I would first like to go over a few housekeeping items. Today's event is being recorded and will be posted to our investor relations website following the event. I would also like to point you to our Safe Harbor statement and remind you that today's discussion may contain forward-looking statements. Actual results may differ materially from these statements as a result of various risk factors, including those found in our SEC filings. We may disclose information related to development and plans for future products, features, or enhancements, which are subject to change at our discretion without notice. All statements are made only as of today, and UiPath undertakes no obligation to update any forward-looking statements and makes no assurances and obligation to update any forward-looking assumes no responsibility to introduce future products, features, or enhancements described today.
Additionally, we would like to note that this is a product webinar, and as such, we will not be taking any financial-related questions. In terms of today's agenda, we begin with a fireside chat with Daniel Dines, our Founder and Chief Executive Officer, which will be moderated by Hitesh Ramani, UiPath's Chief Accounting Officer and Deputy CFO. Following this, Graham Sheldon, UiPath's Chief Product Officer, will provide a product demonstration. After the product demonstration, if time permits, we will take questions from the audience. Please use the Q&A feature to submit your questions, or if you're on a mobile device, please submit through the chat function. With that, I would like to turn it over to Hitesh.
Thanks, Jake. Hey, Daniel, how are you?
I'm great, Hitesh, and hello, everyone. Thank you so much for joining us.
Daniel, where are you dialing from today, by the way?
I am dialing exactly from the place where we have started our journey, where we have built our first robots in Bucharest, Romania.
That's great. That's great. Let's get started. Daniel, we pivoted from an Act One RPA company to Act Two agentic automation company. A lot has happened. Maybe it will be helpful if you can walk us through our journey towards agentic automation and why we believe it was such a natural next step for UiPath.
Since the beginning of our company, our vision was to play into end-to-end process automation. RPA was, in all fairness, just a really good gate into enterprises. It's very difficult to go and say, "I'm going to play directly into API automation, system integrations." RPA provided us an incredible, easy gate to build great relationships with customers and be taken seriously. Ten years ago, in this city, we were living, you know, ten people in an apartment. It was an incredible chance. We were not limited to RPA throughout existence. We started with RPA, but then we extended into APIs because API automation is one of the cornerstones of process automation. It cannot exist. One of our strengths is that we combine RPA, which means using the user interface of applications like humans do, with a very powerful, world-class API automation story.
Moreover, we moved into intelligent document processing because many, many processes start with a document, semi-structured documents. We later moved into emails, which is a form mostly of short messages. We have built AI to understand short messages. Now, as you know, we have moved into understanding long, complex documents. Agentic is a natural extension because we call it second act because all the other previous technologies, in fact, were rule-based, even if they were AI-based. The process could have been described in rules. That was inherently limiting, limiting the entire capability of our platform. Now, when LLMs arrived on the, you know, on the world stage, we see an incredible opportunity to go after cognitive-like processes. I want to make a really important point. Our technology was based always to imitate how people work.
What AI is doing when you apply in the context of processes aims also to imitate how people work. The way AI is going to make a decision, even if it's, I don't know, a loan application decision, the AI is going to emulate a human's mind. I think it's a very important, but another big difference between how LLMs work and how our traditional automation works is that automation is reliable and deterministic. AI is non-deterministic. That was the biggest challenge that we actually had. How can we take, you know, the power of GenAI and deploy it into a deterministic fashion? Here we enter agentic automation. We've built all of the frameworks in order to take the non-deterministic technology and apply it into deterministic processes. Again, this is a natural extension for us that will enhance our robotic capabilities.
That's very impressive. In fact, I see when we say that we wanted to always emulate human behavior. I recall we were defining it as a digital FTE, and the natural extension to that is now agents. That's very impressive. Now that we know why it is a natural extension to what we have been doing, what, in your view, is our right to win in the market, Daniel? Also, what are some of the use cases? How are we helping our customers solve some of their complex problems through this technology?
First of all, we are already there in the context of enterprise processes. I think many, many companies would love to have our expertise in understanding business processes, manual business processes, because in the end, if you where is going to be the most applicability of agentic AI? Take an existing manual business process, emulate people, and get all the benefits of AI, right? We are already in the context of we have 10,000 customers. Our robots work in the context of the business processes. We have the understanding. Now, for us, it's a matter of extending the robotic capability. So it's as simple as going to where are the processes where you have robots? Let's look left and right and see what kind of tasks couldn't be automated before. Let's bring agents that can help people.
Actually, they will reduce the human input on these tasks and make these agents work with the existing robots. It's a natural extension. If you think if you have already robots installed and you build agents, it's very handy to manage agents and robots within the same platform. Because robotics, it's also an emulation of human technology, we had to build a lot of security and governance that are specific for this type of technology. Nobody really has it at our level, this level of governance that is specific to emulation software. We've been from the beginning, the way a robot accesses an application, how they handle people's passwords. You need to give robots passwords to access different applications. This is unusual for workflow companies. They don't have to deal with this type of scenarios, but we had to.
We are applying the same set of governance and security rules to agents. It's much more easier to handle and manage everything in one platform. It's like, think about in your HR, you don't have two HR systems, one for, let's say, contractors, one for your full-time employees. You manage them on the same platform. It's normal. You, from the, you know, giving rights to your documents in the company, from everything, it's one platform. This is kind of the same with the robots. It's a natural right to be really there. Agents without actions are nothing. Most of the agents that we are seeing today are actually conversational agents. It's chat in, chat out type of interfaces. They don't run autonomously in the context of a business process. They are originated by human users that work on specific tasks.
The agents that we want to deploy and we are deploying are called and are instantiated by enterprise workflows. They run autonomously in, you know, in data centers, in the cloud. You have to put a lot more rules and governance around that technology in order to make sure that this is a reliable technology. Again, we have the means to do it. Like people, when an agent naturally touches multiple systems, if you look into an enterprise context, is it rare when a business user lives only inside one business application? Is it rare you live only in Salesforce or in SAP? You connect to multiple business systems in order to carry on even most of the tasks. That means that the actions that agents need in order to complete their goal touch multiple systems. This is where we shine.
We are the Switzerland of integrations with different systems. We are agnostic. We are not going to provide better support for Salesforce or better support for SAP or better support for Dynamics or Oracle or whatever. We will be agnostic, and we will provide equally great integration support to all platforms. It is natural. I heard it from CIOs of big healthcare companies, financial companies saying, "I'm not going to put my data from one system into the other in order to power the agents." I like your approach that is agnostic, and I like your capabilities to integrate all systems. It is, in the end, why RPA exists, because there are systems where it is very difficult to communicate with because maybe they do not have APIs, or maybe they are very complex APIs, very difficult to implement that require, you know, high-level expertise, where there are legacy systems.
We have mainframes in banking systems, financial systems. We routinely automate tons of mainframe applications. I don't think they will disappear. They stay for 30 years. I don't think they will disappear in the next year. We are really best in class when we connect to these systems. In many enterprises, agents will connect to multiple systems, some legacy systems, some modern API systems. It makes sense to have a platform that handles everything. Why would you take a platform that handles APIs, and then you have a different platform when you'll need RPA? It makes no sense. It creates a lot of security imbalances. It's much more difficult to manage the security. It's a natural way for us to expand. You asked me also about use cases.
We're seeing really use cases across the board and within always into the finance department, order to cash, procure to pay, huge processes. Now, customers have the chance to get an amazing level of automation. We are using with one of our largest customers in Japan, a large bank in Japan. They want to achieve in their order to cash process 95% accuracy. It's a huge number, and they started with automation around under 50. So it's a great opportunity. We are seeing in healthcare, in the context of revenue cycle management. We are working with many customers, client claims denials, prior authorizations. There are a lot of use cases and many more. I think, Hitesh, you are in charge with our own internal automation. Maybe you can elaborate a bit what we are doing internally.
Yeah, no, absolutely. Daniel, as you mentioned, we started with first transforming our finance processes using automations as part of our Act One. There were several areas where we could not actually solve some of our pain points, which we are now actually using agentic solutions to help us. I'll give you an example. Within our order to cash process as a company itself, there were still manual reviews of contracts that the revenue recognition team was performing. Now we are using our IXP technology along with agentic so we can actually transform how we are reviewing these contracts by putting human in the loop. In my mind, there are two things that actually help us really lay a strong foundation for autonomous workflows.
One is the fact that the solutions which we could not solve in the past, now with agentic, we are able to solve it. Using agentic along with robots and Orchestrator, we are actually able to embark on our autonomous journey. I feel like we are at a very great spot in terms of the evolution of technology. Maybe with that, Daniel, it will be helpful, I guess, for the folks on the call, if you can explain how important Orchestrator is for agentic workflows.
From our perspective, this is the essential component in order to deploy enterprise agents. Let me explain why. We thought a lot about it. Our foray into agentic AI started actually a year ago. We thought, "How can we deploy these agents?" We realized, first of all, our blueprint is agents will not be allowed by our enterprise customers to take directly actions that can provide important side effects and cause security issues or cause, I do not know, people will not let agents move money just by agents or maybe in very small quantities. We realized we need a framework to have humans in the loop, a very powerful framework. We need a framework that will allow agents to connect with agents and agents to connect with robots. We need a great, or we call it, an orchestration engine.
RPA had an orchestration engine, but it was not adapted to the modern, to the way that the modern way to describe processes and to facilitate this integration. We created from scratch a new orchestration engine. I think we really were ahead of the market. A year ago, nobody was even talking about agentic AI, and nobody was talking about orchestration. Now everybody is talking about orchestration because everybody realized that you need orchestration in order to deliver agents. This orchestration emulates also how human organizes in groups. If you think there is a reason why we create processes, why we create workflows that are governed by rules and by fixed path. It's mostly the individual tasks that one person is doing that require knowledge and, you know, cognitive approach. The end-to-end workflow, it's more of a fixed path.
You need to have this engine, and you need to have an engine that is modern and is capable of connecting agent to agents or agents to robots and putting humans in the loop. We invested a lot in this technology. It's not, it's an orchestration, means orchestrating tons of workflows. A process, you can think of a business process, end-to-end business process, is comprising hundreds of sub-workflows. We have great analytics that makes like a 360 view of all your process instances that comprise agents and robots and automations, APIs, documents. Everything is fully auditable. You see one single audit trail for everything. Even our process mining technology can connect to backend systems, and we can integrate these data sources. You can completely have a 360 view of the transactions in a system. Again, this resonates with our customers.
I'm explaining to customers, "Guys, this is how we are seeing. What do you think? Do you think we are wrong? Do you really believe? Are you ready to deploy a swarm of agents that you don't use? It's not clear how they talk to each other and somehow magically they will deliver results." It's not. People want this reliable way to deliver agents. This is what gives comfort to enterprise customers. You put humans in the loop in orchestration. You watch how the agents perform. People approve. Agents become actually better and better every time they see they learn from how people interact with them. Once you become more comfortable, you surround them with more rules, but give them more agency.
Like I'm comfortable that in case of a loan originating for loans less than, I don't know, X amount of money that can, you know, with people with this credit score and that different characteristics, I can bypass humans. Yes, but only after I need to have the orchestration, first of all. I need to have the audit. I need to see. I need to have the confidence. This is why this is tremendously important. This is why we are winning deals today, because our orchestration is really thought for the agentic era.
Yeah. No, that's helpful, Daniel. Now that, Daniel, we have GA-ed our agentic products almost three months. I also am aware that we had several of our customers which had taken part in a private preview. Since then, they have been actively working on use cases. What has our early feedback that you are hearing back from our customers, the early feedback on the products that they've started using? Also, where do you see them evolve, like go from here on forward?
Even before we launched our product in GA, we had customers asking us to GA because they wanted to put in production some of the agents and orchestrations. For a few of our customers, we had the controlled GA. We gave them the assurance that we are going to support the product even before being GA. They put it in production. It is the highest number of POC and pilots since our early days of RPA. Overall, it is a higher number. We are a much bigger company. This is our bottleneck of really fulfilling the POCs and pilots requests from the customers. The interest of this technology is tremendous. It is still, I want to make clear, it is still early days. Customers need to get confidence. Trust is a very big word here. Many customers still start small.
They put, you know, agents into a small part of the process, into a small context in order to get the confidence. We and our partners and customers are in the ways of figuring out what are the blueprints of large-scale deployments. My estimation is that throughout this year, this trend will continue. We will learn a lot more. We are building more and more vertical solutions. Agentic is very suitable to build more vertical solutions. Because in a way, I want to describe to you a bit how I see an agent. It's more like a college grad that goes to their first job. They know really the public space. They have a lot of knowledge, but they don't know their industry. They don't know any company specific. If I go into banking, I need to learn a lot about banking specific.
If I'm becoming an investment banker, I need to learn about. This is an industry specific. If I go to a specific bank, I need to understand the specific city of that bank. As we work with our customers, we capture a lot of industry knowledge. We already have great industry knowledge from our RPA. Now we are working and we capture more industry knowledge. We come with better industry solutions that we can replicate to different customers. We are in the process right now to understanding, again, what is the blueprint of large-scale deployments.
That's very helpful. Daniel, thank you so much for your insights. This was very valuable. With that, let me turn over the call to Graham, our Chief Product Officer, who will provide us a product demonstration.
Great. Thank you. Good morning or good afternoon, everyone. Thank you for joining us today. I'm Graham Sheldon, the Chief Product Officer at UiPath. I'm really excited to walk you through how our customers are actually able to unlock AI transformation and drive real ROI and time to value through our platform. I'm going to start this morning by giving you a view of our platform and how we've reimagined it in this new act that Daniel described around agentic automation. There we go. The UiPath platform that we're building is built on our strong foundation in automation. Our customers can now easily combine the best of what robots have been able to do about deterministic rules-based kinds of work with what agents and LLMs are now capable of doing, the more dynamic and goal-seeking type of work.
It is the combination of that in an end-to-end process that can really help transform the way that we do business and the way that work gets done. People are still at the heart of this, and they can really then focus on the critical decisions and the high-value work. That together runs seamlessly across all of the workflows, all of the systems, all of the people in an organization in an enterprise-grade governance framework so that you can feel confident and have the trust that those processes are running the way that you want to, and you remain in control, fully integrated into your systems, regardless of whether they be legacy ones or newer ones as we move forward together. Why is this platform uniquely able to deliver these AI transformation results for our customers?
First and foremost, we believe in what Daniel described for these mission-critical workflows, you really want to have specialized agents, not ones that are swarms of agents that are going to go run around and do things that you can't understand or control. Our customers demand that they have real insight into what those agents are doing, specialized for particular tasks so that you can trust that they're working in a way orchestrated across the workflows with what people need to do and what robots will need to do so that you get the highest level of accuracy, reliability, and governance. Those agents are only as good, though, as the actions that they take. Those actions are based on the best-in-class automations and tools that they have.
Again, over any system or any type of data, we've built into the platform those capabilities, and we believe that they remain best-in-class. That is work that is done that is orchestrated with people and with robots to be able to operate across an entire workflow through end-to-end agentic orchestration, which I'll show you in a few minutes. The UiPath platform supports Procode, so professional developer tools, and those can come from any vendor, including our own, so that you never get locked in and you can use the best-in-class models, best-in-class agents from any provider. We also have a low-code approach so that it can scale out and you can have your business users and domain experts participate in the development of those agents.
Underpinning all of that, we're investing a ton into making sure that you can track what those agents are doing, govern them, and manage them, and they achieve the highest levels of accuracy and reliability. Now, we've developed a ton of new innovation on the platform, but I want to focus on a few highlights that take these promises and make them a reality for our customers. First is a product we call UiPath Maestro. This is that concept that Daniel described in terms of agentic orchestration. With Maestro, we're not just simplifying the automation itself, but actually how teams manage it.
It's actually a collaborative surface where you can design, you can manage, and then you can optimize these workflows and plug in agents for the right tool for the right job, plug in robots for the right places, and allow people to participate as well, orchestrating all of that seamlessly in our system. Next is Agent Builder. Agent Builder is a low-code interface for designing these specialized agents and has an integrated experience to help you test and evaluate that over time so that you get the highest level of accuracy and reliability from those agents. Last but not least, we know there's a ton of data that's still locked away in unstructured or semi-structured data, like documents and communications. IXP is a product that now allows you to sort of extract and make that data really useful in these end-to-end automations.
All of these capabilities come together. I'm going to show that to you in just a moment on the platform. This platform is trusted by over 10,000 organizations worldwide. We have customers like WEX that are using these agents to tailor customer profiles and improve sales engagement. You've seen lots and tons of examples of this across many different industries, many different departments to truly transform the way people work and get incredible results. Now, I'm going to focus in on one example of that in the context of insurance claims processing. In this example today, people are doing a lot of manual work still. You're gathering critical information. You're doing some research. You're updating systems of record and systems of engagement, communicating back and forth with the claimant. It's a really difficult process. It's very messy. There's a lot of swivel chairing happening.
Think about that daily reality for an adjuster with multiple legacy systems simultaneously open. It's really hard to stay on top of this. It's also fairly error-prone. Those mistakes cost time, but they also cost patient outcomes. That's what we want to make sure that we can really transform. Now, imagine a transformed experience where instead of that adjuster doing all of this busy work, instead of that, we agents and robots take care of all of that sort of repetitive paperwork. They can focus really on just the part where the final decision needs to be made, where the agent has already made a recommendation on the basis of all of the data that it's extracted, all of the policies that it's been applied.
They can just do the last mile, the critical decision about whether or not to move forward and any additional information that can help this particular transaction go as quickly and as seamlessly as possible with natural language tools to help them get their job done even more efficiently. How do we make an experience like this one a reality? What I'm going to do is now show you a demo. I want to start with the process. The process, in a very simplified way, again, involves a lot of manual work. With RPA and automation today, robots are already doing a lot of this work, getting the critical information, making some simple rules-based identification of the claimant, and then updating the systems of record. Now with UiPath agentic automation innovation, we can now take out some of the rest of this.
You can take out a lot of these sort of more nuanced and dynamic decisions and make recommendations to people, like doing the initial duplicate claim check or checking for fraud, as well as a determination of eligibility. You want to orchestrate this process end to end. This is obviously an oversimplified view of this process. The process actually is quite a lot more complicated than that. A lot of the time to value is really unlocked when you can describe this in a way that lets your developers and your business users collaborate together to make this a reality. Now I'm going to switch and show you what that looks like in the UiPath platform. Imagine this particular process is now something you can model directly with UiPath Maestro. I'm going to switch over and show you what that looks like in our platform.
Here you go. This is that same HSA process working end to end in the UiPath platform. As I mentioned, you have some robot tasks where you're gathering critical information. You're updating systems of record. You also now have agents doing eligibility checks. We even have LangChain agents here built by professional developers doing a fraud check, as well as ones from other systems and other agent providers, like this one from Agentforce that might interact and notify the claimant, all orchestrated together. All of the actions that those agents are taking, all of the things that the humans are doing only when they're needed to, are tracked, which is super important to make sure that this traceability is happening.
In this case, if we've got an eligibility determination and we want to see exactly why the agent made the decision that it did, which in this case is to recommend that we approve the claim, I can go back and I can see all of the different information that it gathered, all of the different reasoning that it did tracked here in an auditable and traceable way. Now, many of these processes, as I mentioned, have a lot of documents that you're gathering from the claimant. Maestro is helping me to weave these together. Let me show you how these documents are processed in our system. In this insurance scenario, think of all that unstructured data that was here. Let me just hide my meeting controls for a second. There we go.
In IXP, all of those documents, the PDFs that contain the claim summary, the explanation of benefits, and each of the individual claim details are being captured by LLMs. With simple prompts, we're able to get all of that information from structured documents, even unstructured ones like the receipt here. I can actually go into IXP and, much, much quicker than you were ever able to do before, you can see visually where it's extracting some of that data from and be able to check and see how well this model is performing over time. Now that we've seen how effectively IXP can extract some of that critical data from complex documents and communications, let's go and check how the insights seamlessly feed into the broader automation workflow. Going back to Maestro, I'm now going to focus on this agent and how we built it.
This agent was built in Studio as a low-code agent for determining eligibility. You can see that we've got some critical elements. We've got the prompt or the steps that the agent should take in order to figure out whether or not we should move forward. It's got the tools, existing automations, web search tools. We support things like MCP for tools and agent-to-agent communication. You can reuse all of your existing automations, or you can use even other agents as part of the definition for this agent, as well as the critical context from your policies and how to escalate to humans if the agent gets stuck. I mentioned before, we're really focusing on reliability and accuracy.
In line here, it's kind of hard sometimes for people to understand how well their agent is doing and the improvements that they should make. I didn't come up with this prompt myself. I got help from Autopilot, actually. I asked it how to make it more concise. I can continuously look at the suggestions that are here in this particular case, maybe to make this a little more concise. This loop helps me create the best possible agent. How do I prove that? I need to be able to evaluate how this agent is performing both right now and then over time to make sure it continues to meet my quality expectations. I can look at results from many different evaluation sets.
For every one of these, I can go deeper and say, "Okay, in these particular cases, why did the agent make the decisions that it did?" and be able to compare with these test cases what I expected versus what the agent actually did, and then be able to update it and make recommendations back to the agent so that it can perform better next time. It is this integrated experience that we think is really unique to creating specialized agents in this enterprise context that you can trust. Now, excuse me. The last piece I want to show you is that we have just designed this great process. You can see that it is sort of paused at a step where I am going to make a final determination about whether or not to proceed with this claim.
In the UiPath system, we have an integrated way to build an action app. I can bring to the end user the final determination that they can make a decision on. As a business end user, I can take a look at the analysis that the agent provided with the summary. I can inspect the documents myself to make sure that indeed it's making the right decisions. I can provide additional comments so that over time, the agent can get better and better with long-term memory. Last but not least, we know that this process is an evolving process. We want to control the degree of agency and over time improve it.
With the UiPath platform, you can also look at how this process is going over time and be able to show where there are bottlenecks, where there's conformance problems, and even go so far as to try to figure out how to optimize it, to add additional agents, to see where conformance is not happening, and to be able to simulate and rework this to eliminate those bottlenecks, to improve this, and see where I have additional ways to improve this process. To summarize, Maestro is more than just modeling. It's a central command center where agents, where robots, where people are collaborating together in a very highly governed, highly accurate, and trusted way so that for our customers, processes that now took weeks and months take days. You're eliminating all of that busy work. You're focusing people on the stuff that really matters.
For UiPath, it's about our core differentiator, bringing the best of what robots can do for deterministic work with the best of what agents can do and help your employees focus on the stuff that really matters to be more productive, more confident, and to satisfy your customers and hopefully create better health outcomes in this particular case. That is the future of agentic automation with just a snapshot of what is possible today with UiPath on the agentic automation platform. Hitesh, I'm going to turn things back over to you.
Great. Thanks. Thanks, Graham. This is really cool and very exciting, Graham, to see how automation agents are operating along with humans on the side. It's all getting agentic orchestrated with UiPath Maestro. Let's take the next 10 minutes to address some of the questions. We are seeing a few questions come through the Q&A here. If folks on the call have more questions, feel free to ask them. Let's get started with the first question here for Daniel. The question is, do you need to see broad enterprise adoption before seeing agentic orchestration adoption? Why or why not? It sounds like I think folks want to understand, is there a sequencing? Does enterprises need to first embark on agentic adoption and then focus on orchestration, or can that be done side by side?
I think that for the context of autonomous agents that are deployed in the context of enterprise processes that run largely autonomously in the backend, back office processes, you need simultaneously orchestration and agents. We believe there is no other way to deploy them. This is actually something that I said before resonates deeply with our customers.
Great. Now, we saw in Graham's demonstration, Graham, you mentioned about there was a LangChain agent. And Daniel, you mentioned about how we are Switzerland. What do we mean by this and how UiPath Maestro is expected to interact with other agents? Maybe Daniel, if you want to answer that question.
Yeah. So we thought a lot where the world is going in terms of building an agent. And while we have tremendous experience in building low-code applications, low-code RPA, we know that also for sophisticated scenarios, people rely on code. We decided to have a dual approach on building agents, to have our own low-code Agent Builder that really can speed up the development, but also to support most common agentic frameworks, open-source frameworks like LangChain and LlamaIndex and CrewAI and there will be others. Because in the end, we want to offer we always said we are an open platform. We want to offer our customers the ability to build agents in whatever flavor they want. Even with these open-source frameworks, we offer the same level of integration with Maestro, but also with our robots as actions. You can create an agent in LangChain.
You can easily describe robots as tools for the agent. You can pack the agent. You can send it to our Orchestrator. You can see it as an agent in Maestro exactly like you see our own agent. Same security and governance apply to it. Moreover, when building an agent, we realized that one of the most complicated areas is testing the agent. We also have tremendous experience as a testing platform. We focused a lot on building evaluation sets for agents and helping our customers improve the prompts of agents. This set of technologies apply to our agents, but apply also to agents developed with the open-source frameworks. We treat them literally as first-class citizens in our platform.
Awesome. I see there is another question. Given the current rate of change or evolution in technology, why wouldn't the agents eventually be both deterministic and probabilistic and completely replace any need for deterministic robots?
Yeah, I think this is a very important question. It's because you can say humans, yes, can do both non-cognitive and also rule-based tasks, I would say. Why agents cannot do this? I think it's related to the existing limitation of technology. GenAI is extremely good at finding patterns in data. You can scan tons; it sees a lot of data. GenAI doesn't follow rules in a traditional human sense. It's very difficult to give a set of steps and rules and have GenAI follow these rules precisely every time. It's not going to deviate from the rules. I've seen even recently, I've seen an interesting study published by researchers of Apple where they try to convince GenAI to follow a simple algorithm for this problem of towers of Hanoi that is moving disks from poles to poles in a certain order.
GenAI, even if they said, "I can give you exactly the step-by-step, the algorithm," GenAI cannot follow it. That is the reality of the technology today. Also, another point is when you try to, it's more difficult to build something in plain English than to build something that is reliable in a programming language, especially in a low-code language like what we have in our studio. In English, it's much more difficult to reason about every single change that you, even a comma can change the outcome of the agent. Changing it from an LLM to the other can change the outcome. You never are capable of comprehensively testing an agent because it's in code. You can go through all the branches, and you can have a fair understanding of how the code works.
It's easier to train a developer to follow, to describe a rule-based process in code than to create an agent. That's clear to me. Why would you use, and even philosophically, why would you use a technology that is so sophisticated to do something that is rule-based?
Yeah. Yeah. I mean, it's like for in my viewpoint, Daniel, why would you hire a PhD to perform a clerical task, let's say that way, number one? The other thing is when we talk about governance and trust layer, for me, it's very important to always get a consistent answer. If there are certain automations that will give me a consistent answer every time, I would rather like to use that versus an agentic. As you said, I think agentic, it's not guaranteed that every time you get the similar consistent answer. I see there is one more question out there, which is, you know, when we talk to our customers, how do we talk to them about ROI with regard to Orchestrator, UiPath Maestro, and how difficult it is for our customers to build their own orchestration platform like Maestro?
Building a platform like Maestro, it requires years of engineering and teams of hundreds of people, Hitesh. This is not something that is easily achievable. First of all, I want to make sure that the audience understands we've built Maestro on the top of our existing Orchestrator. We started building our Orchestrator around 2015. I think it's achieved maturity probably around 2020. On the top of our Orchestrator that can provide the security and the governance and deployment and managing of robots, we've built the Maestro, which is more like a workflow engine, but it's combining. Building these capabilities is extremely difficult. There is no point in doing this. In this way, you can start building from scratch any other. It applies to build and buy to most of the software. It is not a simple system. This is an extremely complex enterprise system.
Yeah. Daniel, as I think about ROI from my viewpoint, RPA by itself or agents by itself or maybe Orchestrator by itself has relatively limited value. When you integrate all of them together along with the ability to go and play the Switzerland role, in my mind, that actually lays a strongest foundation for autonomous workflows. I feel like that has the maximum ROI from the way that.
Yeah. Because at the end of the day, what customers want is outcomes. And we provide the outcomes. You cannot have random agents deployed and spread around. You need to put them in the context of an end-to-end process with the workflow, with orchestrations, with humans in the loop. In the end, that provides the outcome.
That's right.
I think the other thing maybe perhaps that's worthy mention is time to value. Yes, in theory, could someone build, there are basic frameworks that help you string together agent work or some robotic work. It does not have the governance, the observability, the controls, the evaluation sets, the ability for you to inspect and then collaborate on it together. Those are things that the scarcest resource in most of our customers today are the data scientists. People just do not have a lot of folks ready to go build agents themselves. To have a surface where you can have developers and business end users looking at that diagram together, being able to figure out what needs to be built, and then getting it into production within days. We are talking about days for some of our biggest customers.
One of the biggest health care providers was able to turn around that claims scenario just like the one we did in 48 hours.
Yeah. That's great. I guess the other question is we recently acquired Peak. What are some of our learnings from this acquisition? How do we expect our vertical solutions to evolve over time?
Peak is looking like a really good acquisition for us. It's well received by customers. I think we can accelerate. We are seeing an acceleration of their pipeline. Our go-to-market is excited about adopting Peak, especially in our manufacturing verticals, mostly in the U.S., but also in EMEA, Germany in particular. It is also a lot of lessons learned from Peak. Peak used the model like forward-deployed engineers in order to facilitate their agentic. This is a model that we are looking more and more to adopt ourselves into helping our customers in the early phase of agentic. We will use their model to also build other specific vertical agents. We already identified a few areas. We are putting in place similar teams to build the technology.
Yeah. Daniel, there's one more question here. For customers who are using other RPA technologies and you mentioned such as Blue Prism for a minute, what do you think are the key advantages of our offering versus theirs?
I think we are seeing a bit of an acceleration of customers that want to migrate their RPA solutions. Blue Prism in particular, I think, has not evolved in the last quite a few years. I always said it's a good solid platform for RPA alone. When you try to extend it, it's a very difficult platform to work with. Now customers have realized the importance of having an overarching platform. Again, I want to make a case. Combining many agents will need RPA robots. Managing agents and robots in the same platform offers tremendous advantages. You can have even the existing center of excellence is working. It's an extension. It's a natural extension of your already automation programs in place. This is why some customers are seeing, well, maybe this is the end of the road with this particular technology.
Why are we not switching to a technology that offers me a future?
Great. I see we are at the top of the hour here. Graham, thank you so much for the demonstration. Daniel, thank you so much for your insights. It was very valuable. Daniel, do you have any closing remarks for the audience in the call?
I would like to thank everyone for their time and for being with us today. We are looking, as always, to connect with many of you throughout the quarter. We are already here maybe to provide more information and everything you guys need. Thank you so much.
Thank you.