Liz Bearce: Thank you for joining our webinar, co-hosted with UiPath and Lydonia, to discuss the future of work, the power of agentic automation.
Logan Emerson: So, agents can do a lot. But they’re usually 3 types of workers that are going to be delivering on a process. You got agents, humans, and robots. They’re going to perform specific steps of every business process. You want to get as well depending on your needs as a business. But often the idea is to get as little human touch as necessary. So, the humans can do more complex in-person work while the more boring stuff, or slightly more boring stuff goes to the agent. And then the extremely boring stuff, say, maybe data entry, goes to a robot.
But how do you know where a particular instance of one process stands? So let’s say again, in one of Vinnie’s cases. Where is the work bottlenecked right now? How do we understand what type of worker should perform this or that step? Agentic orchestration is going to be the answer to that. So at UiPath, we’ve got a history of orchestrating and monitoring RPA workflows. We don’t really shout this from the mountaintops quite enough, but the design of our orchestrator for RPA processes is necessarily unique because it needs to give actionable information. Why did this workflow fail? What should I do as a result of this failure? How do I action on this information?
This isn’t just about raising error codes, which you could do with any sort of monitoring. It’s about making errors fixable, and errors in workflows like RPA workflows, not even agentic workflows, but RPA workflows. They’re meant to emulate. They’re meant to emulate what people used to do before the automation was there. And that means that there’s more information than just a simple error code. It’s like, “Hey, I’m here. This is what’s happening. This is the context before here. Here’s what still needs to happen.” And potentially, here’s the value of not fixing this problem in a quick timeframe.
So, orchestrator just for RPA. Pretty cool orchestrator or agentic orchestration, orchestrating agentic workflows, and really agentic workflows from not just UiPath, but say some of our competitors who are doing similar things. We want to be able to be the house where you can manage all of that and understand what’s going on with any of your agents, any of your workflows, and your human processes altogether. That’s the idea of agentic orchestration.
So, we’re looking right now at a view of agentic orchestration. So, this is really multifaceted. What you’re seeing here is what can be developed in Studio. So, you can create a process diagram and understand, “Hey, this is how this process ought to work.” We already have a PDD, let’s get this using BPMN style tooling. So, we’ve had a lot of requests to have BPMN integration. This is that. You can import BPMN, you can export this as BPMN, and this also—anybody that’s familiar with process mining—this has similar functionality to that. You’re able to actually watch tokens of work. Instances of a particular work stream flow through this and understand: Where are we in the process? What’s going on? Who’s going to take it? Is it a human that’s taking this next? Is there a robot that’s up to work right now? Is it an agent that’s making a decision about what to do next?
This is all in one space, and we really envision all agents that you may have, again, the competitors have them, we have them too, bringing your agents into the UiPath platform and understanding who’s doing what, why, and what’s the value of them doing that work. Beyond that, I do think there’s a bit of a philosophical point that I think is interesting here. We’ve always had to, you know, with artificial intelligence, we’ve always had to understand, how do we communicate with artificial intelligence? Well, lately, it’s been with the English language, with the advent of large language models, and it’s been pretty exciting since that happened. But the English language isn’t a perfect way to describe a process. It’s a pretty imperfect way to describe a process. We could look at this process diagram on the screen right now and understand pretty well the function that’s happening here. I won’t read through it. I know it’s not the point, but it’s definitely going to be easier for us if we did the exercise to understand what process is on the screen.
Now, with that BPMN workflow, as opposed to, let’s say, two pages of text thereabouts, or even one page of text just reading it straight through, we’re going to not have the visual of how the work is moving. This is a better mechanism. This BPMN is a better mechanism for communicating what is happening with a process. And we think that communicating with artificial intelligence with BPMN may actually be a more fruitful language to share here. I’m not saying that we are changing the game in any particular way as a result of this idea, but it is some of what our roadmap, part of what we’re thinking about as we continue in this communication with AI as we go forward.
All right. So why should you choose UiPath and Lydonia for agentic automation? I think there’s one build that we can look at on this one, Liz. Awesome. So, I won’t belabor all of them. Yeah, you can go to the next one right there. So, enterprise tools, human in the loop, low code, vendor agnostic, and end-to-end orchestration. I’ve hit on a few of these already, and I do think they are the most important ones. So, having enterprise tools, you’ve got the whole UiPath platform, and Lydonia is as good as anybody at leveraging the whole UiPath platform. So, it’s not just API calls from your agent to a service. It’s, “Hey, sometimes a UI is going to need to get involved.” You need that core RPA to be a tool for the agent. You might need intelligent document processing or communications mind to take care of part of the job.
Large language models are great for a lot of things, but these tailored tools for communications, binding document understanding, UI automation, they’re already better and will always be better because they have better communication methods with what needs to get done than a large language model would be at doing the work. The large language model, to be clear, is going to be understanding context and deciding the next best action given its toolset, and that toolset is the UiPath platform.
Vendor agnostic, I’ve touched on this a couple of times, but there’s a meme going around internally at UiPath, talking about us being the Switzerland of agents. So if you want to use a third-party agent or an open-source agent, we welcome that. There are reasons to choose them in some scenarios, there are reasons not to, and to use a vendor-agnostic, at its core, agent built with UiPath. But we don’t really care what kind of agent you want to use, and if you want to use your own tenant or model that you have a license for, like, we’re going to have you able to use whatever model you want to use to power your agents. But if you’ve already got your own license key for, name a model, a large model that you want to be the engine for, lack of a better term, for your agent, you can do that. We’re fine with that. So we’re not trying to lock you into our platform like some others are. And then end-to-end orchestration, I just got off the last slide talking about that. I think it’s an important point, but maybe I’ll tease for a future webinar. We’re going to talk more about our end-to-end orchestration, and then we can move on to our platform differentiators, which I won’t belabor. I want to make sure we have time for Q&A.
But quality and accuracy. So, you’ve trusted us with your RPA for some folks, for a decade. We go into systems of record. We do work with robot compute. That’s not really changing. That’s not significantly changing. Feel free to jump into QA and challenge me on that. But what’s fundamental here is that, again, the agents are making decisions about what to do next, and that could be, and will often be, to fire off core RPA to get some job done, and it will give it the context to be able to get that job done better than it could have had it not been kicked off by an agent. So that’s an important piece of this: runtime evaluations, design-time evaluations. We make sure that you are comfortable with how your agent is going to be behaving and what it’s going to get up to before you put it into even dev, staging, test, broad QA, whatever. We want to make sure you’re comfortable with this before you’re launching.
Management monitoring. I think that’s relatively self-explanatory, but again, when we think about that BPMN tooling, you’re really going to be able to see, where is this process, this instance of this process right now, who’s next, why did this happen, why did this particular instance of this workflow take this path versus whatever three invoices ago it followed this other path? There will be answers for that, there will be reasoning as to why that happened, and that will be surfaced to administrators who are looking to monitor these processes. And governance and trust. I think that ties pretty well into that as well. We’ve got deployment, flexibility, compliance, and security standards are all there. And really, it’s just the same stuff that you’ve already come to trust with UiPath: just better decision-making about when to fire, why to fire, and what to fire as far as your RPA workflows doing work against your business processes.
So I hope that’s illustrative of why Lydonia and UiPath are a really good choice for agentic automation going forward. How to get started? Really, so we’re not generally available yet. We have some private previews kicked off, but we’re going to go GA, or generally available, in our 25.4 release sneak peek. I believe we’re looking at early May for that, but don’t hold me to it. There will certainly be announcements. You can get on an email list, and Lydonia will certainly keep you posted on when things are coming out. But how to prepare for that time? Building tools. So tools are going to be those RPA workflows that kick off really any work that needs to be done. Still, you, UI. Some folks could start with agentic and not do any RPA, but I don’t think that makes a lot of sense. Your processes that are already working well, make sure that you monitor them, that you manage them well, and that you have good tools ready to go that could be used by an agent. Maybe that means cutting them up into smaller chunks and saying, “I want an agent to be able to do parts of this process versus the whole process,” and decide when to do which parts of this process. These are things that we can consult on as well.
Unlock more data, start with supervised. So, starting with supervised if you wanted to use autopilot for everyone. Now, that’s officially our 1st agentic experience. Instead of handing the keys over to an agent to make the next decision, you could use the agentic experience of using UiPath assistant and autopilot for everyone, which is more of a chat interface to say, “Hey, I need to get this job done.” Then, the agent on the UiPath Assistant interface will say, “Great. Give me the context I need,” and you will have a more transactional, step-by-step process. As you become more comfortable with that, you can hand more of the process over to a less attended agentic experience or a more unattended one. That’s not the official language we’re using for agentic, but for those who are more familiar with robot terms, I think that makes it pretty clear: supervised is more like that attended assistant agent experience, whereas the agent, builder agents, and 3rd party agents are more like unsupervised or unattended agents. Excellent, and that is what we have to present.
Liz Bearce: Logan, Vinny, thank you both so much for your time, and again, thank you to everyone who joined today.
