 Welcome back everyone to theCUBE and our live coverage of Forward 6, UiPath. I'm Rebecca Knight, your host, along with my co-host and analyst, Dave Vellante. We are joined by Graham Sheldon. He is the Chief Product Officer at UiPath. Thanks so much for coming on theCUBE. Thanks for having me. So you were on the main stage this morning. You made a lot of product announcements. We're going to dig into in this interview. But I want to first ask about you and your role because it's a relatively new role. You're obviously a tech veteran. You are Microsoft for many years. I'm interested in your impressions of the company and its business and how things are going so far for you. Yeah, absolutely. Yeah, as you mentioned, I joined UiPath back in December and it's been a really wild ride. I found that their values about being bold and fast and yet humble are ones that really resonated with me. And I spent a bunch of time working on AI and on Microsoft Teams beforehand, but automation is a fairly new space for me and it's so broad and vast. So I've had a lot to learn over that time and so coming to events like this, being able to learn what matters most to people is really helpful to understand what's going on. I love what you said about the values in terms of being fast but then also being humble and having humility. So how do you practice those values as a Chief Product Officer? For sure. I mean, I've always been a big believer in the mantra that you have to have strong opinions loosely held. And what that means is that you have a really strong hypothesis about how to solve a problem or attack a challenge or go after an opportunity. But you need to be really open to learning and the data that you get if those things don't work out the exactly way that you want to. And then be able to iterate on that and build the product with customers that they really want in the first place. So let's talk about the shift from point product to platform. That's largely done. I mean, in terms of you've never done but you've made that transition. And so now you're building on top of that platform. That's right. So you've made some announcements today. Autopilot is the big one. Explain the news, what you guys announced and we can maybe get into some of the functionality and value. Yeah, so we announced several things today. The first and foremost is that we're big believers in the generative AI movement. And the combination of generative and specialized AI is what makes UiPath really unique. For generative AI, our approach is to support all the latest and greatest stuff. So we announced a bunch of new connectors. So regardless of which platform you've bet on, you can use the latest and greatest stuff. LLM, you're saying? In terms of, yes, those foundational models. Exactly. It's like Lama 2 or Claude, the bedrock on AWS. If you use Microsoft with open AI, if you use Google and their Vertex or you use AWS and all the panoply of models that they have, we've got you covered. So that's sort of the first thing. The second thing we talked about was something called the UiPath AI trust layer. So with this new generative AI capabilities, people are still a little concerned to make sure that we use their data the right way. And so that AI trust layer makes sure that when you use those third party models that none of the data that you don't want to expose gets used inappropriately. So that means having rules for governance, auditing and monitoring to make sure you know when the data is being used and then being able to mask and filter some of the data so that it never gets to those models in the first place if you need to. And then the last thing, as you mentioned, but not least was a really round this autopilot thing. So combining the best of generalized and specialized AI for these brand new scenarios to help developers be more productive, to help testers be more productive, analysts and soon everyone, that's going to be the really magic moment that opens up the next wave of automation possibilities. Okay, so you're delivering with the connectors, you're delivering optionality. That's right. You don't care what LLM, we support them all. The second piece here, the trust layer is really interesting because you want to make, basically you want to fence off the data from the LLM vendor. You don't want the LLM vendor to have access to that. Plus you're doing this across multiple clouds. So there's variability across clouds. One might assume it was the cloud vendors' responsibility for doing that, but it sounds like it's a shared responsibility to use a cloud. Yeah, we want to make absolutely sure that our customers have come to know that UiPath is a company they can trust. And we want to make sure that that standard of responsibility is met by all of the vendors that we end up working with. And so yes, we add value added services on top of that, including some of the things I already mentioned. Yeah. Okay, and then I'm looking at those article on SiliconANGLE, a company said autopilot is so powerful that it'll be able to transform paper documents into business applications within a single click. Okay, you got me, explain that. Sure. So yeah, on stage today, Nupur and Ani and I got to show folks that with a form that was previously just a PDF, that people would have to print out and go through and fill out, she was able to just take that one PDF, upload it, and then we would use our specialized AI and our generated AI to be able to create an entity, a data record for that, to create a form that you can fill out, digital form with an app, as well as a workflow behind it that can then orchestrate how that data goes across all of the systems that you need. And in her four minutes there, which she maybe spent filling out those forms and processing them, she created an end-to-end solution. Maybe it wasn't just one click, Dave, you got me there, but it was one step to get that done. But the impressive thing is the whole workflow piece comes because you can operationalize that now. That's the big difference. The other interesting part of that is the specialized capability. I wonder if you could talk about that a little bit more because I've written about this as generative AI, a tailwind or a headwind for folks like UiPath, you have to make it a tailwind. You know, or else you're going to be in trouble. And Daniel spends a lot of time focusing on that. The product folks obviously do that. So I wonder if you could explain that specialized model because that's a way that you're ahead of the generic models today, but of course it's like cybersecurity. It's like this game of leapfrog, right? So what specifically are you doing to create or enable those specialized models and how do you stay ahead of the mainstream market? Absolutely, I'm glad you keyed in on that. So for specialized AI, we've been working on that for many, many years. Before I joined, where it started was on computer vision so that you could really help computers understand what's on a screen and how best to access it when you're doing a UI-based automation. Well, it evolved from there and we added document understanding. Last year we added re-infer which became communications mining. And we really deeply understand screens, tasks, documents and the end-to-end workflows in such a way that you can both build your own or use one of our 70 plus built-in specialized models that are already tailor made for the processes that matter to most companies. And what I actually, I forgot this actually in the announcement that we talked about before. In the past, one of the bigger challenges with specialized AI was how much data and how much time it took to train one of those new models for your business or for your process. Well, now that we've reduced that by up to 80%. And in cases where you are building, you need to build one of your specialized AI models because it's faster and more accurate. You can use active learning in document understanding or generative labeling to do that labeling before you. What that means is that you don't have to have a machine learning degree or even know how to do data analysis. Just with a few clicks, you're working with the system to go build a specialized model in hours and days, not weeks and months. The interesting thing about that is data quality is obviously an important ingredient to having good AI. I was looking at a survey, I have to look it up. It was like the number one use case was using the AI to help us improve the quality of our data. Maybe through some auto categorization or maybe some cleansing. Are you seeing that? How important is that in terms of getting the fidelity that you want out of the models? Yeah, it's critically important. And I think you've touched on something that a lot of people are starting to think about, which is how do we build models that are going to be really robust to changes that happen over time? Invoices don't all look the same. Emails certainly don't. People change the language that they use all the time. New topics come in, new products get launched. You need to stay ahead of that. And so only with a platform like UiPath are we building these models, the specialized models in a way that they adapt over time. They learn from the feedback you give them when they get used. So using things like validation station. If the model is kind of confused and maybe it hasn't seen something before and you want to pass that through to a human expert to make sure, you can use us to do that. In addition, if you have a critical decision that needs to be made and maybe you don't want to have that, the AI just make the decision for you, you want a human in the loop, you can use Action Center to be able to go ask someone to actually sign off on that check with an extra, that might have an extra zero in it or make that really critical decision about who to hire, right? Those are the kinds of things where data quality and the ability to adapt over time are super critical. Do you ever think that it's going to make us all lazier though, I mean, just because as you said, needing a human in the loop, which I agree, there needs to be an expert. There needs to be a human, not for everything, but for those critical decisions. But do you think that we will end up deferring more to the technology or just become ignored to, it can take over so much of our tasks, I've almost forgotten how to make these decisions. I mean, because I'm a believer that it can take away the mundane task that we do. I'll answer that. I'll answer that. I'll answer that. Well, okay, yeah, all right. I mean, I think that if that is a, what do you, what's your take? It's a really great question. It really is. Honestly, you know, any time a new technology is introduced, you're not really sure exactly what the ramifications you're going to be that you haven't thought of before. And so I think we need to be thoughtful about that. I do have a lot of optimism though. So first of all, you probably saw this morning when Rob had his keynote, he had put his hand up and asked people, hey, who, you know, who's looking for more work to do? Well, this guy, but yeah, maybe Dave. I'm just looking to be more productive, that's all. People are inundated with work today. I think that's just everyone's reality. And so no one's really like being able to focus on the stuff that really matters is what is the real end game that we're trying to get. So that you can, you know, solve the customer's problem so that you can grow the business so that you can, you know, bring your service members home. Those are the kinds of, you know, things that people ought to be focusing on. But my optimism about that specific question, because it's an excellent question, is if you're using specialized AI in particular, you can actually understand how confident the model is that it's making the right decision. And if it's 99% confident that it's got something right and maybe it's a smaller invoice that you don't need a human to look at, great. Send it through. But if the model's only 80% correct and maybe there's a big amount that's there, well, instead of the human getting lazy, you need to be able to build your experiences and your automations in such a way as to be like, hey, you should take a pause here, make sure that this is right. So you talked about the Gen AI connectors, the UI path, AI trust layer, so many acronyms here. Lots of AI. And Autopilot, what is next for the platform? So I think where we're going to go next is we're going to still double down on specialized and generative AI for sure. Autopilot has a few tricks left up its sleeve that haven't been announced yet. They'll come tomorrow. For the normals, like me, yeah. Yeah, for everyone. And you're far from normal. But no, we're going to be doubling down on AI for sure. Then we're going to be thinking about how to think more about end-to-end processes. So there's this movement afoot to try to go find and discover the right processes to go after and how those things are working. We have process mining and task mining today, and those are things in the discovery part of our product that are really important to folks. Well, I think we can go further than that and really try to stitch together and help people understand the connections between different tasks. So based on what people are doing and the data that we're seeing and the business entities that go across those processes, we can start to build a picture and visualize and monitor what's actually going on in your organization. Today's tools are really good at sort of the top down. They're not as good at the bottoms up for how to paint that picture and then help people take action and automate that. So I think that's one of the other bigger pieces of what we're going to do. And then we're going to stay on top of our game. We're going to continue to have the best in class automation for UI, for APIs, and obviously AI. And try to bring that all together in a cloud platform that's going to be even more trustworthy and governable and transparent as possible. So for instance, we're very soon going to be FedRAMP certified and be able to go through public sector. There's lots more certifications and security and privacy things that we're going to go after to make sure that people can make that bet on the cloud with us. So Graham, when you take that system view called a system view that you described, it reminds me of what's that famous picture book from the manufacturing? I don't know if you've ever seen it. Sorry, I can't remember the name, but the basic premise is if you just optimize on one part of the system, it could create bottlenecks elsewhere. You're just squeezing the bottleneck somewhere around. So you have to really take that system view. My question is if you've got that system view, can you actually begin to predict what the impact is, do the what-ifs and help the humans ultimately decide what they should put into operation? Yeah, it's a fantastic observation. It's only when you're looking at the full end-to-end picture that you can then do those things. So I think you're keying in on something that we're very excited about, which is being able to do things like simulation, decision making, and using the intelligence of AI to help you figure out not just where a local optima is going to be, but where a global optima is likely to be. So that you're not doing exactly as you described, optimizing one part of the system at the expense of maybe the other parts of the system. So you guys obviously are the kings of software robots. Do you see, or maybe it's happening already, that moving into the physical world, people, places, and things where the physical world is now increasingly automated using your software? Yeah, it's a great question. I haven't personally thought a lot about that. I've been up to my eyeballs trying to figure out the software robot space, but I know we have lots of customers in manufacturing and retail who are blending some of that stuff together, for sure. Yeah, I wonder, I mean, because that would increase your TAM, probably by 10X. We've got our hands full with the software part. I'm not sure if it's applicable. Yeah, but once you get to five billion, you'll have another problem to solve. But I don't know if it's applicable. I would imagine it would be. I mean, it's automation, it's AI, it's understanding systems. It's certainly conceivable. I mean, a lot of the techniques are applicable, but it's an interesting idea. I'll mention that to Daniel in a couple hours. There you go. Yeah, these are our topics for Forward 7. Yeah, we don't want to get ahead of ourselves. Graham, thank you so much for coming on theCUBE. A really, really interesting conversation. Thank you both. I'm Rebecca Knight for Dave Vellante. Stay tuned for more of theCUBE's live coverage from Forward 6.