我们独特的优势：UiPath作为编排层的技术依据

好的。

Okay.

我在做自动化。

I am doing automation.

我在做AI智能体。

I am doing AI agents.

我怎样才能确保智能体不会偏离轨道？

How do I make sure that the agent doesn't veer off a trajectory?

还有，我怎样才能确保有防护措施？

And how do I ensure that there are guardrails in place?

AI的潜力是真实的。

The promise of AI is real.

在组织中让它真正落地，同样是个挑战。

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

欢迎收听《前进之路》，在这里我们探讨AI时代工作的未来。

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

很高兴今天能请到Gautam Char和Raghu Malpani。

I'm pleased to be joined by Gautam Char and Raghu Malpani.

下午好。

Good afternoon.

下午好。

Good afternoon to you.

很高兴来到这里。

Nice to be here.

好的。

Alright.

我们今天要聊的是编排。

We're here to talk about orchestration.

那么，Raghu，谈谈现在AI发展到什么阶段了。

So, Raghu, talk about where we are in this moment right now with AI.

编排在其中扮演什么角色？

Where does orchestration fit into it?

对。

Yeah.

这个问题我一直在思考，你知道的。

It's a question that I think about all the time, as you know.

所以对我们UiPath来说，编排就像是一家公司拥有一个卡车车队，和拥有一家物流公司之间的区别。

So to us at UiPath, orchestration is the difference between, say, a company owning a fleet of trucks versus, owning a logistics company.

卡车负责干活，但物流公司决定卡车做什么、去哪里、怎么处理爆胎、司机生病或者路线上发洪水这些情况。

The trucks do the work, but the logistics company decides what the trucks do, where they go, how they handle, you know, a flat tire or a driver falling sick or a flood happening in a route and so on.

所以编排就是物流公司，而干活的是那些任务。

So orchestration is the logistics company, and the workhorses are the tasks.

所以在这个世界里，我们发现，无论AI代理是在哪个平台上构建的，可能是UiPath，也可能是客户选择的任何平台。

So in in this world, we find out that as you have AI agents built on any platform, could be UiPath, could be any platform of a customer's choice.

我们的确定性无令牌软件可以在任何软件公司构建。

Our deterministic tokenless software could be built in any software company.

它可以在UiPath构建，也可以在别的地方构建。

It could be built in UiPath or anywhere else.

重要的是要对由这些确定性的软件和这些代理组成的流程有一个全局的视角。

It becomes important to have a global view for a process that constitutes of these deterministic pieces of software with these agents.

这样你就可以压缩周期时间，拥有端到端的全局视图，并且具备可审计性和可追溯性。

So you can compress cycle times, have that global end to end view, and have auditability and traceability.

这就是我们现在看到的编排所扮演的角色。

So that's really the role that we find orchestration play here and now.

Gautam，同样的问题，但从你在Omega Healthcare的角度来看。

Gautam, same question, but from your perspective at Omega Healthcare.

跟我说说你在业务中看到编排在哪里。

Talk to me about where you see orchestration in your business.

也许我应该稍微回顾一下我们公司的历史和我们的重点。

Maybe I should rewind a little bit into our history of the company and our focus.

随着我们公司的发展，公司的第一步实际上是，如何通过自动化，比如从A点到B点，来降低我的成本结构？

Now as we've been growing the company, the first step of the company was really, how do I reduce my cost structure by automating, say, point a to point b?

所以我们最初走上了这条路，让我们把RPA放进去，那时候我们成为了UiPath的客户，我们可能把它用到了极致。

And so we went down this path initially of let's put RPA in there, and we became a UiPath customer in those days, and we probably used it to the hilt.

随着时间的推移，我们发现RPO花了我们很长时间，然后我们在那之上增加了AI代理能力，让我们以更智能的方式从A点到B点。

Over time, what we found was RPO took us so long, then we added the AI agent capability on top of that to take us again from point a to point b in a smarter manner.

现在，再往后看一点，公司发展得更大了，我们的客户现在期望我们真正成为我们代理、机器人、第三方解决方案、他们的第三方解决方案以及他们自己的AI解决方案之间的协调者。

Now, fast forward a little more, the company has grown even more, and our clients now expect us to be really an orchestrator between our agents, our bots, our third party solutions, their third party solutions, and their own AI solutions.

所以现在，我们业务的核心部分实际上是在各种技术和我们公司里的人员之间进行协调，以交付客户期望的突破性成果。

So now, a core part of our business is actually orchestrating between various technologies and the humans that we have in the company to deliver breakthrough results that clients expect from us.

所以我们在AI之旅上已经走了至少三年了，认真地说，一年或三年。

So we've been on this path for AI journey at least three years now, one or three years, seriously.

但真正认真地进行协调这件事，大概是在一年前开始的，我们一直深入参与那个过程。

But the fact that we orchestrate very seriously has ticked in maybe about a year ago, and we've been really deep down in that part of the process.

那太好了。

That's great.

所以我想谈谈流程的概念，特别是分解一个流程，了解当你考虑应用协调时，其中包含哪些不同的部分。

So I wanna talk about the idea of process, specifically breaking down a process and understanding what are the different pieces that go into that when you're thinking about applying orchestration.

是的。

Yeah.

我觉得从自动化领域的角度来看，这样看会很有帮助。

I think it's helpful to look at in the in the automation landscape.

先看看任务是什么，再看看编排层怎么把这些任务串联起来，这样会比较好理解。

It's helpful to look at maybe what the tasks are, and then how a orchestration layer glues together the tasks.

就拿我们UiPath来说吧，大概十年前，我们通过RPA找到了我们的产品市场。

So in our world, UiPath, we found our product market for, as you know, a decade ago or so with RPA.

RPA是确定性的软件。

RPA is deterministic software.

它能让你通过企业部署的各种应用，包括遗留应用和现代应用，实现企业级的覆盖。

It allows you that enterprise reach through the mixed estate of applications that enterprise deploy, legacy applications, modern applications.

所以任务层就变成了这个覆盖层，你用UI自动化和RPA来访问你的遗留应用。

So the task layer then becomes this reach layer where you use UI automation and RPA to get access to your legacy applications.

然后你用基于API的自动化来访问你更现代的软件。

And then you use API based automations to gain access to your more modern pieces of software.

MCP服务器，这是与代理交互的最新形式。

MCP servers, which are the newest incarnation of how you interact with agents.

所以，综合来看，你有AI代理，它们本质上是认知型的，还有确定性软件，也就是RPA和API工作流，这些共同构成了任务层。

So together you have AI agents, which are cognitive in nature, deterministic software, which is RPA and API workflows, forms that task layer.

而且你知道，在UiPath，我们一直都有人员参与的能力，让人类来监督、审批发票，或者根据自动化系统的建议采取纠正措施。

And as you know, at UiPath, we've always had humans, human in the loop capabilities to have humans supervise, approve invoices, or really take corrective actions based off of recommendations from an automation system.

现在，这些离散的任务，无论是确定性的还是代理型的，都分布在整个企业中。

Now you have these discrete tasks, deterministic, agentic tasks that are split over the entire enterprise.

这为局部孤立的自动化提供了视角，但你不知道如何压缩周期时间，因为需要人工交接，有人需要把任务从一个系统、一个流程转移到另一个。

It gives a view for localized isolated automations, but you don't know how to compress cycle times because there is a human handoff necessary where someone needs to move a task from one system, one process to the other.

流程编排让我们能够把企业中所有这些任务整合起来，形成一个统一的、全局的视图，你可以检查和询问流程的整体状态，了解优化机会在哪里，并推动这些效率提升，因为你有了全局视角。

Process orchestration allows us to glue together all of these tasks that exist in the enterprise into sort of one unison, one global view, where you can inspect and interrogate the overall state of a process, understand where optimization opportunities may be, and drive those efficiencies because you have that global perspective of how things are.

所以在我们看来，正如我们定义的编排，你有这些确定性的能力，还有无令牌的能力，与代理和人类一起协同工作，推动流程从左到右朝着目标前进。

So in our world, as we've defined orchestration, you have these deterministic capabilities, also tokenless, capabilities alongside agents and humans, together work in concert to drive a conclude to to drive a process along from left to right towards the goal.

Gautam，当你听到Raghu描述这些时，有没有一个流程或者你业务中的某个部分，让你立刻想到可以与之对应？

Gautam, when you hear Raghu describe that, is there a process or is there a part of your business that immediately springs to mind that you can map this against?

有趣的是，我们是从一个业务问题开始的。

It's interesting because we started off with a business problem.

我们的大多数客户都是市场上的医疗服务提供者。

Most of our clients are healthcare providers in the market.

他们的利润率非常低，只有2%到3%。

They run at a very low margin, 2%, 3% margin.

所以，作为我们给客户带来的商业价值，我们的目标是让他们的收入提升5%到10%，同时降低他们的成本结构。

So, as a business value prop we give our clients, our goal is to give them a five to 10% uplift in the revenue while we take out cost structure at the same time.

如果我们采用那个框架，看看我们的业务组合，比如90%的收入周期管理业务，我们首先把它分成前端、中端和后端。

So, if we take that framework, so we look at our book of business, say 90% of RCM, look at the RCM business and we said, Okay, let's first separate it into front, middle, and back.

我们为每一部分都设定了目标。

And we set goals for each one of those.

比如，我们设定了收入目标和成本削减目标。

So, for example, we set a revenue goal and a cost takeout goal.

这些是一些子目标。

These are some of the subsidiary goals.

然后我们召集团队和产品团队，说：好，现在确定在收入周期的后端中，哪些子流程我们可以用成本有效的方式去除，因为并非所有AI用例都是便宜的。

And we took the team and the product team and said, Okay, now identify which sub processes within, let's say, the back in the revenue cycle do you think we can take out in a cost effective manner because not all AI use cases are cheap.

所以我们就是这样为前台、中台和后台的用例设定目标的。

So that's how we came back with targets for use cases in the front, middle, and the back.

现在我们已经把这些目标按年度划分了，第一年做什么，第二年可能做什么，以及后续的计划。

And now we've segmented that into what we're going do in year one, what we might do in year two, and move forward.

这就是我们在创造影响力方面采取的方法，我觉得。

So that's the approach we've taken in terms of how we create impact, I would say.

收入周期管理是那种，我觉得对大多数人来说，听起来就像是最宏大的事情。

Revenue cycle management is one of those things that, I think to most people, sounds like the thing, like the biggest possible thing there could be.

从根本上说，编排对一个看起来如此庞大、如此复杂、涉及这么多不同参与者的流程，到底能做什么呢？

Fundamentally, what does orchestration do to a process like that that seems so big, that seems so complex, that seems to have so many different actors?

对。

Yeah.

对我们UiPath来说，我们——我先描述一下编排的几个类别，然后具体谈谈收入，比如RCM领域。

To us, at UiPath, we we I'll first describe a couple of categories of orchestration and then specifically talk about revenue, like the the the RCM space.

你看，我们认为编排大致可以分为两大类。

Look, think we think of orchestration in sort of two big buckets.

一种是，你知道，主要是定义好的流程，变化很少。

One is, you know, you're mostly a defined process with very few variations.

比如员工入职流程，或者信用评分高的贷款申请。

This is where think of an employee onboarding process, so think of a loan origination for people with high credit scores.

这些流程路径相对明确。

They take a reasonably well defined path.

所以你可以预先设定工作流的形状，确定哪些任务是确定性的，哪些是渐进的，人在哪里介入，异常可能出现在哪里，等等。

So you can you can pre establish, predefine what the shape of the workflow looks like, what the deterministic tasks are, what the asymptotic tasks are, where the humans come in, where the exceptions might be, and so on.

在UiPath，我们开发了一个叫Maestro BPMN的东西。

At UiPath, we've built this thing called, you know, Maestro BPMN.

这是BPMN表示法，用来构建流程，非常适合我刚才提到的那些案例。

This is a BPMN representation of how you build that process out, and it works very well for these examples of cases that I talked to you about.

还有第二类情况，编排也很重要，那就是每个流程实例都有不同的变化，因为流程本身非常不确定和复杂。

There's a second category of cases where orchestration becomes important, which is every instance of a process takes a different variation because the process is so undefined and complex.

收入生命周期管理就是一个例子。

Revenue life cycle management is an example of it.

拒付就是一个例子。

Denials is an example of it.

没有固定的路径。

There is no fixed path.

这些流程是混乱的。

These processes are chaotic.

它们非常动态。

They're very dynamic.

从产品角度来看，UiPath 采取的是基于案例的方法，你定义、理解并接受这些流程是动态、混乱的事实。

This is where, from a product perspective, UiPath, we take a case based approach where you define, you understand, and embrace the fact that these processes are dynamic, chaotic.

对吧？

Right?

所以你把这些问题分类成几个阶段。

So you you categorize these these these problems into stages.

我们有这个叫做案例管理器代理的东西。

We have this thing we call a case manager agent.

我接下来会讲得稍微技术一点，但我觉得这很重要。

I'm going to a little bit of a technical video, but I think it's important.

这个案例管理器代理能随时了解流程的状态，然后决定下一步的最佳行动是什么。

Is this case manager agent understands the state of the process at any given moment in time and then defines and determines what the next best action is.

对吧？

Right?

而且这个最佳行动，对于流程的每一个实例，可能都是不同的。

And that next best action can literally be a different action for every instance of the process.

对吧？

Right?

所以，把 Maestro BPMN 用于定义明确、结构化的流程，而 Maestro Case 用于更混乱、动态的流程，这样 UiPath 就能提供一个满足客户需求的编排平台。

So taken together, Maestro BPMN, for the more well defined and structured process, and Maestro Case, for the more chaotic dynamic processes, allows UiPath to offer an orchestration platform that meets customers where they are.

收入生命周期和生命周期管理，它绝对属于第二类，因为没有固定的路径，而且拒付代码的数量，你比我清楚得多，对吧？

Revenue life and life management comes comes it's it's categorically in the second category, where there isn't a fixed path, where the number of denial codes, you know a lot more than I do, right?

差异如此之大，以至于案例代理必须为每个流程实例采取非常确定的路线，而基于案例的方法对我们来说最适合这种流程。

It's so significantly different that the case agent has to take a very deterministic route specifically for every instance of the process, and the case based approach to us makes the most sense for that kind of process.

我该补充一下。

I should add to it.

我们做这行生意之所以这么复杂，对吧？

The reason we are in business is so complicated, right?

实际上，说正经的，保险公司和付款方的数量加起来，任何一个客户都有几千家，再加上病人接受的治疗种类、就医的场所，比如医院、门诊、理疗、手术中心等等，这些组合起来，还有雇主那边。

In reality, putting it more seriously, the combination of number of insurance companies, the payers, which are in the thousands in any one client, and the number of treatments patients go in for, number of settings you could go into hospital, outpatient, physical therapy, ASC, you know, the combination of that, plus the employers.

每家保险公司都和雇主单独谈判协议。

Each insurance company negotiates a separate deal with employers.

政府还有自己的医疗保险和医疗补助之类的。

And the government has their own Medicare, Medicaid, and things like that.

这些组合起来，可能的组合和排列太多了。

The combination is very too many combinations, permutations possible in this thing.

所以，真的很难有人能什么都懂，这就是为什么AI用例、编排等等能派上用场，因为那种知识你根本没法复制。

So, it's very hard for someone to actually know everything, which is where the use of AI use cases orchestration, etcetera, comes in handy because there's no way you can replicate that knowledge.

所以，我们就得靠你所在的公司来帮我们走完这个过程，你懂的。

And so, that's where we depend on a company that you are part to help us through this journey, you know.

你真的没法谈医疗保健。

You can't really talk about healthcare.

你真的没法谈收入周期管理。

You can't really talk about revenue cycle management.

不谈合规、数据安全和访问权限，这些都没法谈。

Can't really talk about any of these things without talking about things like compliance and data security and access.

所以我想谈谈治理。

So I wanna talk I wanna sort of talk about governance.

所以当我说治理这个词，以及人们把它和编排联系起来时，Raghu，你想到什么？

So when I say the word governance and people associated with orchestration, what comes to mind for you, Raghu?

你怎么看这个问题？

How do you think about this?

在现代编排的故事中，治理扮演什么角色？

What's the role of governance in the modern orchestration story?

对。

Yeah.

这个问题非常棘手。

It's a it's a very tricky question.

我希望我能用两分钟回答，但我会尽力。

I wish I wish there was a two minute answer to but I I will try.

你知道，说实话，我们接触的CIO们最常谈到的就是，好吧，我在做自动化。

You know, I think to be honest, the the single most important thing that CIOs who we speak to talk about is, okay, I am doing automation.

我在做AI代理。

I am doing AI agents.

不管是在UI平台还是其他地方，我如何确保代理不会偏离轨道？

Whether it's on the UI platform or elsewhere, How do I make sure that the agent doesn't veer off a trajectory?

我如何确保有防护措施等等？

And how do I ensure that there are guardrails in place and so on?

我如何确保代理只能访问被授权的特定信息子集？

How do I make sure that the agent can only access a certain subset of information that is authorized to access?

我如何确保代理在追求目标时，不会偏离方向，去做我不想让它做的事情，坦白说？

How do I make sure that as the agent is goal seeking, that it's not veering off and doing things that I don't want it to do, quite frankly?

所以作为平台层面的能力，我觉得我们要用业务用户能理解的方式来阐述治理的力量，这确实很关键。

So as a platform level capability, I think we think about articulating the power of governance in ways that our business users understand, which is gut wrench.

但这里要明确的是，一个智能体能做什么，以及严格来说，它不能做什么。

But here's the things that an agent can do, and strictly, here's the sets of things that an agent cannot do.

而且顺便说一句，这不仅仅是智能体的问题。

And it applies to by the way, it's not an agent thing.

它也适用于像机器人、API工作流这类东西。

It also applies to things like bots, things like API workflows.

它适用于所有任务。

It applies to all tasks.

对吧？

Right?

所以我们采取一种非常慎重且细致的方法，确保任务和流程能做的事情都经过仔细审查，并让企业能够决定和确定智能体在操作时有哪些访问控制。

So we take a very deliberate and pretty fine grained approach to making sure that the sets of things that tasks and processes can do is inspected of the fine toothed comb and give businesses the ability to determine and decide the sets of access controls that exist in what an agent can do.

所以治理其实只是一个简单的杠杆。

So governance is sort of just one simple lever.

另外，我们引导客户和用户去做的一件事是，在构建代理时，我们帮他们构建那些目标单一的代理。

The other thing we guide our customers and clients to do is as you are building agents, we help them build agents that are single-minded.

我的意思是，你可以构建一个非常复杂的代理，但这意味着它可能会偏离轨道，去做一些你不希望它做的事情。

What I mean by that is you can build a very complicated agent, which just means that it'll probably veer off and do things you don't want it to do.

如果你让代理更简单，就像桌面终端上的一个人类员工，如果只让他们做一件事，他们可能会比让他们同时处理五件不同的事情做得更好。

If you make an agent simpler, just like a human worker on a desktop terminal, if have them do just one thing, you know, they'd probably do a better job than asking them to multitask and do five different things.

对吧？

Right?

所以我们的平台帮助客户更好地管理这些单一目标的代理。

So our platform helps our customers better these single mounted agents.

我觉得这非常有帮助。

That is extremely helpful, I think.

在安全和治理方面，我们遵循所有的治理能力。

And in terms of just security and governance, we follow all the governance capabilities.

例如，我们符合HIPAA标准，对于那些在美国关心这一认证的客户。

We HIPAA compliant, for example, for those customers in The US that care for that certification.

SOC 合规性是我们首先想到的。

SOC compliance is something that comes to mind.

我们在云管理方面提供了强有力的治理。

We offer strong governance in how we manage our cloud.

我们有一支规模相当大的安全专家团队，他们的全职工作就是确保客户资产在我们云上的安全。

We have a pretty large group of security experts employed in our payroll whose full time job is to make sure that customer assets operating on our cloud is taken care of.

这些专家在安全和自动化领域都是顶尖的。

And these are some of the leading experts in security and automation.

所以对我们来说，治理不仅包括平台层面的能力，还包括可信赖的人员、对云运营的深入了解以及介于两者之间的一切。

So governance to us is, yes, of course, the platform level capabilities to also trusted personnel, a high degree of knowledge in terms of how we operate the cloud and everything in between.

对。

Yeah.

所以我想问你，Gautam，一个类似的问题。

So I want to ask you, Gautam, a similar question.

当你考虑你试图推动的转型的潜力与医疗行业对治理和安全的需求之间的平衡时，你会怎么想？

When you think about the balance between the potential of the transformation you're trying to effect and the need for, especially in healthcare, for governance and for security?

你如何平衡这两件事？

How do you balance those two things?

还有，有没有那种——我不想说灵丹妙药——但你会给那些同样在纠结如何平衡这两件事的人什么建议呢？

And is there, like, is there, like, a a I don't wanna say a silver bullet, but, what advice would you give someone maybe who finds themselves at that same place where they're trying to figure out how to balance those two things?

这很有意思，因为我们作为外包商，还有额外的复杂性，我还得让客户相信我们是安全的，而且你知道，医疗行业对PHI（受保护健康信息）有额外的担忧。

It's interesting because we've got a little extra nuance because we are an outsourcer, so I've got to also convince my customers that we are safe and healthcare, as you know, has a little extra concern in terms of PHI.

如果代理失控，它可能会做出奇怪的事情，对吧？

If an agent goes nuts, it can go do something strange, right?

所以我们有额外的考量。

So we got that extra lens.

所以，我们不仅要自我管理，还要向客户展示我们是如何管理的。

So while we have to govern for ourselves, we've got to share how we're governing the client.

所以，就我们的业务而言，还有两件事要补充。

So there are only two things just to add to us in our business.

第一，我们有一个AI委员会，定期审查以监控性能，检查哪个代理，检查哪个代理，等等。

One, we have an AI committee that reads regularly to monitor the performance, who checks on what agent, checks on what agent, etcetera.

所以我们公司里就有这么一层。

So we are that layer that we've got across the company.

这是一个AI治理委员会，专门处理这类细微问题。

It's an AI governance committee just to deal with that nuance of that problem.

第二件事是，在我们的业务中，由于政府合规问题的敏感性，或者因为可能错失收入的问题，比如说你去看医生，医生实际编码时写的是：你接受了这种治疗，这次算作常规门诊。

The second thing is in our business, because of sensitivity of compliance issues from the government or money left on the table issues, because let's say you're visiting a doctor and the doctor actually codes saying, You saw this treatment, this one came into a regular office visit.

如果我编码过高，支付方或政府可能会审计你，甚至以欺诈起诉你，或者标记你为欺诈。

Now, if I code it aggressively high, the payer or the government can audit you and sue you for fraud or whatever it is, or flag you for fraud.

反过来，如果我非常保守，不敢少报，你就会错失收入。

On the flip side, if I'm very conservative can't do less, you leave money on the table.

对。

Yeah.

所以，我们还有另一个与人类相关的治理细节。

So, we've got that other nuance of governance that is tied to humans.

所以，即使软件可以自动化并给出建议，我们也必须有人工来审计，看看它的表现如何。

So, even though the software may automate it and suggest themselves, we have to have humans sort of audit it to see how it's performing.

所以，即使软件可能认为它是自动化的，我们仍然需要有人参与其中。

So, we've got that human in the loop requirement even though the software may think it's automated.

所以，我们得处理这个细微差别。

So, we have that nuance to play with.

而且在很多情况下，我们必须与客户分享一些发现，因为在当今世界，每个人对AI都既兴奋又担忧。

And in many cases, we have to share some of the findings with the clients because in today's world, everyone is hyper excited and worried about AI.

我说的是客户群体。

I'm talking about client base.

因此，虽然有信任，但他们也倾向于去验证。

And therefore, there's trust, but they are prosperous to verify it.

所以，在这些方面我们多了一层治理。

So, we have a little extra level of governance on those lines.

对。

Yeah.

我认为关于编排的一个常见误解是，人们认为代理之间的交互就是编排。

I think one of the common misconceptions about orchestration is people assume agent to agent is orchestration.

我们显然有不同的看法。

We obviously have a very different view.

Raghu，你能谈谈那个不同的观点吗？为什么代理对代理的方法有限，而业务编排方法才是你真正需要走的路？

Can you talk about that, Raghu, talk about that different view and why an agent to agent approach is limiting, and why business orchestration approach is the real way you need to go?

是的，是的。

Yeah, yeah.

这是个好问题，也是我在每次客户会议和EBC中都会被问到的问题，我们经常讨论这个。

So it's a good question and one that we get asked by every client and every EBC that I'm in, have discussions on this.

说实话，我觉得这个问题的答案正在不断演变，就在我们说话的时候。

Look, I think the the answer to this is quite honestly evolving, as we speak.

你知道，一年前，当我们刚开始进入编排这个领域时，Daniel、我自己，还有我们一群人，我们构建了一个叫Maestro的产品。

You know, a year ago, when we first entered this, this world of orchestration, Daniel, myself, and a bunch of us, you know, we built this product called Maestro.

那时候，我们认为我们理解，甚至我们的客户也理解，你需要明白业务流程是复杂的。

And and at that time, we think we had the understanding, and our customers had the understanding even, that you needed to understand that business processes are complex.

有些事情是确定性的。

There are things that are deterministic.

有些事情是确定性的，有些事情是认知性的，而你不应该用认知性的方法去处理确定性的任务。

There are things that are cognitive, and you don't do cognitive things for things that are deterministic.

所以，如果你从CRM系统里查信息来确定付款金额，你就不应该用智能体来做，因为你希望这一步尽可能确定、可靠、稳定、快速、便宜。

So if you're retrieving information from a CRM system to determine what the, you know, what the the payment amount is, you don't want to use agents for it because you want that step to be as deterministic, as reliable, as dependable, as fast, as cheap as possible.

对吧？

Right?

然后，你真正应该用智能体的是那些更认知性的任务。

And then you want to really use agents for your more cognitive tasks.

比如从文档中提取信息、处理、评估贷款申请等等。

Know, you're extracting information from documents, processing, evaluating a loan application, and so on.

所以我觉得这种关注点分离非常非常关键，就是你有确定性的软件组件，也有认知性的软件组件，当然还有人来监督，确保无论用智能体还是确定性软件，都能采取最佳行动。

So I think that separation of concern is super, super critical, where you have deterministic pieces of software, cognitive pieces of software, of course, human supervising it, to to make sure that that that whenever agents are used, or even when deterministic pieces of software are used, the the best course of action happens.

所以，就像我跟你说的，这是一年前的情况，但现在的答案依然如此。

So as I mentioned to you, this this was a year ago, but the answer is still true.

我们发现，那些在业务转型目标上最成功的客户，正是像我描述的那样使用业务编排和自动化的。

It's very much the case that the customers that we find have the most success with their business transformation goals use business orchestration and automation in the way I describe it.

要有谦逊的态度，明白在合适的地方用确定性软件，只有在做认知性工作的时候才用智能体。

Have the humility to understand that you use deterministic software where it makes sense and use agentic only when you're doing cognitive stuff.

如果混着用，很可能你的流程目标和转型目标会遇到麻烦。

If you mix it up, it's very likely that your process goals, your transformation goals will will run into duress.

对。

Yeah.

Gautam，我想问你一个类似的问题。

Gautam, I wanna ask you a similar question.

当你决定在哪里应用编排，在哪里用智能体而不是自动化，就像Raghu描述的那样，你是怎么做出这个决定的？

When you're making the decision about where you're going to apply orchestration, where you're going to apply agents versus automation, the way that Raghu described, how do you get there?

比如说，你真的会坐下来在白板上画出来，或者用思维导图之类的吗？

Like like, do you literally sit down and map things out on a whiteboard or go through a mind map or something like that?

还是说这是AI委员会的角色？

Or is that part of the role of the AI committee?

对你们来说，这个决策过程是怎样的？

What's that decision making process like for you guys?

其实，我们的做法有点不一样。

Actually, quite it slightly differently.

首先，我们有一批顶尖的思考者，他们会回头问：我该如何重新构想整个收入周期？

Our first thing is we got a bunch of leading thinkers that go back and say, How do I reimagine the whole revenue cycle?

真的是从头开始。

Literally.

因为现在的运作方式不一定就是未来该走的路。

Because the way it's run today is not necessarily the right way to move forward.

所以，我们有一群人专门做这件事。

So, we've got a bunch of people that just do that.

然后，我们有大约三百五十个客户，对吧，放在这个背景下来看。

So, then we go back and we have like three fifty odd clients, just put it in context, right?

每个客户的运作方式都略有不同，但我们现在认为，如果我要重新构想整个流程，比如，让我先构建一个后端产品。

Everyone runs this slightly differently, but our now view is if I were to reimagine the whole thing, let me build a back end product, for example.

我直接把它放进去，然后围绕它做剩下的工作，用人力或其他软件来配合。

I'm just going to plop it in place, and then I'm going to do the remaining around it to stitch with humans or with other software around it.

所以我们退一步看看，我们要建的那个部分到底是什么？

So, we step back and look at, okay, what is that piece that we're going to build?

当你弄清楚那个部分怎么建的时候，我们再拆解Raghu刚才提到的那些部分，对吧？

And when you figure out what sort of built that piece, then we break down the pieces that Raghu was just talking about, right?

我们用代币、AI、自动化等等。

We use tokens, we use AI, we use automation, etc.

所以这就是我们的处理方式。

So, that's the way we are approaching it.

然后随着时间推移，我们放在RCM运营结构中间的这个部分，我们会不断扩展、深化、优化，等等。

And then as time goes on, this piece that we put in the middle of an RCM operations structure, we sort of widen it, deepen it, sharpen it, etc, etc.

所以最终，我们的目标是，三年左右，如果以前需要100个人做的工作，现在只需要70个人。

So, end of the day, our goal is in three years or so, if I needed 100 people to do a job, I need about 70 people.

所以我们正朝着这个方向努力，但只要有成本效益，我们就会系统地逐步推进。

So, we're driving towards that journey, but we're chipping away at it systematically as long as it's cost effective.

说得好。

That's a great point.

可以，当然可以。

Can I Yeah, absolutely?

当然，绝对可以。

For sure, absolutely.

对，对。

Yeah, yeah.

你看，我觉得我之前跟你提到过，那些在使用平台编排方面最成功的客户，是那些有谦逊态度，或者我们有谦逊态度去向他们解释如何拆分代理型和确定性软件的客户。

Look, I think I was talking to you about customers who find the most success with using orchestration on a platform are ones that have the humility or we have the humility to explain to them how you split agentic and deterministic software.

但我觉得你提出了一个非常好的观点，即使在你做那件事之前，让编排转型成功的第二个要素是真正理解为什么，真正理解怎么做，以及真正理解第一性原理。

But I think you make a very, very good point in terms of, even before you do that, the second ingredient of what makes orchestration transformation successful is really understand why, really understand how, and really understand first principles.

用例，你必须确保你正在追求的用例是真正基础性的，真正高回报的，并且真正用第一性原理解决的。

Use case, you must be sure that the use case that you're chasing after is one that is truly foundational, truly high ROI and truly first principles solved.

我认为这是一个非常重要的观点。

I think that's a very important point.

绝对同意。

Absolutely.

对。

Yeah.

我觉得这是个很好的总结方式。

That's a great way, I think, to sort of to to wrap up on.

所以我想问你们两位一个问题。

So I wanted to ask you guys both a question.

我想问Raghu一个问题，就是关于我们UiPath怎么看待这个的。

I think there's a question I'll ask you, Raghu, which is about sort of, like, how we, as UiPath, think about this.

当你想到下一个编排时刻或下一个编排突破时，考虑到各个行业，你觉得下一个时刻会是什么？

When you think about the next orchestration moment or the next orchestration breakthrough, As you think about, like, the industries, what's the next moment that you think is coming?

我们已经看到的一个是构建一个基础平台的想法，这让我们能够真正构建所有用于映射复杂业务流程的工件，并使用编码代理来打印那个流程。

The one that I all we already see is this idea of building a substrate, which is how we really allow the the construction of all the artifacts that go into the mapping of a complex business process and using coding agents to print that process.

我说的打印，真的是指把代码写进去，然后让业务用户来修改。

When I say print, I literally mean literally write the code out into it and have the business user modified.

以前需要几个月的事情。

Things that will take months before.

我们想象一个世界，希望不会太遥远，将复杂的流程转化为现实，转化为优化的目标追求函数，只需要几天、几周，而不是几个月和几年。

We imagine a world, which is hopefully not too far out, converting complex processes into reality, into optimized goal seeking functions will take days, weeks, not months, and years.

这就是我们最近宣布的，UiPath 在编码代理方面正在推进的转变。

So that's the transition that we're rolling on with UiPath for coding agents, as we've announced recently.

这个概念就是打印最复杂的业务流程。

This concept of printing the most complex of business processes.

还有一件事我们非常关注，也是许多客户非常关注的，就是真正成为他们的思想伙伴，帮助他们识别所在垂直行业或子垂直行业中最高价值的用例。

There's another thing that is very actively top of our mind and top of the mind of many of our clients is really being their thought partners in helping identifying the highest value use cases for the vertical or the sub vertical that they're in.

我们作为一家公司，正在积极思考，为最高层管理者打造这样一个代理式的发现伙伴、思想伙伴。

And we are actively thinking as a company to build that agentic spotting partner, thought partner for the most senior executives.

对。

Yeah.

为企业中最资深的自动化专家提供帮助，让他们驾驭企业中存在的复杂性，并进一步阐述最高价值、最优化的用例。

The most senior entrenched automation specialists in an enterprise to help them navigate the complexity that exists in the enterprise and help articulate further the highest value and most optimal use cases.

这是我们正在进行的非常早期的探索工作，我们有一些客户正在与我们合作，共同构思这个想法。

This is very early discovery work that we're doing, and we have a few clients that we're working in in sort of ideating it.

对。

Yeah.

那你呢，Gautam？

What about you, Gautam?

你看到的下一步是什么？

What's the next moment that you see?

好消息是，我们只有70%的软件开发是自己做的，这很好，因为我们的观点是，我自己做70%，然后我会寻找周围有创新能力的公司，把它们纳入我们的技术生态系统，来交付产品。

The good news is we only do 70% of our software development ourselves, which is good because our view is that I'm going rise 70% myself and I'm going to find innovative companies around me that I want to incorporate into our tech ecosystem to deliver stuff.

目前，我脑子里没有一个能从根本上改变我们世界的灵丹妙药。

Now, I don't see a silver bullet in my mind that's going to change my world fundamentally.

我看到很多人们提出的点解决方案，我可以更快地整合它们。

I see a lot of these point solutions that people produce that I can incorporate much quicker.

所以我们正沿着这条路走，我们一直在观察行业格局。

And so, we're going down this path, we keep looking at the landscape.

我们也在努力自己创新，寻找切入点，但现在这些都是微创新，你知道，它们不会彻底改变我们的世界。

We're trying to innovate ourselves, find points, but they're all micro innovations now, you know, so they're not anything that will radically change my world.

它会加速我的世界。

It will accelerate my world.

所以我们走的路线是，要么我自动化处理，加一些检查，因为在我们这行，收入或合规是个问题；要么我提供人工辅助，因为这太复杂了。

And so, we're going down the path of either I'm going to do it automated with some checks because in our business, revenue or compliance is an issue, or I'm going to provide a human assist because this is so complicated.

我不需要那种做了二十五年同一个支付方、一直做BCBS的人来干活，我可以随便找个只上过一年大学的人来做这个工作。

I don't need someone twenty five years doing the same payer, BCBS forever to come and do the work and get someone literally off the street with one year college degree to sort of come and do this work for us.

所以，可能还是辅助，这样知识门槛就降低了，任何人都能做，对吧？

So, probably assist so that the knowledge is brought down to the level that anyone can perform, right?

这就是我们前进的方向。

So, that's the direction we're headed to.

希望有一天，病人来看医生，看完医生马上就能拿到钱。

And hopefully, one day, we can get a patient come to see the doc, and the moment that's done, the doc gets paid.

但我觉得那还很遥远。

But that's I think that's quite far away.

这不是一朝一夕能实现的。

It's not a grind that's going to happen.

是的。

Yeah.

不过我看到了一路上的渐进式改进，我们就在事情进行中把它们纳入进来。

And so but I see the incremental improvements along the way, and we just tuck them in as things are going on.

那太好了。

That's great.

非常感谢你们两位的见解，谢谢你们能来。

I really appreciate both of your perspectives Thank on you so much for being here.

谢谢你们精彩的回答。

Thanks for the great answers.

谢谢你们带我们了解了这个编排之旅，无论是在UiPath这里，还是在你们每天开展业务的现实世界中。

Thanks for walking us through what this orchestration journey looks like, both here at UiPath, but also out in the real world where you guys are doing business every day.

谢谢。

Thank you.

谢谢。

Thank you.

感谢邀请我们。

Having us.

非常感谢。

Appreciate it.

感谢收听关于未来道路的分享。

Thanks for listening to the path forward.

我们路上见。

We'll see you on the journey ahead.