Google Cloud Next 2025 | Dallas Dolen & Dan Priest, PwC

Clips
More from Google Cloud Next 2025

Dan Priest

Chief AI Officer

PwC

Dallas Dolen

Global Google Cloud Alliance Leader, US Tech, Media & Telecomm Industry Leader, and Board Member

PwC

play_circle_outline Enhancing Specialized Tasks with Agentics: Multi-Agent Models and Continuous Learning for Autonomous Applications

play_circle_outline Differentiation between hard-coded microservices and Agentic software

play_circle_outline Utilizing Agents for High-Stakes Health Systems: Importance of Trust and Top-Down Goals in Decision-Making

play_circle_outline Importance of responsible AI framework and security architecture in deploying agents

play_circle_outline Role of diverse teams of humans and agents in solving complex problems.

play_circle_outline Transforming Contact Centers, Finance, and HR Teams with Pre-Configured AI Agents for Compliance Tasks

Info
Transcript

Dallas Dolen & Dan Priest, PwC

Dan Priest

Chief AI Officer PwC

Dallas Dolen

Global Google Cloud Alliance Leader, US Tech, Media & Telecomm Industry Leader, and Board Member PwC

Exploring Agentics in AI Innovation at Google Cloud Next

Join us for an engaging discussion with Dallas Dolen of PwC and Dan Priest, an AI strategist at PwC. Recorded at the Google Cloud Next 25 event, this video examines the innovative landscape of AI and its transformative potential within enterprise frameworks.

In this video, we speak with Dolen and Priest, passionate leaders from PwC, about the latest advancements in AI technology and their implications for business. Focusing on Agentics, they illuminate how generative AI has evolved fro... Read more

explore Keep Exploring

What do agents do and how do they create value in a multi-agent model? add

What are the differences between static hard-coded microservices and agentic software, agentic assistants, and agentic digital workers? add

What is the process for deploying high-stakes AI agents and ensuring trust in their decision-making capabilities? add

What considerations should be made when granting access to data for AI agents? add

What are some benefits and potential applications of diverse teams consisting of both humans and agents from different platforms in solving complex high-value problems? add

What are some areas where artificial intelligence and automation could have a significant impact right now? add

bolt Powered by CUBE AI

Dallas Dolen & Dan Priest, PwC

search

Dallas Dolen

>> To come by and make faces while we're up here.

Savannah Peterson

>> Good afternoon, Cloud. Whoa, excuse me. Good morning, CUBE community and welcome back to fabulous Las Vegas, Nevada. We're here already off to a spicy start. So spicy. I don't even know where I am this week. My name's Savannah Peterson. We're bringing you all the laughs and all the insights with Dave Vellante. Dave, how is your extension cord collection?

Dave Vellante

>> Oh, garlic bread for lunch. That's what I know. It's a joke.

Savannah Peterson

>> You got to watch the segment that just aired to get that one. That'll get you to tune in. Speaking of laughs and fun, very excited to bring back a CUBE alumni this time. Dallas, thank you so much for coming.

Dallas Dolen

>> Thank you.

Savannah Peterson

>> And bringing your friend, Dan, to join us today. This is going to be fun. Last year when you sat on this stage, we talked about tax and sustainability, which was one of the more unexpected conversations for me personally. We had Justin on with us. That was great. And so I'm very curious to hear what you would like to talk about today. Do you have a surprise?

Dallas Dolen

>> Well, we may still talk about tax later, but that's a different conversation we'll have today. I think Dan and I are going to talk about Agentics and really what's happening and has really happened over the last 24 months. We go from no one knew what generative AI was all the way through to everyone has their favorite Agentic and their friend that's waking up and telling them what emails they should read and what TV shows they should watch, and which garlic bread they should eat. So we're moving a whole different direction here as far as tech is concerned. But Dan is our AI leader for the firm, and so brought him along to really bring some substance to the conversation around what PwC is doing with Agentics and AI generally, and how we're working with Google Cloud in developing our Agentics on Agentspace and some of the products that we've been coming to market on around the same.

Savannah Peterson

>> Stoked about it. You have a pretty big job right now, Dan.

Dan Priest

>> Yeah. It's fun.

Savannah Peterson

>> A couple people care about what you're doing right now and they're talking about it in this room.

Dan Priest

>> 2025 is better than 2024. 2024 was about proving it works. 2025 is about scaling it up, getting value from it.

Dave Vellante

>> It being agents or it being GenAI?

Dan Priest

>> I'd say 2024 was very much focused on proving GenAI use cases. To Dallas's point, 2025 is very much about activating and realizing value for the enterprise through these Agentic workflows, Agentic organizations, these agent frameworks that we talk so much about.

Savannah Peterson

>> So how are they doing that? How are they achieving that value? Give us some examples.

Dan Priest

>> So agents, in short, they're basically programmable applications that you bound them, you give them a task, and within those bounds they can work autonomously to achieve the outcome you want. And what we found was if you design these multi-agent models, so imagine any other team, you'd have a planning agent, a design agent, a due agent, a test agent, an orchestrate agent, and they're all doing different things in a workflow. And that multi-agent model creates an incredible amount of value. It's kind of like a human team. The more team members you have, the more diverse that team is, the higher performing typically that team is. And that's what agents do for us.

Dave Vellante

>> And when you say programmable, people might think, "Oh, it's a static hard-coded microservices." But that's not agents. And then you've got Copilots, digital assistants, you've got agents, you've got multi-agents, you've got bots. We know they're not bots. Help us understand the difference between sort of hard-coded programmed microservices and Agentic software, Agentic assistants, Agentic digital workers.

Dallas Dolen

>> I'm actually Dan's agent, by the way, just as a separate point.

Dan Priest

>> He promotes me very well.

Dallas Dolen

>> Everybody, every time.

Savannah Peterson

>> Actually wait just for a second. I thought this was really cool. One of my girlfriends shared recently with me that she turned a GPT into a hype woman for her to help her get pumped up and motivated for interviews and whatnot. So trained it on her resume and her skillset. And then now when she goes into interview for jobs, she has her hype person. So I mean, Dallas could be your hype.

Dallas Dolen

>> I'm Dan's hype guy. That's actually what I am. Yeah, that's a great description of it.

Dan Priest

>> But it is funny, you're not the first person I heard that from. People are developing relationships with their agents and they joke with them. They get coaching advice.

Dave Vellante

>> Argue with them.

Dan Priest

>> Argue with them.

Savannah Peterson

>> One of my other girlfriends is using it as a dating coach, which is very interesting as well.

Dan Priest

>> I have some friends who need it. So you can do a lot with agents, but the difference is, so we have here Agentspace, Google Agentspace, it's a platform you can assign agents to a human. And those agents, when I say programmable, this is a very light program. This is basically in natural language, "I want you to perform these things. Here's some instructions. Here's the data you have access to. Now go." And then you can keep going back to that agent and that agent learns and deepens its specialty and then it starts colluding with the other agents to share information and improve the performance of whatever you've tasked them to do. So it's a little bit more dynamic than the traditional hard-coded solution.

Dallas Dolen

>> And I'll layer on to that.

Savannah Peterson

>> Yeah. Please.

Dallas Dolen

>> As we start thinking about what happens, I mean joking aside, how individuals are using it. If you think about the enterprise application, which is the big componentry that Google's working on here today, that's a place where as a part of Dan's team and the work that he's doing in the US, we've looked at our strategy to say, "Hey, what traditional services do we do from a strategy point of view, from a workforce point of view, and also from a responsibility point of view? How do we take all those different perspectives we bring to our customers and actually layer on the use of this new technology to help people be more effective? Again, whether it's hyping themselves up, their dating life, how they interact with other humans, getting good type of advice, or it's how are they answering emails? What are they doing to get organized for a given work day?" I met with the CEO of a company a couple of weeks ago. He said that the Agentic that he's built that operates on his email platform allows him to get up in the morning when he has 50, 5-0 actionable emails. He says, by the time he even thinks to open his email application, 30 of them have been responded to. The other 20 have been organized by prioritization based on what the Agentic knows about him and how he deals with these people and gives him a point of view on how to deal with them both operationally, what do I do first and what he might say to the person in response to it. So you start thinking about hyper-customization, but also making people much more effective. The workforce dynamic, I think there's a huge opportunity to make people way better at their jobs and make them a part of what they're doing, not replacing people, which I think is an important part of this whole story too.

Dave Vellante

>> So how did that solution happen? What did that individual have to do to create that outcome?

Dallas Dolen

>> Yeah, I mean, Dan, you're probably better to speak to how these things are being built, but it's almost zero to no code, right? We're talking point and click as far as capability.

Dan Priest

>> Yeah, I mean you have to have a platform to build on. That's why Agentspace is so important. You have to have a basic knowledge of prompting and tasking, but it's all in natural language. Read my emails, identify the ones that have an action for me, identify the ones that are addressed to me and tell me where I have action. These are like you would task any team member to do something. And to Dallas's point, it is less about the technology and the important focus should be on creating high-performing teams. And so high-performing teams, what we know about high-performing teams is they tend to be diverse. They tend to be from different parts of the organization. They can be from outside of the organization, but a high-performing team comes together around a task. It's very cohesive. It's easy to work with. And that's what we're building with these Agentic workflows. We take agents from one platform from another. Most clients have multiple platforms that they're trying to work within. We can pull it all together in a seamless, pretty elegant solution that all these agents are working well within.

Dave Vellante

>> So I envision an outcome and then what? Watch me or let's try it. You go through and give me a test run. I'm going to give you some human input, thumbs up, thumbs down, whatever, adjust it. And then over the course of, I don't know, some amount of time, doesn't sound like it's a long time, get to the point where you're comfortable saying, "Okay, you handle it. If it's a client or a customer, I want to be able to review it before it goes out. But otherwise, if it's internal, go ahead. If it's trivial, ignore it or whatever." Is that the process?

Dan Priest

>> That's the process. And we are deploying them into some high stakes scenarios. So for instance, we've built agent solutions for health systems and based on the ICD-10 codes, the diagnostic codes, the agent can present to the clinician, here are the natural pathways as options. The clinician is still treating and making the medical decisions, but they, based on the diagnostic code, can give treatment options that the clinician could consider. And they've got the benefit of calling all the best treatments, the most effective treatments, bringing it to the edge to that clinician and making a difference to the patient. But that's a high stakes scenario.

Savannah Peterson

>> I can imagine... To build on that, I can imagine in these scenarios, there's a lot of trust that folks have to put into you as you're deploying this out. Talk to me a little bit about the trends you're seeing there and how you ensure that people have the confidence to deploy agents like that.

Dan Priest

>> It is a thousand percent about building trust in these solutions and the humans are the hardest part of getting the value from these...

Savannah Peterson

>> Shocking.

Dan Priest

>> Shocking, right?

Dallas Dolen

>> Present company excluded.

Dan Priest

>> You've been very pleasant this morning. But you have to make clear that that person in that workflow has the decision rights, is going to have an enduring value proposition in that workflow and is exactly as Dallas framed it, this is a performance boost to them. It'll get the best performance out of that individual and that team possible because it's bringing a lot of intelligence to the decisions they're making. And so building trust is essential. The key to doing it is being clear about the human value proposition.

Dave Vellante

>> So at what point, or maybe we're at that point, will Agentic be able to be guided by a set of top-down goals, revenue growth, maybe there's margin constraints, and then execute with agency such that it will drive whatever, greater productivity, greater revenue according to those sort of top-down hierarchical goals? Are we years away? Are we there today? Help us understand that.

Dan Priest

>> I don't think the human is going anywhere anytime soon. Humans are very good at first principles thinking, creative problem solving, scientific methods, relationship management, strategic decision making, and all of those things will be true now, tomorrow, next year. And I think what will happen is there's a lot of scut work typically in workflows that will get delegated into the agent workforce and the human can elevate to their highest value proposition. And so I'm not worried about the human going anywhere.

Dave Vellante

>> Yeah, my point wasn't taking humans out of the loop, it actually was keeping them in the loop. So for instance, take your clinician, human clearly in the loop there.

Dan Priest

>> That's right.

Dave Vellante

>> I'll give an example. Supply chain, I've got a SKU, might be air conditioners. In New England, you don't really worry about that until May.

Savannah Peterson

>> And so you really worry about it. Yeah.

Dave Vellante

>> But maybe there's a heat wave in March or April. Well, and I want the agents to give me a plan to reroute the air conditioners to New England and where do they take them from? Take into consideration distance, my margin profile, some of those top-down constraints and objectives that I put forth, and then maybe make here's plan one, here's plan two and then the human says...

Dan Priest

>> That's today. That's today.

Dave Vellante

>> "I like plan two," or "I'm going to..." Okay, so let me keep going. "I like plan two, but I'm going to change it. There's an exception here. So I'd like you to go back and redo it." Okay, the agent redoes it. That's today. Can the agent today learn from the reasoning traces of that human and then incorporate it in the next example, so when that exception comes up, the agent can handle it?

Dan Priest

>> Yes.

Dave Vellante

>> You're saying that's today.

Dan Priest

>> That's today.

Dallas Dolen

>> A hundred percent.

Dan Priest

>> That's the value of specialized agents too. So I was talking about multi-agent models, get that agent really good at the planning process, get another agent really good at the doing process, executing that plan, get another agent really good at testing that they did that plan or they executed per the plan. So that specialization and that continuous learning is happening today. All of that is well within the capabilities.

Dallas Dolen

>> Yeah.

Dave Vellante

>> What if- Go ahead.

Dallas Dolen

>> Well, just to layer on, I mean, I think you have another component of that, which is the cost of running all these things. Because Dan said, "Hey, well just go out there and get five agents and they can do this for you." What's probably going to end up happening is you'll have the ability to get really efficient on the building and the pricing around it. So it's not like you're going to be incrementally paying significantly more to bring each one of these things in. You're effectively customizing your LLM, which by the way, you might be pulling from different models because the use cases and the reasoning and the decision making might be different and the knowledge base from the models might be different. That's actually a place where and something, a product that we've come up with called Agent OS orchestration, but also what Agentspace does too, and all the announcements that Google has made saying, "Hey, we want to make sure that you, as the end user, have access to just about any model you want to have access to." Because some might be better at dealing with uncertainty like you're talking about in terms of perhaps weather or perhaps the responsiveness. And even, I mean, back to the individual interaction points, which one's friendlier? The most shocking conversations I have with people is, well, I kind of like the data that's in this model, but to be honest, this one's nicer to me. And so when I interact with it, this is the one that I want to interact with more because I find it to be polite in what it's doing. If you stop and actually think about if you're going to be interacting more often with these things and the prompts are happening and you're responding to them and vice versa, hey, chances are, that may be your interface, but behind it, maybe all these other things because the models they have access to and/or the data they have access to, like it's being real-time from a weather point of view or otherwise or supply chain, those are the ones you'd want to use. And I think that's where customization really comes into play. You'd pay a premium for that part of it. You wouldn't pay a premium for the nice thing probably, although maybe you would a little bit.

Savannah Peterson

>> I was going to say, do I have to bribe it to be nicer?

Dallas Dolen

>> Yeah, you do. Yeah, yeah, yeah.

Dave Vellante

>> But on your first point, if I understood it correctly, you're saying if you set a foundation for building these agents, the marginal cost of building the nth agent goes down and you get operating leverage from that foundation. What is that foundation? What are the first principles? What are the prerequisites that a customer needs in order to be successful at building out an Agentic framework?

Dan Priest

>> So first, so you started going down the trust path. Responsible AI is an important sort of prerequisite, knowing which data they have access to. You wouldn't grant broad access to all data. You would grant access to the data that's required to complete the task. And so you have to have a framework to do those permission-based assignments. You have to have a way to register when an agent is deployed and report on the performance of that agent and turn that agent off when that task or that job is done so you don't have a bunch of agents running around doing things endlessly. Right?

Savannah Peterson

>> Hamster wheel.

Dave Vellante

>> Agent party.

Dan Priest

>> Yeah, exactly. There's some groundwork around a responsible AI framework, a security architecture that gives people trust that these things will perform the task within certain bounds and there's a beginning and an end and you can monitor them and report on their performance. That would be that. The other piece is the data. Getting the agents specialized depends on getting them connected to the right high-quality data sources. And so I don't know how the dating app did it, but there's probably some dating resources out there that keeps pinging and it's a learning loop.

Savannah Peterson

>> And some of what I've seen from these sort of recommendations is they're pulling book resources or blog posts or courses on courtship, I guess in this case.

Dallas Dolen

>> There's lots of them.

Savannah Peterson

>> I'm sure, and I'm sure they call it courtship.

Dave Vellante

>> Can I follow up on that? Because the email example, you've got a confined set of data, but in order to take action, I might have to go into my Salesforce automation, I might have to go into my HR system, I might have to go into my ERP. So that's the access issue. But then the other piece is how do you harmonize all that data? So does there have to be... I mean, today humans have to do it and oftentimes they don't do it well.

Dan Priest

>> That's right.

Dave Vellante

>> So that seems to be a hard piece to automate, that data harmonization or semantic layer, some people call it.

Dan Priest

>> So two things. So I talked about the human value proposition and the workflow, and there's an agent value proposition too. So humans, interestingly, for better or worse, we can make decisions without data, right?

Savannah Peterson

>> Facts.

Dallas Dolen

>> We do it all. It happens all the time.

Dave Vellante

>> Yeah. Exactly.

Dallas Dolen

>> Gut feel.

Savannah Peterson

>> Story of my life.

Dan Priest

>> Agents need data. Humans don't like complexity. That's why simple, elegant user interfaces are so important. Agents, they do really well with complexity. They actually navigate those complex data architectures much better than... You just have to make sure you've given them permission to the right sources and establish the boundary conditions. And they can go into the HR system or they can go into whatever planning system that... They can access it and take the actions that you've given them permission to do.

Dave Vellante

>> And that's probabilistic AI, that's LLMs and GenAI, because... I just push back a little bit because when I ask an LLM, "What's the revenue of this company? What's Google Cloud's revenue in 2024? Go read its annual report." Maybe it's the particular LLM that I was using. It'll give me 2022 data or 2023 data when I go and validate it. So if I tell an LLM, "I want ARR, I don't want calendar year or a fiscal year. I don't want bookings. I want ARR." I don't trust the LLM to get it right. My question is, is there work that you have to do, maybe you guys are good at this, in making sure that the LLM gets it right?

Dallas Dolen

>> Sure. There is, but I mean, keep in mind the following. We're talking on a weekly basis. You have companies enhancing these models. Even in the last couple of months with Google, not only are they giving access to all the other non-Google-specific models, there was a big announcement around Anthropic earlier today.But you also have each evolution of Gemini coming. It's Gemini 2, you got 2.5, you have Gemini and the small models. Now you have to your question, deep research and the functionality there, which is actually where I think a lot of people are going to go get to the crux of the, okay, no, I really want to understand this thing because you're right, at a superficial level, quickly prompting something and hoping to get a good answer isn't always going to work. But with these new models that have the deep research componentry where you can see everything they're doing as you go along, you actually know how they're getting to the right answer. I think that's the big difference maker.

Dave Vellante

>> That's fair. I mean, it's getting much, much, much better.

Dallas Dolen

>> It's getting much, much better. And look, it's getting cheaper and cheaper. I mean, even on the downloadable application that you can throw onto your phone, the version of Gemini that you get is the latest and greatest. You're not paying anything for it, which is remarkable. Now if you want to start customizing it and you start working the model in a way where it's making sure that it's not giving you old data, it's not like you can actually start fine-tuning these models. That's what companies are doing. And you mentioned the healthcare application, right? We're really proud to say that we're a partner of the year around healthcare with Google this year because of-

Savannah Peterson

>> I see that pin right there.

Dallas Dolen

>> Yeah. Right. On the lapel.

Savannah Peterson

>> Yeah.

Dallas Dolen

>> Because of what we did actually in the healthcare space with them and a customer. That's a really good application of we didn't just take any generic model. We actually had to do a lot of work to make sure what the clinician was getting and how they were using it in that specific use case is something that you would trust. Back to the trust point too, and I think it's getting better and better and better and better. The thing holding it back of course, is speed of compute and how much are you willing to pay for that? How much can you get at the edge? How many people can access it at a given time? That's been a problem in the last couple of months. We've seen-

Dan Priest

>> That'll be solved though.

Dallas Dolen

>> That'll be solved, right. You got these stories of GPUs melting, right? It's a real thing when you talk about the power going through.

Savannah Peterson

>> We'll suss that.

Dallas Dolen

>> Yeah, exactly. But I think that the reality is as these models get better and as the ease of building the Agentics gets easier, you're going to have a situation kind of where companies, again, back to the enterprise use case, they're going to have to start getting a lot more programmatic about how they're allowing these things to be deployed and what's happening with them in their environments. But it's going to make their employees much, much more efficient back to my CEO example that I gave a little while ago.

Dave Vellante

>> But they could really screw it up if they don't think it's-

Dallas Dolen

>> That's right.

Dan Priest

>> Two things. One, again, the multi-agent model is powerful because you can have a tester agent saying, "Are you pulling the right data, and getting it right, and testing the quality of the-"

Dallas Dolen

>> Right.

Savannah Peterson

>> Like that middle manager keeping them in line.

Dan Priest

>> Exactly. Yes. And not to put it back on you, but the other thing we've learned is the prompts make a difference too.

Dallas Dolen

>> Yeah. It's true.

Dan Priest

>> It's true. I've run into the same problem with just the way I prompt and the way I ask the agent to perform a task. We'll either send them in the wrong direction or I get a good quality answer. So I've-

Savannah Peterson

>> It's like bad management. There you go.

Dan Priest

>> Exactly. And like human teams, these agents do better with more instructions and clear instructions. I've written some prompts, literally they've been a page long and you bold things in caps to really get the point across. "Do this, don't do that."

Dave Vellante

>> Sure. And then you lose internet and you have to write the whole thing over again. That's happened.

Dan Priest

>> But it is exciting. I think this notion of diverse teams coming together, humans and agents from different platforms to solve complex high-value problems, giving that performance boost to the human, that human is ultimately accountable for the performance of that process or the achievement of that outcome. It's really kind of exciting about this phase we're entering into and the promise it holds.

Savannah Peterson

>> It is. It is. And I love thinking about a highly specialized team around to support me in my many nuances without impacting the wonderful humans in my life.

Dan Priest

>> Yeah, exactly.

Savannah Peterson

>> This was wonderful.

Dan Priest

>> They have... And we've recently written agents on health advice. You can literally have health objectives and the agents will help you achieve your health objectives through diet, exercise.

Savannah Peterson

>> Little coaches.

Dan Priest

>> Vitamins.

Dallas Dolen

>> We should save me from having to give Dan health advice, which is exhausting.

Dan Priest

>> I will make it any day.

Savannah Peterson

>> Last question for you, Dan. And Dallas, I'll ask you actually to put this in the lens for Agentic and then you can put it in the tax lens for our next segment.

Dallas Dolen

>> Sure.

Savannah Peterson

>> What do you hope to be able to say? Since this is a consistent thing now, Dallas will tell you, I mean here he is on twice. We met last year. It's a whole shebang. What do you hope to be able to say this time next year at Google Cloud next 2026 that you can't say today?

Dan Priest

>> That we've scaled it. So we describe AI as a jagged frontier. Not all use cases are created equal. These solutions aren't ready for every part of your enterprise. But to Dallas's point, they are getting phenomenally good very fast. And we are developing pre-configured agents for contact centers, finance teams, HR teams, and this will move into product solutions that make the customer experience better. So in a year from now, what I'd hope to have is a full-blown library of pre-configured agents across the enterprise. Today we have pockets of excellence that we're building on and we're expanding around.

Savannah Peterson

>> I like that. Great way to put it. What about you, Dallas?

Dallas Dolen

>> Well, I'll just say we're focusing on four areas where we think it could be most impactful right now. They're largely compliance-oriented, things that people would have to do, small tedious tasks here and there redundantly throughout the day, starting at 8 AM and finishing at 2 AM. It's like how do we make our staff's life easier? You start taking away those tasks, right? Which that's the place where we think AI and Agentic starts doing. That is the next evolution. So by the time we get to 2026, we'll have solved for those primary areas that we're looking at, the big four. And then also how do we weed that all together throughout the big compliance process? And it's not just tax, by the way, right? You have lots of things going on in SOCs and internal audit and controls testing. There's tons of time spent on that. It is manual. It is tedious. It happens four times a year. No one likes sitting through walkthroughs. No one likes documenting it. So the application sets to all these areas within business gets really good, and I think that's what's going to happen in the next 12 months. We come back here. We have the same conversation. I'm sitting right next to Dan and I'm going to say, "These are the 10 problems that we've solved, and this is how many hours we take out of these compliance processes." That's the answer I promise to remember to you next year.

Savannah Peterson

>> I look forward to it. I love the promise.

Dave Vellante

>> Good value too.

Savannah Peterson

>> That was fantastic. You committed to that, Dallas.

Dave Vellante

>> Hard commitments here today.

Savannah Peterson

>> Yeah. I loved it.

Dave Vellante

>> Show me the money. Love it.

Savannah Peterson

>> Well, this has been an absolute thrill. Thank you both for taking the time today.

Dallas Dolen

>> Thank you as well, guys.

Savannah Peterson

>> This was a ride as always. And thank you, Dave, and thank all of you for tuning in to our three days of live coverage here at Google Cloud Next in Las Vegas, Nevada. My name's Savannah Peterson. You're watching theCUBE, the leading source for enterprise tech news.