from the beginning where we're building these proxy servers and actually exposing to the agent only the things they need to see.

实际上，那个最高报价的金额，智能体根本看不到，也无从谈起去谈判。

And actually Max Buy, the max amount of money the bot can actually see or actually negotiate is not even exposed to the bot.

我们没把它暴露出来，而是用外部的谈判算法，让智能体就像人一样，老老实实问一句：‘我去请示一下上级。’

We were not exposing that, we were doing external negotiation algorithms so that the bot would just ask for permission literally the same way a human would like, Hey, let me ask my boss.

它其实就是调用一个工具，请求授权再进一步操作。

And it was really just calling a tool and asking for permission to do more.

然后我们再把结果反馈回去。

And we would inject back the RAID.

不是吗？

No?

所以这类情况，我们不是直接把信息丢给模型让它自由发挥，而是采用更确定性的方法来处理。

So those sort of things, instead of, like, just putting it in the context, we're not having the LLM just freestyle it, we do it in a more deterministic approach.

关键是要在概率性和确定性之间保持平衡，该交给AI的才交给AI。

So it's always that mix of probabilistic plus deterministic where you need to let everything to the AI.

不是吗？

No?

这是在为真实世界做设计。

It's building for the real world.

真实世界本来就是一团乱麻。

The real world is messy.

总有人会试图绕过限制，想拿到那个最高金额的权限。

Those things are gonna happen where someone tries to jailbreak the agent and get that max amount of money that they can get.

所以我们很早就得搭建好这些安全机制，才能真正去跟像C这样的机构合作。

But we needed to build those guardrails very early on so that we could actually go to the likes of C.

嗯。

H.

像罗宾逊运输和Uber Freight这样的大公司，只有当我们真正打造出扎实的技术时，才会信任我们。

Robinson or Uber Freight and all of these big players that would only trust us if we actually were building real technology.

这一点对我们来说非常明确。

That was pretty clear for us.

我们知道，不能把精力放在物流和运输的长尾需求上，因为那个领域太复杂了，但我们清楚必须服务好运输、供应链和物流领域的大型企业。

We knew that we didn't wanna focus on the long tail of logistics and transportation because it's a very tricky space, but we knew we needed to serve the enterprise in transportation and supply chain and logistics.

这一点对我们来说非常清晰。

So that was very clear.

这直接决定了我们该开发什么样的产品，该构建什么样的基础模块。

That that shaped the type of products that we had to build, the type of primitives that we had to build.

这很有意思，因为Happy Robot在语音AI和企业级智能体这两个领域都算是早期入局者，这点很棒。

It's so interesting because Happy Robot was very early to both voice AI and enterprise agents more broadly, which is great.

而且，现实情况也在快速变化，模型本身正在飞速演进，就像你提到的，微调、提示工程，接下来该怎么做，都在不断变化。

And also, it's like the ground has been shifting under our feet because the models are themselves are kind of changing and evolving so rapidly to your point about fine tuning versus prompting versus kind of what to do next.

也许我们可以聊聊几个具体的使用场景，说明光靠一个更智能的模型本身是不够的，为什么你必须得买一个完整的平台。

Maybe we can talk through a few of the use cases you have where it's very clear that a smarter model by itself doesn't just do it and like why you need to buy a platform.

我可以讲讲昆纳加勒这个案例。

I I can bring up the Kununaggle use case.

我们最近刚宣布和一家顶级货运代理公司达成合作。

We recently announced our partnership with the marquee, like, freight forwarder.

他们是很棒的合作伙伴。

Great partners.

我曾经和他们的空运负责人一起吃午饭。

I was having a personal lunch with their head of air.

顺便提一下，我在塔斯马尼亚的伊姆维家做客时，和他聊了很多。

Shout out to our friend, Imve at Kununaggle at his house in Tasmane.

从他们的运营中我了解到，这根本不是那种简单的客户服务——不是在Zendesk提个工单就完事了，也不是靠知识库回复一下就能解决的。

And what I learned from their operations is that this is not a simple customer service type of create a ticket in Zendesk and you're done, or you reply based on a knowledge base.

像物流、运输、货运代理这些实体经济领域，还有电信、公用事业等行业，客户支持远没有那么简单，光靠知识库回复根本行不通。

Customer support for these real economy industries like logistics, transportation, freight forwarders, broader supply chain, even other industries like the telco space or the utility space, it's not as easy as just replying based off of a knowledge base again.

之后还有很多工作要做，才能把更新反馈给客户。

There's a lot that happens afterwards that really has to get done to provide that update to the customer.

比如货运代理Kununaggle，他们服务的都是大客户，具体是谁我不能说。

So example, freight forwarding, Kuninagal, they are serving customers, very large customers, I cannot name who.

但你可以想象，你是Kununaggle的大客户，你问：我的货现在在哪？

But imagine that you are a big customer of Kuninagal and you ask, hey, Where is my shipment?

这时候，客服人员得去帮您查清楚。

What happens now is an agent has to turn around and go find it.

这个查找过程非常复杂。

That go find it is very complex.

需要很好的协调。

You need that coordination.

你需要一个调度型的智能代理来处理这些事。

You need basically an orchestration agent that is, okay.

这是空运货物，自然就跟航空公司有关。

This is an air shipment, so, obviously, relates to airlines.

这个货件是哪家航空公司承运的？

Who is the airline on this shipment?

明白了。

Okay.

我上去查一下航空公司的官网。

Let me go to the airline's website.

我们有专门的爬取程序，会去抓取航空公司的网站信息。

So we have browsing agents that go and scrape the website of the airline.

糟了。

Oh, bummer.

没找到。

It's not there.

没有任何更新。

There's no update.

真烦人。

Damn.

我得发封邮件过去。

I need to go send in an email.

好的。

Okay.

我给航空公司发封邮件。

I'm sending an email to the airline.

两个小时过去了，还没回音。

Two hours later, no reply.

好的。

Okay.

我得好好想想这事儿。

I need to reason that.

要是现在还不回复，我就要错过给客户的服务时限了。

If they don't reply now, I'm gonna miss my SLA with my customer.

所以现在我得给他们打电话。

So now I need to call them.

我会一直打下去，直到航空公司有人接电话，告诉我我的货到底在哪。

I'm gonna keep calling them until someone picks up at the airline and tell me where the hell is my shipment.

这种协调工作，正是运输和物流领域真正需要的，也正是我们当初在AV公司不得不开发这款产品的根本原因。

So that is the sort of coordination that we need to make happen for for for transportation and and and logistics, really, and how and and that has shaped the type of product that we we had to build at an AV.

是的。

Yeah.

不是。

No.

我全都订阅了。

I subscribe everything.

再举个谈判的例子吧，就像我们一开始说的，那些演示视频后来火遍全网。

I guess another example on, like, the negotiation, which is always like how we started and all these demos went pretty viral.

真正的问题在于，光靠智能还不够——比如在谈运力时，如果有十家承运商或买家同时打电话进来。

And I guess one point of how raw intelligence really wasn't enough is when you're negotiating, for example loads and there's like 10 carriers or 10 like buyers calling in at the same time.

你不可能让每个代理都各自为政，而人类在这种情况下往往就是这样做的。

You cannot have all those agents like doing work independently which is what happens to a certain degree with like humans.

他们都在同一层楼，大声喊来喊去，说‘这个货很急，赶紧谈，有人想接’。

They're on a floor and sure they shout to each other and they're like, hey this is a very hot load, please negotiate hard, I have someone interested.

你懂的。

You know?

这些信息，模型里根本就没有。

All this information is really not in the model.

所以我们开始这么做：当多个来电都针对同一个货单时，就把信息互通起来。

So what we started doing is when you have inbound calls for the same load, you can start like sharing context across them.

比如提醒对方‘有人很感兴趣，赶紧加把劲，这单很急’。

Like hey I have someone they're very interested please push harder like this is a hot load.

所有这些信息共享，其实就是你随时放在上下文窗口里的内容。

Also all this information sharing is literally what you put in the context window at any point in time.

普通智能或原始智能根本不知道，有没有别人也在盯这个货单。

Like general intelligence or the raw intelligence doesn't really know if someone else is calling on that load.

所以关键在于，你对这个行业到底有多了解，对谈判策略又掌握多少。

So it's all that about like what do you know about the business, what do you know about the negotiation strategies.

也许你知道，因为这是跨境业务，所以更积极地跟进会更有优势，诸如此类的情况。

Maybe you know that pushing harder on this load because it's like cross border is gonna be better or whatnot.

这根本不是通用智能。

Like that's not general intelligence.

这是非常具体的经验，不同公司运作方式都不一样。

That's very specific and different enterprises operate differently.

你不能随便训练一个智能体，调个参，就指望它能适应任何公司。

Like you cannot just build an agent, fine tune it, and have it work at any type of company.

所有这些细节都超出了模型的能力范围，而我们正想构建的就是这一层上下文信息。

All those nuances is outside of the model and it's that context layer that we're trying to create.

实际上，我们可以通过讨论如何真正开展工作、如何执行任务，来获得这些东西。

And that actually like we can talk about how actually doing the work and executing the work is what gets you that.

这就像通过实践来学习。

It's like learning by experiences.

你去做一件事，从中学习，不断探索上下文的边界，这样才能持续进步。

You you do something and you learn and you explore that space of the context layer so that you can keep learning.

不是吗？

No?

真的很有趣。

Really interesting.

你刚才提到了两件不同的事，帕布罗。

So you you talked about two different things there, Pablo.

你谈到了一个跨职能的工作流程。

You just talked about a very cross functional workflow.

路易斯，你谈到了精通销售的复杂性。

Luis, you talked about the complexity of really mastering sales.

你知道的。

You know?

你们最初是从销售和支持做起的。

You guys started as sales and support.

所以，相比一些竞争对手，你们从更复杂的起点开始，还遇到过哪些意想不到的情况？

And so what are some of the other surprises that you've had having started with more complexity, I think, than some of your competitors?

我们最近从一家最大的卡车运输公司了解到，通常情况下，当我们采购技术时，都是先看看能用在哪儿。

So one thing that we heard from one of the largest trucking companies recently was, typically when we buy technology, we see where we can apply that.

但跟你们合作不一样，我们是先有具体问题，然后来找你们，因为我们知道，你们这个平台几乎能为我们打造任何类型的智能代理，不管是销售、客服、后台支持、运营，甚至催收都可以。

With you guys, we actually have a problem, and we come to you guys with that problem because we know that as a platform that you've built, we can pretty much build any type of agent for our operations, from sales to customer service, back office support, and and operations, and even collections.

所以有些客户来找我们时会说，嘿。

So some of the use cases that customers came to us were, hey.

我们有个很大的催收难题。

We have a huge collections problem.

你们能做个智能代理，通过邮件或电话联系客户，把钱收回来吗？

Can you build an agent to reach out to customers via email or voice and collect money?

我们当然说没问题。

We're like, of course.

我们还跟一家大型供应链公司讨论过类似场景——需要打电话给客户，催缴包裹的关税。

We talked about these use cases with one of our largest supply chain companies and customers where we need to call customers to recover duties on parcels.

现在我们每天都在开展2万到5万通电话的外呼活动，专门催缴包裹关税，要是客户不交，这笔钱就收不回来了。

And today we're running campaigns of 20 to 50,000 daily outreach to customers, collecting duties on parcels that otherwise they would not get if they don't pay the duty on the parcel.

所以客户给我们的一些反馈，其实挺让人意外的。

So that sort of surprises, if you will, we've gotten from customers.

比如，我也需要招司机。

Like, yeah, I also need to recruit drivers.

你们能做这个吗？

Can you do that?

我们当然可以开发一个代理，不仅帮着招司机，还能跟运营系统打通，这样他们就能提前安排车辆服务了，因为现在司机已经到位了。

And we obviously can build an agent that not only just recruits drivers, actually connects to the operation so that now they know they can service a truck with customer earlier, because now they have a driver to move that.

各个功能之间其实有好多这样的联动关系。

So there's all sorts of interesting connections between the functions.

我再给你举个例子吧。

Maybe I'll give you another example.

我们开发了一个代理，用来联系维修厂，确认卡车什么时候能修好。

We built an agent to reach out to maintenance shops to see where a truck or when a truck was ready.

你当然可以把这个代理单独搁着，就让它主动联系维修厂，打听卡车什么时候能修好。

You could just leave that agent in a silo and just have an agent that is proactively reaching out to those repair shops to see where the truck when the truck is ready.

结果发现，越早知道卡车什么时候准备好，就能越早把它投入市场，作为运力供客户使用。

Well, it turns out that the sooner you know when the track is ready, the sooner you can put it in the market to sell it as capacity for your customers to actually move things.

这让我们意识到，销售和维修这两个环节其实紧密相连。

So that was a very interesting realization of how sales in this case and maintenance were tightly connected.

这就是我们所说的背景关系。

So that is the context that we talk about.

企业里各个部门之间必须共享底层信息，这样才能实现整体最优——不管是追求全局最大值还是最小值，取决于你具体想优化什么，而不是只盯着单个部门的问题去解决，你明白吧。

There's there there has to be an underlying context sharing across the different functions in a business so that the whole business optimizes optimizes for a global maxima, if you will, or a global minima, depending on what optimization problem you're trying to run, versus just minimizing the the problem in one function, if that makes sense.

对。

Yeah.

那你能说说，这些工作流程是怎么被发现的吗？

And then maybe can you talk about, like, how do you discover these workflows?

是谁发现这些流程的？

Who discovers them?

又是谁来搭建它们的？

Who builds them?

那这些流程是怎么建立起来的呢？

How do they get built?

我的意思是，

I mean,

Luis，你来聊聊这个吧。

maybe Luis, talk a bit about that.

对。

Yeah.

我们一直采取前线部署的模式。

So we're very forward deployed.

很早就意识到，要真正解决客户的痛点，必须开发能适应他们实际运营的软件，而不是像过去那样，先做出产品，再要求客户按你的想法去调整业务。

So we very early on understood that really to solve the customer's pain point, we had to build software that adapts to their operations and not the other way around, which is like the old era before AI was you build something and ask people to, like, run their business however you think they should be run.

但我们觉得，应该反过来才对。

But we think it should be the other way around.

所以从一开始，我们就着手招聘并组建了前线部署团队，也就是FDE——现在大家都开始谈这个了。

So from the very beginning, we started, like, hiring and building this forward deployed motion with the FDEs, forward deployed engineers, everyone is talking about them now.

但我觉得关键在于真正以客户为中心，专注解决他们的核心问题，亲自去他们办公室，深入了解他们真正需要什么。

But I think it's about like really being customer obsessed and really focusing on like the value add and their problem and really sitting down with them and going to their offices and learning what they need.

当然，行业内有很多共通之处，从一个客户那里学到的经验可能对另一个客户也有用，但很快我们就意识到，没有一种方案能适合所有情况。

So sure there's a lot of like synergies in the industry and what you learn from a customer might be relevant to another but very soon we realize that there's not a one size fits all.

哪怕是在 inbound carrier 销售这个流程里，企业间的操作差异也很大，可能存在长尾需求，但企业之间的运作方式真的天差地别，所以我们才打造了一个足够灵活的平台，能适配任何企业的流程。

Even within inbound carrier sales, even like within this workflow in the enterprise maybe like there's like a long tail where this might apply but enterprises operate very, very differently and that's why we built a platform that is flexible enough to adapt to anyone's operation.

我们曾经试过直接把为一个客户开发的功能套用到另一个客户身上，结果根本行不通。

And it's because we were trying to like plug and play what we built somehow with a customer to another one, it didn't really work.

他们想要的是完全不同的东西。

Like they want something different.

他们想调整整个流程。

They wanna change the procedure.

有人希望这个工具能被重新命名，有人则希望在承运商未通过审核时自动触发升级机制。

They wanna call this tool, they wanna escalate whenever the carrier is not vetted and someone else wants to do it automated.

正因为这个行业里各种细节千差万别，我们才不得不打造一套高度通用的技术平台，你懂的。

So we really had to build up almost like horizontal technology because of the variety of all the nuances in this industry, know?

所以我们打造的平台不是为了优化某个具体任务，而是更注重支持实际的工作流程。

And that's how we create a platform that is not optimized for like specific tasks but more so optimized for like doing work.

我们的核心构建模块是工作流、数据、集成和标准操作流程、提示指令。

So our primitives are around workflows and data and integrations and SOPs, prompts.

你不会看到具体任务被固化下来，因为那样太主观了。

You don't see like, particular tasks being modeled because that's almost too opinionated.

客户不希望他们的供应商对如何运营他们的业务指手画脚。

And customers don't want, like, opinion like, their vendors forming opinions on how to run their operations.

他们早就在这行干了多年了。

Like, they've been running it for a long time.

他们对自己的业务了解得比我还深。

I'm I they don't know more about their business than I do.

我带来的只是技术，只想帮他们解决问题。

I just come with the technology and I just want come to, like, solve their problem.

你懂的。

You know?

是的

Yeah.

有意思。

Interesting.

我觉得，对于AI应用层的公司来说，前线部署的模式至关重要。

I feel like the forward deployed motion has been crucial for AI application layer businesses.

这也引发了很多关于利润率的问题。

And also it's prompted a lot of questions about what are margins?

服务和产品之间的界限到底在哪里？

What is like a service versus a product?

长期的竞争优势和护城河又在哪儿？

Kind of where is the long term alpha and moat?

你能聊聊你们是怎么把FDE们的工作产品化的吗？我觉得这正是Happy Robot的独特优势。

Maybe walk us through how you think about productizing the work that your FDEs do, which I think is kind of a unique strength of Happy Robot.

前线部署这个过程，是从哪里开始，到哪里结束的？

Where where does the forward deployed motion start and end?

你们会为某个客户做定制化的工作吗？

Do you do custom work for one customer?

我很想了解这具体是怎么操作的。

Would love to understand how that works.

也许我们从头开始讲起吧。

Maybe let's start with in the beginning.

是的。

Yeah.

我当初其实根本没意识到，自己就是第一个前线部署工程师。

I was the first forward deployed engineer without knowing it, I guess.

对。

Yeah.

这基本上就是任何创始人都会做的事。

Which is pretty much what any founder would do.

你会直接去客户那里，待上一周，找到真正正在做你想要自动化那件事的人，然后跟他们一起跟进。

You Like, you just go to your customers, spend a week there, and just chase down the people that are actually doing the thing that you wanna help them automate.

对吧？

Right?

所以我就这么做了，后来就经常去联系这些人，跟他们说：你们得把这个功能做出来，这能让我省不少事儿，而且别的客户也会用得上，因为我已经亲眼看到了。

So I did that, and now it would be, like, pinging these guys, like, dude, you need to build this thing because it's gonna make my life a lot easier, and it's actually gonna be replicable across customers because I've seen it.

所以你们赶紧把它开发出来吧。

So please build it.

他就会问：真的吗？

And he would be like, really?

真得开发这个？

Do I need to build that?

那时候，前线工程师的行动和产品团队之间就形成了这种良性张力。

So there was that good tension between kind of that forward deployed motion and the product team.

我们就这样维持了挺久的两个独立体系，我带着前线工程师团队，还有后来我们意识到必须有的部署策略团队。

So we kept going with these separate worlds for a little bit, where I would be leading the FDE team and the deployment strategists that we realized at some point we actually needed.

这其实是个挺重要的顿悟，算是插个话吧。

That was a bit of a realization, a bit of a parenthesis here.

我们一开始只安排了前线工程师，结果客户就问了：等等，你们有这些人一直在开发，那谁来统筹呢？

We started just with forward deployed engineers, and then they're like, the customer's like, Wait, you have these people building, but who is managing?

我说，这某种程度上就是部署策略要解决的问题。

I'm like, I guess that's the deployment strategy to some degree.

部署策略的作用就是明确问题的边界，好让前线工程师能更专注地做开发。

So the deployment strategy is just a figure that scopes the the problem so that the forward deployed engineer can spend more time on building.

不过现在我们发现，真正有效的前线工程师或部署策略，必须具备很强的跨职能能力。

Although now what we see is that the right FD or deployment strategies, they have to be very cross functional.

对。

Mhmm.

关于这种角色类型，先说到这里。

Close parenthesis on the type of of profile.

后来我们意识到，前线团队和产品团队之间脱节得太厉害了。

So when did what ended up happening is we were too disconnected from, like, the the forward deployed world and the product team.

所以我们决定，让路易斯也接手这部分工作，让前线工程师团队真正成为产品团队的延伸——本来就应该如此。

So we realized that that needed to be part of Luis's world so that the FDA team would actually be an extension of product, which is what they should always be.

这是产品团队的延伸，这样我们能更快地落地产品，更快地收集客户反馈，从而比任何人都更迅速地掌握这些信息。

It's an extension of product so that we can implement product faster, we can gather the feedback faster from the customers, and hence, capture that context faster than anyone else.

这其实是一个迭代循环，正是通过这个过程，Luis才真正意识到我们有多需要它。

So it's a bit of these iterate the iterative loop that we let that Luis really realize we needed.

对。

Yeah.

我再补充一点。

I'll add to that.

我们从最根本的问题出发：我们到底在做什么？

I mean, we go with, like, first principles, what are we doing?

我们在企业里各个职能和渠道部署智能代理。

We're deploying agents across different functions and channels in the enterprise.

所以我们的产品，本质上就是为部署这些智能代理而设计的。

So our product is built for the deployment of of an agent.

我们非常了解部署的全过程，因为我们跟客户贴得很近，实际在亲手部署这些代理。

Like, we really understand the deployment life cycle because we work very closely with our with our customers, and we are actually deploying these agents.

而且有个挺有意思的现象是，你的用户实际上就在你家里，因为我们主要是在为财务部门做产品。

And something cool that happens is that you have to a certain degree your user in your house because we're building for the FDs for the most part.

当然这也不是绝对的，有些企业确实非常看重平台，这个我们以后再聊，他们提到，能同时拥有一个可信赖的平台和一个值得信赖的财务智能代理，这种组合其实非常少见。

And it's not entirely true, some of course some enterprises they really appreciate having a platform and we can talk about that later, about how interesting the mix of coming with a platform they can also use and an FDE that they can trust is actually something very rare and they mentioned that.

不过说到部署生命周期，我们真的非常清楚部署一个智能代理到底需要哪些步骤，对吧？

But I guess to the point of like the deployment life cycle, we really understand what it takes to like deploy an agent, no?

首先得明确需求，然后开发、测试，

There's a scoping phase, there's a building, There's testing.

接着是监控，

There's monitoring.

还要有自我学习的闭环。

There's, a self learning loop.

所以我们的每个功能都是围绕这个部署流程来设计的，而要确认这些设计是否有效，唯一的办法就是紧贴部署现场。

So I guess the point is where every feature we're building in the product is optimized for that deployment life cycle, and the only way to know if that works is being being very close to the deployment.

对吧？

No?

而且这些工程师在做部署，或者从其他团队那里获取反馈。

And if these are doing these deployments or they're getting feedback from the other team.

所以与其说FD要贴近产品，不如说我们的产品本身就是平台和前线部署模式的结合，少了任何一个都不成立。

So actually more than the FD is being very close to the product, I think it's more so like our product is a combination of a platform and a forward deployed motion and it would really not exist.

现在大家总在讨论服务之类的话题。

There's like this conversation about like services and stuff.

不同之处在于，前线工程师更像是价值实现的催化剂或加速器，而最终留给客户的，是正在运行的代理程序。

The difference is that the forward deployed engineer are like catalysts or accelerators to value, but what we're leaving in the customer are like agents running.

还有一个平台。

There's a platform.

FD完成工作后，留下的是一套能正常运行的系统。

Once the FDs have done their work, they leave a working thing.

所以你花时间部署的，不是FD个人的成果，而是直接交付了一个在平台上运行的代理程序。

So it's almost like you spend that time, you deploy a thing, but you're not delivering like an output that the FD has done, you're literally delivering the agent working on a platform.

这跟纯粹的服务有本质区别——我们提供的是前线部署加上长期运行的平台，持续创造价值。

So it's a very different distinction of pure services versus like a forward deployed implementation plus the platform running the value forever, hopefully.

太棒了。

Awesome.

我觉得另一个常被讨论的话题是，在人工智能时代，记录系统的价值到底在哪里？

I feel like another thing that has been a topic of discussion is kind of what is the value of systems of record in the AI era?

每个应用公司都必须成为这样的系统吗？

Does every application company need to become one of those?

我知道你们在Happy Robot对记录系统和行动系统或执行系统之间有明确的看法。

And I know you at Happy Robot have a view on kind of systems of record versus maybe systems of action or systems of execution?

我们很想听听你对这个话题的看法。

So we'd love to talk through kind of your view on that topic.

那我先简单说一下。

Maybe I'll start quickly.

我们把自己定位为执行层，真的就是这样。

We see ourselves as that layer of execution, really.

真正的魔法就发生在这里。

Like, that's where the magic happens.

你得先动手做起来，才能捕捉到这些上下文。

You have to start doing the work to capture that context.

所以我们必须从执行任务开始，先把事情做完，先部署一个智能体，再部署第二个，然后通过这个上下文层把它们连起来。

So it's very important that we start with executing work, with getting the thing done, implementing one agent, implementing the second agent, connecting them through that context layer.

但这个上下文层，是在你真正开始做事之后才产生的。

But the context layer happens after you're actually doing the work.

如今，执行层的重要性比以往任何时候都更关键。

More than ever, the importance is on the execution layer.

对我们来说，Luis 可以补充更多，这部分数据确实很重要，但它是在智能体之后才出现的。

So for us, and Luis can comment on that more, that data piece is a very important piece, but it happens after the agent.

所以我们打造的产品叫 Twin。

So what we've built is Twin.

对我们而言，Twin 就是那个数据层，它能把客户的各种系统——比如 CRM、ERP、运输管理系统，或者你的 Snowflake 实例——全都连起来，同时智能体也能在这里写入或存储自己的上下文。

Twin for us is really that data layer where we connect systems of record of the customer, your CRM, your ERP, your transportation management system, whatever it is, your Snowflake instance, and where agents can also populate their own or re store their own context.

这几乎就是 Happy Robot 原生的数据点。

It's almost Happy Robot native data points.

所以我们实际上建立了一个数据层，既保存客户记录，也保存Happy Robot代理生成的信息。

So we've basically created these these data layer that holds both customer records and happy robot agent created records, if you will.

对。

Yeah.

在部署代理之前，花多少时间清理数据，和直接部署代理、边执行边清理，这两者之间其实存在一种有趣的张力。

Think there's an interesting tension in how much time you need to spend ahead of deploying the agents on cleaning the data versus just deploying the agents and cleaning the data through doing the execution?

我觉得这需要两者结合。

And I think it's a mix.

我们发现，这些代理会生成大量以前从未被记录过的信息，而这些信息根本无法融入现有的系统，因为它们更像高维、语义化、近乎记忆智能的东西。

I think what we realize is these agents are creating a lot of information that really hasn't been captured before, and it really doesn't fit in any of these systems because it's more like high dimensional, semantic, almost like memory intelligence.

所以我想说的是，很多企业都在等把数据源清理干净了，才敢启动这套代理工作团队。

So I guess the point is many enterprises, I guess, are waiting to clean their data sources so that they can power this workforce of agents.

但我觉得，只要开始行动，让代理真正去执行任务，数据自然就会在过程中被逐步清理干净。

And I think by doing the work and by actually having agents execute the work, you're going to clean the data as you go.

当然，人类很厉害，但他们也有不少局限性。

Because humans are great of course, but they have a lot of limitations.

它们好像同时存在于两个地方，还留下了不少问题。

Like they kind of been in the same place, in two places at the same time, they drop a lot of threats.

它们在把数据录入正确系统这方面并不够严谨。

They're not very diligent in putting the data in the right system.

有时候你忘了，有时候又随手记下来。

Like sometimes you forget, sometimes you write it down.

就算你先把所有数据源都清理干净，再靠人工操作，数据也很可能很快又变得混乱。

So actually you can clean all your data sources and then you can still run with humans and it's actually gonna probably get dirty very, very soon.

AI的好处在于，只要我们让它处理数据，它就会非常认真地完成。

The good thing about AI is it's very diligent where we push data.

通过实际执行任务的过程，你会逐步清理所有数据源，因为你终于能看清这些细节了。

So it's through the process of executing work, you're gonna progressively start cleaning all your data sources because you're gonna get visibility into all these things.

这不只是把不同系统里的数据——比如表格里的行和列、不同实体——连起来，更重要的是在它们之间建立起关联。

So not only are you connecting the data like the systems of record, like rows and columns and different entities, it's more so creating relationships across them.

所以，运输管理系统里的发货记录，本质上只是一个条目。

So again, the shipment in the TMS is just a record.

那其实根本不是IP地址。

Like that's really not the IPs.

同一个记录可能在很多不同企业里都存在。

How that record might exist in many different enterprises.

比如经纬度、费率什么的，这些本身根本没什么意义。

Like it's latitude, longitude, rate, whatever it is, like that really doesn't mean anything.

真正有意义的是，企业打算怎么用这些数据。

What means something is how an enterprise is going to what the enterprise is going to do with that.

一旦数据进到系统里，关键看他们的流程怎么设计，员工会怎么处理这些信息。

Like once it gets into the system, how their processes are built, how their humans are going to deal with that.

这些内容其实都不在系统里。

So all that is really not in the system.

很多背景信息都只存在于人的脑子里，属于那种只有老员工才懂的潜规则。

It's more so like in people's brains, a lot of these contexts is like tribal knowledge, the operator's whole.

某种程度上，这些信息非常零散。

And to a certain degree, they it's not it's super fragmented.

不是吗？

No?

实际上，通过动手做，我们会深入了解这种更像对话式的记录或智能，同时也会因为持续一致地执行，逐步清理掉那些末端系统中的混乱数据。

So actually by doing the work, we're gonna learn learn a lot about this more conversational record or intelligence, but also we're gonna start, like, cleaning the the end systems of record just by doing the work very consistently.

不是吗？

No?

你知道吗，路易斯，这个话题真有意思。我的直觉，或者说有点天真的想法是，关于执行的信息可能是短暂的，它的价值会随时间衰减。

You know, Luis, that's such an interesting topic because it's sort of like my intuition my naive intuition is that information about execution is maybe ephemeral or the value of it decays over time.

但我觉得你描述的恰恰相反——它的价值其实是随着时间积累的，甚至会让系统中的信息变得更丰富。

But I think what you're describing is how the value actually compounds over time, and maybe that actually enriches the information in the system of record.

哪种说法是对的？为什么？

Which one is true and why?

对。

Yeah.

我的意思是，你通过执行操作，正如我之前说的，一方面能更清楚地理解这些不同实体之间的关系。

I mean, I think you're so what you're doing by doing the execution, as I said, is one, creating a better understanding about the relationships of all these different entities.