下一个重大转变：我们的业务编排愿景

我们刚刚经历了自ChatGPT问世以来最大的转折点。

We have just had the biggest inflection moment since original ChatGPT moment.

我能猜一下吗？

Can I guess?

人工智能的承诺是真实的。

The promise of AI is real.

但在你的组织中实现它同样充满挑战。

So is the challenge of making it work in your organization.

欢迎收听《Path Forward》，在这里我们探讨人工智能时代工作的未来。

Welcome to the Path Forward, where we explore the future of work in the age of AI.

今天很高兴邀请到UiPath的首席执行官Daniel Dines和首席营销官Atalla。

I'm pleased to be joined by UiPath CEO, Daniel Dines, and CMO, Atalla.

欢迎两位。

Welcome, guys.

谢谢，兄弟。

Thanks, man.

很高兴来到这里。

Good to be here.

好的。

Alright.

对于所有正在收听的朋友，对于所有可能把我们看作某种公司的人来说，丹尼尔，是你发明了RPA。

For everybody who's tuning in, for everybody who may know us as a certain kind of company, Daniel, you invented RPA.

我们今天是什么样的公司？

Who are we today?

RPA不是我发明的。

I haven't invented RPA.

那是一场运动，但我觉得我们引领了RPA运动，最终我们赢得了一个大品类。

It was a movement, but I think we have led the RPA movement, and eventually we won a big category.

但我想说，从一开始，我们的愿景就是RPA是我们进入大型企业生态系统的入口。

But I would say from the beginning, our vision was that RPA, it's our gate into the big enterprise ecosystem.

因为我们一直想成为端到端流程编排和自动化的提供商，为公司提供基础，让他们能够以某种方式构建整个工作流程。

Because we always wanted to be a provider of end to end process orchestration and automation to provide the substrate that companies can build their entire work in a way.

今天，我觉得我们离那个最初的愿景更近了。

Today, I think we are much closer to that initial vision.

所以我们是从RPA开始的。

So we started with RPA.

在RPA之后不久，我们就转向了API自动化。

We moved fairly soon after our RPA foray into API automations.

我们带来的优势是，我们在同一个产品里结合了API和RPA。

And the power that we brought to the table was that we combined in the same package API and RPA.

我说的RPA，主要指的是基于界面的自动化能力，比如点击、输入。

And by RPA, I largely mean the capability to do UI based automation, clicking, typing.

就是自动模拟人的操作。

So emulating a human automatically.

直接模拟人的操作。

Emulating a human directly.

然后我们进入了文档理解领域。

And then we moved into document understanding.

所以我说，这就像一种痒，你知道，更深入到认知层面。

So I say it was an itch, you know, higher into the cognitive layer.

但首先，我们处理的是半结构化文档，用的是我所说的精准机器学习。

But first, we've been into semi structured documents where we use more what I call precision machine learning.

然后如果沿着这个轴继续，我们进入了长篇幅的非结构化文档，这些完全基于生成式AI。

And then if I continue on this axis, we moved into long unstructured documents, which are fully based on on generative AI.

然后那是在基于任务的基础上。

And then that was on the task based.

我们不得不做一个重大决定。

We had to make a big decision.

我们是打算沿用让我们在RPA上成功的同一技术，并将其应用到流程中，还是推出新东西？

Are we gonna take the same technology that made us successful in RPA and leverage it into process where we are coming with something new.

我们做出了，我认为，大胆的决定：让我们去构建一个新的。

And we made, I think, the bold decision, let's go and build a new one.

我们把它构建在，我认为，当今最先进的工作流引擎之上。

And we build it on the top of, I think, the state of the art workflow engine that exists today.

我们在其之上构建了一个界面，连接了能够描述流程的业务用户和能够进一步将这些流程描述编写成自动化并交付到该包中的开发者。

And we've built on the top of it a surface that connects kind of business users that can describe processes and developers that can further author these process descriptions into automations delivered into this package.

所以这就是我们在流程领域的首次尝试。

So that was our foray into into process.

现在我们进一步将流程描述为案例和阶段或状态，如果你愿意的话。

And now we are moving we are moving further away into describing a process as case and stages or states, if you want.

这是一种更动态的流程描述方式。

It's a more dynamic way to describe a process.

这种方式能更好地映射流程在业务中的实际样子。

In a way that maps better the way a process actually looks in a business.

是的。

Yeah.

我认为这自然就是人类扩展的方式。

And this was how, I think naturally humans scale.

在阶段内，可以有依赖于上下文的动态操作。

Within stages, there can be dynamic actions that depends on the context.

从一个阶段到另一个阶段，也是动态的。

Moving from stage to stage, it's also dynamic.

但一旦你构建、一旦你把流程表示成这样，它就会给你很多力量。

But once you build, once you represent the process as this, it gives you a lot of power.

所以我们构建了，你知道，这更像是在这个工作流引擎之上搭建的一个脚手架。

So we've built you know, this is more like a scaffolding on the top of this workflow engine.

这是一种表示。

It's a representation.

对。

Yeah.

现在如果我再谈谈认知方面，我们进入了任务和流程两个维度的认知部分。

And now if I talk more about cognitive, so where we went into the cognitive piece on both tasks and process dimension.

在任务方面，你知道，我们从三年前开始，在任务层面引入AI代理，这些AI代理就像RPA一样，可以模拟人类的操作部分，而AI代理可以模拟判断部分，比如分析长文档、得出结论、提出意见。

On tasks, you know, we started by three years ago to bring AI agents at the task level, AI agents that, like RPA, can emulate the action piece of a human, the AI agent can emulate the judgment piece where, you know, things like analyzing a long document, coming to conclusions, opinions

尝试做决定。

Trying to decide.

总结一下。

Summarize it.

文档角落里的那个小编辑，页边空白处的那个小词，是什么意思？

That little edit in the corner of a document, that little word on the side in the margin, what does that mean?

它应该是什么意思？

What should it mean?

对。

Yeah.

我让它变得相关还是不相关？

I make it relevant or not?

对。

Yeah.

所以我们带来了这些AI代理，并构建了我们自己的框架，比如低代码，但我们也一如既往地希望保持平台开放。

So we come with these AI agents and we build our own framework like low code, but we also, as always, we wanted to keep our platform open.

所以在我们的平台上，你如何构建代理并不重要，但我们准备好托管它、管理它、治理它、保护它，并将其与操作配对，而我们确实是操作方面的世界级专家。

So in our platform, it doesn't matter how you build your agent, but we are ready to host it, to manage it, to govern it, to secure it, and pair it with actions that and we are really the world class specialists into actions.

一个不能采取行动的代理，其实算不上真正的代理。

And an agent that cannot take an action, it's not really an agent.

那只不过是个聊天机器人。

It's just it's just a chatbot.

对。

Yeah.

而且我们这个平台能够以更自动化的方式交付这些代理。

And we have this platform that is capable to deliver these agents in a more unattended way.

这差别可大了。

It's a huge difference.

我觉得我们需要向企业客户强调这一点。

And I think we need to point this to our enterprise customers.

这是一个非常大的转变，我们可以帮助人们应对。

That's a very big transition that we can help people to manage.

这就是我在内部开发的东西，我自己的关于AI采用的思考框架。

And this is what I have developed internally, my own framework of thinking of the AI adoption.

所以我把AI的采用看作一个四阶段的框架。

So I see AI adoption like a four stage framework.

在第一阶段，AI只是回答一些问题。

In stage number one, it's more AI is just providing answers to some questions.

所以它就是个聊天机器人。

So it's a chatbot.

这是个生产力工具。

It's a productivity tool.

这是个生产力工具。

It's a productivity tool.

它是个人发起的。

It's initiated individually.

坐在那里等着别人来调用。

Sitting and waiting for someone to call on it.

对。

Exactly.

第二阶段，同一个人开始与AI进行对话，但他们期望AI也能采取行动。

Second stage, the same person initiate a discussion with AI, but they expect that AI can also take actions.

AI可以使用工具，能读取你的内部文档，如果你允许的话，它还能用工具去做一些事情。

AI can use tools, can read your internal documents, can use tools to do something, actually, if you allow them to do.

所以我认为这是一个强大的新结构。

So this is a is a powerful new construct, I would say.

我觉得很多企业会进入这个第二阶段，但这仍然是由人类发起的。

I think many enterprises will go to this stage two, but it's still human initiated.

对。

Right.

我们能做的就是帮助他们进入第三阶段，即智能体不是由人类发起的。

What we can do is help them progress to stage three, which is when the agent is not initiated by a human.

现在有很多关于AI巨额投资的讨论，但没人看到实际收益。

So many discussions are right now about huge investments in AI, but nobody's seeing in the bottom line.

是的。

Yeah.

对。

Yeah.

你得把一个人的任务拆成小块，然后以完全自主、无人值守的方式去执行。

You will have to take chunks of one person tasks and move it in a fully autonomous unattended way.

或者，也可以是一整块

Or or chunks of

业务，比如整个业务组合。

the business, like whole business portfolios.

主要来说，我认为这在我的框架里是第四阶段，因为在第三阶段，我认为范式是智能体提议，人类决策，自动化执行。

Mainly, I would say this is this is stage four in my framework where because in stage three, I think the paradigm is agents propose, humans decide, automation execute.

在第四阶段，智能体提议，并且智能体决定是否要询问人类处理异常情况。

In stage four is agents propose, and agents decide when to ask a human for a for an exception or not.

如果不问，它们就会直接调用操作。

And if not, they will call the actions.

对。

Yeah.

所以这个框架更难落地。

So it's even more difficult framework to put.

你必须建立更多的信任。

You have to build way more trust.

对。

Yeah.

但这就是第四阶段，我们看到企业工作交付更加自动化。

But that's the stage four when we are seeing more of an autonomous delivery of work for enterprises.

对。

Yeah.

我想从里面挑出很多东西，但我觉得最重要的一点是你引入了生态系统的概念。

There's a lot of things I I wanna sort of pick out there, but I think the biggest one is you introduce the idea of an ecosystem.

基本上，我觉得大多数人想到生态系统，就会想到那种描述方式，比如你有自动化、有代理、有API，所有这些。

Primarily, I think most people think of ecosystem as sort of the way you would you would describe where you have you have, you know, automations, you have agents, you have API, and all of that.

但人们会考虑生态系统吗？我觉得这可能是问你的好问题，迈克尔。

But do people think about ecosystem and I guess it's probably a good question for you, Michael.

大家会不会把生态系统理解为由合作伙伴主导，或者说，是与其他公司合作来帮助你构建那个东西？

Do people think about ecosystem as a partner led or a a, like, a a a collaboration between other other companies to enable you to build that

生态系统？

ecosystem?

可以说是，也可以说不是。对于任何客户，实际上，对于任何公司来说，要在人工智能时代取得成功，

It's sort of yes and yes for any customer well, actually, any company to be successful in this era of AI.

毫无疑问，他们首先必须认识到，这需要一个生态系统。

It is unquestionable that they must first and foremost recognize that it is an ecosystem that is required.

对。

Yeah.

我的意思是，在科技行业的每一次演进中，总会有一些公司。

I mean, there will always be, in every evolution of the tech industry, there has been, a set of companies.

今天，他们可能是云服务提供商、超大规模云厂商，或者业务应用提供商，他们会引导客户相信，他们实际上可以成为一个全面的解决方案，在他们的平台上解决世界上所有的问题。

Today, they might be the cloud providers or the hyperscalers or the line of business application providers who will lead customers to believe that they, in fact, can be a comprehensive solution that solves all the problems of the world on their stack.

他们甚至可能创新到一定程度，以至于在他们的平台上，他们确实相当有吸引力。

And they may even innovate to a degree that on their stack, they are quite compelling.

但实际上，在高端企业和受监管的企业以及大型企业中，仅仅依靠你自己的技术栈是不够的。

But the reality in upper enterprise and regulated enterprise and large business in general is that there is no on just your stack.

对。

Yeah.

总是需要完整的生态系统。

There is always the full ecosystem.

有多个云服务提供商，多个业务应用。

There is the cloud multiple cloud providers, multiple line of business applications.

而且，企业工作流程，就像Daniel描述的那样，会跨越所有这些边界。

And then there's the reality that enterprise work flows, like the one Daniel is describing, cross all of those boundaries.

无论客户如何应对，都必须有一个解决方案来应对整个生态系统，这正是我们在UiPath特别着手解决的问题之一。

However a customer approaches this, there must be a solution for addressing that full ecosystem, which is one of the things that we have specifically set out to to address at UiPath.

对吧？

Right?

我们特别致力于打造一个生态系统平台。

We have specifically committed to an ecosystem platform.

我们在整个行业里建立了各种合作关系。

We have built the relationships across the industry.

我们和一些公司合作，有些人用‘竞合’这个词来形容。

We have partnered with companies that some people use the word coopetitive.

对吧？

Right?

这些公司可能在某些领域和我们有竞争，但在很多重要方面，他们是我们关键的合作伙伴。

The ones that could probably compete with us in some areas, but in really important ways are key partners of ours.

是的。

Yeah.

我们愿意参与这样的生态系统。

We're willing to participate in an ecosystem like that.

我觉得这就是我们选择押注某些合作关系的方式，而且我们在这些合作上投入的技术资源可不少。

I think that's how we choose, to bet on certain partnerships, and and the amount of technology investment we make in those partnerships is is not insignificant.

这确实成本很高，但这是构建一个平台和一套能真正解决客户问题的能力的正确方式。

And it's costly, but it's the right way to build a platform and a capable a set of capabilities that'll solve real customer problems.

对。

Yeah.

我的意思是，对我来说，做了这么久，我知道如果只让客户用一个解决方案，最终会是什么样子。

I mean, if anything, for me, having been doing this for long enough, I know what lies on the other side of making someone use one and only one solution.

最后你会积累一大堆遗留技术，背上很多技术债，还会有些东西因为无法适应你的业务规模而被过度使用。

You wind up with tons of legacy tech, need a lot of tech debt, you have stuff that people are using more because they couldn't scale to your business.

这些正是我们想讨论的，也就是为什么专注于编排和开放平台是一种更好、更不同的方法。

Like, these are the things I think that we're trying to that we're talking about, which is, like, why a focus on orchestration, a focus on an open platform is a better, different approach.

而且我觉得很多人可能有个假设，就像过去其他技术一样，你可以直接从第一阶段跳到第四阶段。

And I think for a lot of people, there's probably an assumption that, like, other technology things in the past, you can just jump from stage one to stage four.

就是那么简单。

It's just that easy.

但事实并非如此，对吧？

It's not, though, is it?

说说为什么没那么简单吧。

Talk about why it's not that easy.

我觉得人们有错误的认知。

People have, I think, the wrong perception.

一个更强大的AI，一旦投入业务，就会突然开始工作。

A more capable AI that can be thrown into the business will suddenly start to work.

我认为那是错误的。

I think that's the that's the wrong

就是模型变得更聪明。

That the model gets smarter.

模型变得更聪明。

The model gets smarter.

对。

Yeah.

对。

Yeah.

因此模型会开始表现得像人类用户一样。

And therefore the model will start to behave like more like a human user.

我觉得人们没有意识到AI的一些局限性，这些局限性不是光靠投入更多GPU训练就能解决的。

I think people don't realize few limitations of AI that I don't think they can escape with just more, you know, flowing more GPUs in training.

因为你想，AI本身没有自我。

Because think about AI doesn't have a self in itself.

我知道这听起来有点哲学，但当你提问的时候，在每次调用之间并不存在一个自我。

I know it sounds a bit philosophical, but when you apply, when you ask a question today, there is no self that exists in between the calls.

你发出一个请求，得到一个结果，自我在哪里？

You ask a call, you get an out, where is the self?

对于人类来说，自我在商业运作中很重要。

In humans, there is a self that is important in in how, I think, business is operated.

自我让我们在游戏中有利害关系。

Ourself give us the stake in the game.

我们从很小的时候就被训练去理解错误意味着什么。

We are trained since very, very early age to understand what mistakes mean.

AI并不理解错误意味着什么。

AI have no understanding what a mistake means.

你有切身的利害关系。

You have skin in the game.

你没法训练AI。

You they you cannot train an AI.

AI不懂失去某样东西的痛苦。

AI doesn't understand the pain of losing something.

不懂吗？

Doesn't?

它不懂。

It doesn't.

你明白一个错误的决定可能会让我们付出代价，你知道，公众舆论可能无法挽回。

You understand the wrong decision maybe can cost us, you know, the public opinion can be unrecoverable.

对。

Yeah.

对吧？

Right?

对。

Yeah.

所以这是技术的一个非常重要的局限，而且我不认为AI能突破这一点，至少不是靠当前的架构，也不是靠规模法则。

So that's a very important limitation of technology that and I don't see AI going there, not with the current architecture, not in by the scaling laws.

还有一个非常有趣的局限，你可以观察到，而且它和这个有点相关。

And there is another very interesting limitation that you can observe there, and it's related somehow to this one.

当AI需要采取的行动有后果，而且代价高昂时，AI就不太行了。

When the actions that AI have to take have consequences, costly consequences, AI kind of doesn't work well.

因为这些大模型的推理方式，它们在组合、理解不同领域并进行综合方面非常出色。

Because how these big models reason, they're fantastically great of composing, of understanding different domains and composing, sympathizing.

然后取一个平均值。

And having to an average.

取平均值。

Into an average.

对。

Yeah.

它们就是一切的平均器。

They are an averager of everything.

对。

Yeah.

它们很难追踪后果，因为需要看到行动的结果才能组合和理解。

It's very difficult for them to trace the consequences because they need to see the outcome of an action to compose, to understand.

这很好，所以它们预测下一个不透明的东西。

It's good, so they predict the next opaque.

它们预测下一个行动，但无法进一步很好地想象后果。

They predict the next action, but they cannot go and further imagine the consequences very well.

如果你想让AI成功，你需要创造一个受控的环境，让行动不会产生重大后果。

If you wanna make AI successful, you need to create an engineered environment where the actions don't have some big consequences.

你不能让AI在你的企业里乱跑。

You cannot let AI running at wild in your enterprise.

从企业的角度来看，这意味着什么？

What does it mean for an enterprise perspective?

这意味着要构建一个定义非常明确的基础架构。

It means building a substrate that define very well defined.

这些是我的业务实体。

These are my business entities.

这些是有人可以对业务实体执行的操作。

These are the actions that someone can apply to a business entities.

这些是影响。

These are the implications.

所有这些操作都应该有审计记录，应该有安全性。

These these are all the actions should have audit rates, should have security.

谁被允许执行？

Who is allowed to do?

谁被允许批准发票？

Who is allowed to approve an invoice?

智能体应该知道这一点。

An agent should know this.

如果这些知识没有被完整地记录和整理，它就无法运作。

It cannot operate if if this knowledge is not fully captured and documented.

这就是我所说的在交付工作方式中的工程化环境。

This this is what I mean by an engineered environment within the how the work is delivered.

由于架构限制，AI无法工作。

AI cannot work because of architectural limitations.

所以我们的客户必须构建这个基础，然后理解劳动力转型非常重要。

So our customers will have to build this substrate, and then it's very important to understand the workforce transformations.

我认为有些公司倾向于直接跳入，比如，你知道，搞个大动作，因为那反过来会在内部产生推动力，促使大家去采用AI。

I think some companies tend to jump into let's let's create, you know, like a big rift because that in turn will create the force, the energy internally to go and adopt AI.

我认为那也是错误的，因为你可能会失去交付AI所需的关键人才。

I think that's also wrong because you risk of letting go key talent that is required to deliver AI.

构建基础。

To build the substrate.

构建基础。

To build the substrate.

你需要这些知识。

You need the knowledge.

所以你得确保把工作做好，把基础打好，然后你看到AI带来的生产力提升，你要明白在AI原生的世界里，哪些人类特质能蓬勃发展，然后再进行劳动力转型。

So you need to make sure that you build the work, you build the substrate, substrate, you bring you see the productivity AI, you understand what are the human traits that thrive into an AI native world, and then you go to the workforce transformation.

那些对公司充满热情、有切身利益、像公司文化传承者一样的人，那些忠诚的人。

People that are passionate about a company, that they have skin in the gig, that they are like cultural bearers of the company, people that are loyal.

对我来说，即使AI会让他们的某些产出完全过时，他们仍然非常有价值，你必须把他们重新部署到业务的其他部分，因为他们是公司结构的一部分。

To me, even if the AI is gonna render some of their output completely few type, they still are extremely valuable and you have to redeploy them in other parts of the business because they are part of the structure.

对于初级人才梯队来说也是如此。

And that's true for for the junior pipeline.

现在人们停止招聘初级员工，因为显然那是最简单的工作，但我认为你这样做会让公司失去一个非常重要的高级人才来源，这些人会延续你的文化，延续你是谁。

People right now stop hiring juniors because obviously that's the simplest work, but I think you severe your company of a very important pipeline of senior people that will continue your culture, will continue who you are.

AI不会给公司带来身份认同。

It's not AI is not gonna give an identity to a company.

是员工赋予公司身份，你必须从一开始就培养他们，让他们在精神上成长。

It's the people that will go there, and you have to take them from the beginning, grow them, you know, in the spirit

公司的精神。

of company.

你的整个前提是，公司需要有一个技术和人文相结合的基础，这个前提基于一个想法：你不能让技术基础或人文基础的任何一部分丢失。

Your entire premise that there needs to be a substrate across your company that is both technological and human is predicated on the idea that you can't allow any piece of your technological substrate or your human substrate to be lost.

你必须找到办法保留它，这需要年轻人来填补我们这些离开职场的人留下的空缺，因为自然减员是存在的。

You have to find a way to retain it, and that requires young people, that will replace those of us that leave the workforce and because there is a natural attrition.

现在，我觉得让初级开发人员去做AI能做的事情是没有意义的。

Right now, I think it's useless to have a junior developer and ask them to build something that AI can build.

为什么？

Why?

这完全是浪费。

It's completely wasted.

但什么不是浪费？

But what's not wasted?

导师制变得更加重要了。

The mentorship is becoming even more important.

你需要指导初级员工，让他们明白与AI合作意味着什么。

You need to mentor junior people into what does it mean to work with AI.

有品味意味着什么？

What does it mean to have a taste?

在软件选择中，什么好什么坏？

What's good or bad in the software choices?

人们将不得不做出很多选择。

People will have to make a lot of choices.

这不仅仅是技术上的时刻。

This is not just simply a technology moment.

这种编排的时刻，是一个真正的转型时刻。

This orchestration moment is, like, a true transformational moment.

它涉及方方面面。

Like, it's it's across everything.

那么，如果把我们UiPath和我们的客户放在那个时刻，你会怎么定位？

So if you were to sort of place place us, UiPath, and then place our customers in that moment?

我们在哪里，他们在哪里？

Where are we, where are they?

我们走的是哪条路？

What road are we?

我们走的是同一条路吗？

Are we on the same road?

我们的目标一致吗？

Are we aimed at the same goals?

我觉得如果看纯AI阵营，他们说的编排是指智能体之间的编排。

I think if you look at the pure AI camp, when they say orchestration, they mean agent to agent orchestration.

对。

Yeah.

就像有一个指挥在监督一群智能体。

Like having a conductor that is overseeing a swarm of agents there.

当我们说编排时，我们指的是流程编排，也就是协调所有参与者来为企业交付结果。

When we talk about orchestration, we mean like process orchestration, and that means orchestrating all the actors involved to deliver an outcome for the business.

那涉及的参与者有哪些呢？

What what are the actors involved?

显然，首先是人。

It's the it's the people, obviously.

还有工作流程本身，它需要经过不同的阶段才能推进。

It's the workflow itself that has to pass through different stages to progress.

嗯。

Mhmm.

还有自动化操作，就是那些你必须采取的行动。

And the automations, the actions that, you know, that you have to take.

以及现有的业务系统。

And business systems that are there.

你需要了解如何与它们连接。

You need to understand how you connect to them.

我可以想象它在任务级流程中运作，甚至像编码这样的场景。

I can imagine it working on a task level process, even like encoding.

现在，我可以有一个指挥代理，它能创建许多子代理，在我构建应用时并行处理不同部分。

Right now, I can have a conductor agent that can create many sub agents that can do different parts of, you know, when I build an application in parallel.

但那是单人的任务。

But that's a one person task.

一旦你有了这些流程描述，就很容易在流程层面引入代理。

Once you have this description of processes, it's very easy to bring agents, agents at the process level.

我有一个病人，某天需要去做手术。

I have a patient that has to go to to have a surgery in a day.

当他们入院时，会去登记处。

When they are admitted in hospital, they go to a registration desk.

所以有登记这个阶段。

So there is the registration stage.

对吧？

Okay?

他们会问不同的问题，检查不同的事项。

They will be asked different questions and check different things.

有些是通用的，比如查保险，有些则根据手术类型而有所不同。

Some things are common, like checking the insurance, for instance, some other are specific to their type of surgery.

对吧？

Right?

这不太容易用规则来定义，但你可以把它定义为你所拥有的行动列表。

It's not so easy to define in rules, but you can define as the list of of actions that you have.

而且还有那个前台，你知道，前台的人会监督这个过程。

And there is the the desk, you know, the human at the desk that will oversee this.

从登记之后，他们进入术前准备阶段。

From this registration they go into like preoperatory.

还有一些人需要检查生命体征，根据他们在这里做什么，工作量会更多或不同。

There is also some people should check their vitals depending on what they are doing there will be more or different type of work.

对吧？

Right?

同样，很难完全用规则来表达。

Again, difficult to completely express in rules.

所以，在登记台有一个人。

So there is a person in the registration desk.

现在我可以引入一个代理，但代理现在明白了，我已经定义了登记台的基础和规则。

Now I can bring an agent, but agent understands now I have I have defined the substrate of the registration desk, the rules.

现在，代理要接手一些原本由人完成的任务就容易多了。

Now it's much easier for an agent to take some of the tasks that the person would do.

对。

Right.

对，因为你定义了案例、阶段，也理解了交接过程，确实如此。

Both because you have defined the case, because you have defined the stages, because you understand the handoffs Exactly.

还有需要采取的行动，都是因为这些。

And the actions that need to be taken because of all those things.

然后，想象一下，在每个阶段，你可以有一个专门的阶段代理，它了解这个阶段。

And then now and imagine, in each stage, you can have a dedicated stage agent that understands the stage.

而且，在我的例子里，差不多会有不同的人来处理。

And it's more or less, again, in my example, there will be different people.

有不同的角色，比如护士负责量生命体征，医生做手术，术后又有另一个护士处理其他所有事情。

There is a different one person is the nurse that is taking the vitals, then it's the doctor that is conducting surgery, and then post surgery, there is another nurse that is doing, you know, all the other stuff.

对。

Yeah.

所以每个阶段都有不同的智能体，但还需要一个总体的案例智能体，来掌握病人的整个治疗过程。

So you have an agent, different agent on each stage, but you also need to have a case agent on the top, that understands the entire trajectory of a patient.

没错。

Right.

因为你基本上是在把编排定义为一种非此即彼的东西。

Because you're basically you're defining orchestration as like this is not a eitheror.

这其实可以是一种并存的情况：一边是常规流程编排，每次都按同样的方式运行；另一边是案例，就像你描述的，要么叠加在上面，要么在事情失控时介入。

This can be an and situation where you have regular process orchestration where it runs the same way every single time, and then you have the case, which is, as you described, it's a case that either layers over on top or it's layers when things get, you know, spin out or whatever.

这挺有意思的，我觉得我们需要比一般组织更好的方式来思考这个问题。

Like, that's an interesting I I need a better way, think, to think about this than maybe the the average organization maybe is thinking.

只不过你一直身处这些场景之中。

Except you live in these places all the time.

世界上的每一位企业领导者，无论他们能否表达出来，都知道这就是他们业务运作的方式，对吧。

Every business leader in the world, whether they can articulate it or not, knows this is how their business runs every Right.

对。

Right.

这就是他们业务运作的方式。

This is how their business is run.

组织就是关于

Organization is about

这就是每天在地面上发生的事情。

This is what's happening on the ground every single day.

但他们必须努力将其正式化到基础架构中。

But they have to make the work to formalize it in the substrate.

为了有效利用AI。

To use AI effectively.

如果他们不这样做，AI就不会起作用。

If they don't do it, AI is not going to work.

对。

Right.

百分之百。

A 100%.

但这种形式化，这种基础架构的建设，我认为对实现AI至关重要。

But this formalization, this building of the substrate, I think it's essential to deliver AI.

你绝对无法回避它。

You cannot cannot avoid it.

我知道我们每天为客户所理解和努力的事情。

I know what we understand and what we strive to do every day for the our customers.

我认为归根结底就是这样。

I think that's what it comes down to.

这就是编排的本质，是的。

That's what orchestration is Yeah.

对我们来说。

For us.

对。

Yeah.

这确实是个很有力的观点，但仔细拆解一下，其实理解起来很简单。

It's it's a really, it's really powerful message, but it's also when you break it down, it's a really simple way to understand it.

就是说，这必须是你构建一切的基础。

Is that this has to be the thing that's under the you're building everything on top of.

如果你不这么做，就像我说的，你只是在做代理之间的对接，无法扩展，或者你停留在聊天机器人的世界里，把聊天机器人的世界误认为是真实、合法的世界。

And if you're not doing that, then like I said, you're agent to agent, and you're not scaling, or you're living in a chatbot world, and you're mistaking a chatbot world for an actual, legitimized world.

这根本不是一回事。

It's just it's not the same thing.

你知道，十五年前，我们开始进入RPA领域。

You know, fifteen years ago, we're we're getting into the RPA game.

十年前，我们扩展业务。

Ten years ago, we're we're scaling the business.

你明白吗？

You know?

我们现在正处在这个旅程中。

We're on this journey that we're on now.

在现在这个时刻，基本上，还会有另一个编排时刻到来吗，还是说这是更广泛时刻的开始，或者说这个时刻正在扩大？

In this moment that we are right now, basically, is there another orchestration moment coming, or is this the beginning of the wider of them the the widening of the moment?

我们刚刚经历了自最初ChatGPT时刻以来最大的转折点。

We have just had the biggest inflection moment since original ChatGPT moment.

我能猜一下吗，编码代理？

Can I guess, Coding agents?

是的。

Yes.

对我来说，那确实是一个重要的时刻，而且它也很好地说明了自动化加认知的结合。

And that's a to me, that was really and it's a big illustration also on the case of automation plus cognitive.

因为编码代理，首先最重要的是驱动模型的难度，它们几乎是同时演进的。

Because the coding agents, it's first and foremost the hardness that drives the model and they evolve kind of simultaneously.

对。

Yeah.

所以我觉得这对我们所有人来说都是改变游戏规则的事情，不仅仅是对内部。

So I think this this is the game changer for for all of us, not only internally.

好的。

Okay.

我们会更快地开发我们的软件。

We will we will develop our software faster.

对。

Right.

明白吗？

Okay?

但对于那些想把AI和自动化结合起来的实践者来说，有一个非常有趣且反直觉的发展。

But there is a very interesting and unintuitive development for the practitioners that want to bring AI and automation together.

他们想要改造他们的流程。

So they want to transform their processes.

你看，我们都知道构建AI代理并不容易。

Look, we all know that building AI agents is not easy.

所以之前讨论的都是构建底层基础，并交付它。

So that all this whole discussion was about building the substrate, delivering that.

但现在，构建确定性自动化变得容易多了。

But what's becoming much easier is to build deterministic automations.

这就是游戏规则的根本改变。

That's fundamentally how the game has changed.

因为以前，当你把自动化集成到系统中时，瓶颈总是在系统末端的分支，这些分支很多而且非常复杂。

Because it was was when you when you build automation into a system, always the bottleneck is the ramification at the end of the systems that are very it's many and very complex.

在编码代理出现之前，情况并非如此。

Before coding agents, it wasn't.

RPA填补了一些空白，但它在某个点就停滞了，因为RPA也需要很多自动化工程师。

RPA kind of it it filled a gap, but it stopped at some point because RPA also required a lot of, you know, automation engineers.

你需要技术技能。

You need technical skills.

嗯。

Mhmm.

而且它们最终会有一定的产出。

And they can and they have a certain output in the end.

对。

Yeah.

它们扩展得这么远。

They scale so far.

但现在想象一下，如果我们从几个月的开发时间缩短到几天。

But now imagine if we if we change from, like, months of development to days.

所以我们正在改变可能性，因为现在有了AI。

So we're changing what is what's possible because AI is now there.

因为那些有效地在技术和人类之间建立基础层的客户，他们将能够以以前无法做到的方式进行流程级和案例级的编排，既有概率性的，也有确定性的。

Because because customers that effectively build a substrate across technology and humans, they will be empowered to do process level and case level orchestration in a way they could not before, probabilistic and deterministic.

我们还能够回到最初的技术，让它变得更好、更快、更完整。

We are also able to go back to the technology that started it all and make it better, faster, and more complete solution.

所以AI赋予客户使用我们平台的端到端能力实际上在两个方向上扩展：它的起源和它的新前沿。

So the end to end capabilities of what AI empowers customers to do with our platform is actually expands on both fronts, its origin and its and its new frontier.

你看，我觉得正在形成的模式跟人类的运作方式非常相似。

And look, it's I think the pattern that is emerging is very similar to how humans operate.

所以当我做一项工作时，你知道，如果是新颖的工作，我基本上处于探索阶段，去理解这项工作应该怎么做。

And so when when I do a piece of work, you know, if it's if it's novel work, I'm basically in an exploratory phase when I understand how the work should do.

我会用很多认知能力。

I use I use a lot of my cognitive power.

对吧？

Right?

同样的工作我做了几次之后，就变得常规了。

I do it a few times, the same piece of work, and then it's becoming routine.

一旦变得常规，我就不需要那么多认知能力了。

And when it's routine, I'm not using so much cognitive power.

我只是在自动驾驶。

I'm just doing on automatic pilot.

我现在看到类似的工作也在这样演变。

I see kind of the same work evolving right now.