TheCUBE Research’s Savannah Peterson talks with Google Cloud leaders Belinda Runkle, senior director of engineering, serverless, Google Cloud, at Google, and Lisa Shen, product manager at Google Cloud, about how serverless computing and agentic AI are reshaping the technology landscape. Part of the Google Cloud: Passport to Containers series, the interview blends career journeys with practical insights on innovation at scale.
Runkle and Shen reflect on their paths into Google Cloud and how their work is driving the evolution of serverless platforms. They unpack the capabilities of Cloud Run, explaining how it enables developers to run containers and code without the complexity of managing infrastructure — while meeting the growing demands of AI workloads.
From accelerating academic research to streamlining auto insurance claims and rethinking financial advising, agentic AI is delivering measurable impact. Runkle and Shen share real-world examples of these advancements, highlighting how Cloud Run provides the agility and scalability that developers and businesses need to build the next generation of intelligent applications.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
Google Cloud: Passport to Containers. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open this link to automatically sign into the site.
Register For Google Cloud: Passport to Containers
Please fill out the information below. You will recieve an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for Google Cloud: Passport to Containers.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
Google Cloud: Passport to Containers. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open this link to automatically sign into the site.
Sign in to gain access to Google Cloud: Passport to Containers
Please sign in with LinkedIn to continue to Google Cloud: Passport to Containers. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Cloud Run & Agents: Fast Starts
TheCUBE Research’s Savannah Peterson talks with Google Cloud leaders Belinda Runkle, senior director of engineering, serverless, Google Cloud, at Google, and Lisa Shen, product manager at Google Cloud, about how serverless computing and agentic AI are reshaping the technology landscape. Part of the Google Cloud: Passport to Containers series, the interview blends career journeys with practical insights on innovation at scale.
Runkle and Shen reflect on their paths into Google Cloud and how their work is driving the evolution of serverless platforms. They unpack the capabilities of Cloud Run, explaining how it enables developers to run containers and code without the complexity of managing infrastructure — while meeting the growing demands of AI workloads.
From accelerating academic research to streamlining auto insurance claims and rethinking financial advising, agentic AI is delivering measurable impact. Runkle and Shen share real-world examples of these advancements, highlighting how Cloud Run provides the agility and scalability that developers and businesses need to build the next generation of intelligent applications.
play_circle_outlineAgentic AI: Defining Intelligent Agents and Their Role in the Future of Software Development
replyShare Clip
play_circle_outlineDiscussion on the edge computing and its impact on user experiences.
replyShare Clip
play_circle_outlineMaximizing Development Efficiency: How Cloud Run Transforms Serverless Application Deployment and Simplifies Infrastructure Management for Developers
replyShare Clip
play_circle_outlineOverview of serverless GPU capabilities on Cloud Run for AI workloads.
replyShare Clip
play_circle_outlineDifferentiation between AI agents and Cloud Run’s core functionalities for scalability.
replyShare Clip
play_circle_outlineThe impact of emerging technologies on democratizing software development processes for all.
replyShare Clip
play_circle_outlineThe emerging paradigm of "wipe coding" for frictionless software development experiences.
Senior Director Of Engineering, ServerlessGoogle Cloud
TheCUBE Research’s Savannah Peterson talks with Google Cloud leaders Belinda Runkle, senior director of engineering, serverless, Google Cloud, at Google, and Lisa Shen, product manager at Google Cloud, about how serverless computing and agentic AI are reshaping the technology landscape. Part of the Google Cloud: Passport to Containers series, the interview blends career journeys with practical insights on innovation at scale.
Runkle and Shen reflect on their paths into Google Cloud and how their work is driving the evolution of serverless platforms. Th...Read more
>> Hello, Nerd Fam. Welcome back to our fantastic exclusive series with Google Cloud: Passport to Containers. My name's Savannah Peterson, here in our Palo Alto Studios for theCUBE. Very excited to be bringing to you two more guests. Today in episode nine of the series, we're going to be talking about cloud run, agentic, serverless, and all the things you need to be agile in this incredibly rapid era of innovation. Without further ado, please welcome Belinda and Lisa. Thank you so much for being here.
Belinda Runkle
>> Thank you so much for having us.
Savannah Peterson
>> I'm going to try not to break the desk while we're hanging out. This is going to be super fun. Bobby, as usual, hyped you up as two fantastic voices. But before we dig in to some of the product story, I want to ask you both one of our favorite questions for the series, and that is when did you fall in love with tech? Lisa, I'll start with you.
Lisa Shen
>> Yeah, sure. I think I just liked the computer science, computer engineering in general when I was out of my high school. It's just so-
Savannah Peterson
>> So already as a high schooler.
Lisa Shen
>> Yeah. Yeah.
Savannah Peterson
>> And look at you now. So did your path follow the trajectory you thought it would or were there some surprises along the way?
Lisa Shen
>> I've been in the tech industry for my whole life, so definitely have seen through a few tech shifts, but nothing that's bigger than what we are seeing right now. The age of the AI agents, where we're basically giving the code a brain, and who knows, tomorrow the code is going to make a coffee for itself.
Savannah Peterson
>> Yes. Ooh, I love that analogy. Giving code a brain. I haven't heard that. We're going to come back to that when we dig in a little bit, but that was great, Lisa. Belinda, what about you? When did you fall in love with tech?
Belinda Runkle
>> So my entry to tech is a little more anomalous. So I grew up in eastern Kentucky, so there was no coding high school class, or anything like that. I actually went to college, and studied studio art and anthropology before I went into any sort of tech. But in that time, I was one of the first people in my class that had a computer. My dad had a computer, my brother had a computer. When I graduated, I wanted my own computer. I spent eons of time looking at... For like all of the parts and pieces that I want.
Savannah Peterson
>> So you built it.
Belinda Runkle
>> This was like 386 era, so this is a while.
Savannah Peterson
>> Oh, yeah.
Belinda Runkle
>> I've been around for a bit. Okay? But, I made a transition actually through tech support. So, I started in tech support, actually moved into an IT role, learned to program a little bit, and actually got into startups. So, it was by joining sort of a pretty technical startup that did electronic discovery, and working there in the trenches for a while, that I kind of had to pick up a lot of coding, programming, database skills, and then got into management, and went from that startup to other startups. And then eventually, like last 10 years, found myself in cloud startups before coming to Google. So, my background is pretty different from most Googlers. There's definitely some other folks that I've run into during my time at Google, that's kind of made me appreciate honestly, that I've got that non-traditional background.
Savannah Peterson
>> I think the .
Belinda Runkle
>> Kelsey Hightower, I don't know, you've probably talked with Kelsey a few times.
Savannah Peterson
>> Yeah. Yeah, yeah. I had interviewed Kelsey, yeah.
Belinda Runkle
>> He's definitely one. Tim Hawkins also, who's very well-known in the Kubernetes space. Same thing. So, folks that kind of came out these very different backgrounds, and ended up at a place like Google.
Savannah Peterson
>> Yeah.
Belinda Runkle
>> Yeah. Well, I have a different way of thinking about things.
Savannah Peterson
>> That diversity of thought I think is really important, especially when we're building the future. We were just talking about how important it is to bring all types of creators on this rocket ship that we're currently in the process of launching. I think what you just touched on is really imperative. I got to ask you a follow up there. Do you find that you use your anthropology and art skills in your current role?
Belinda Runkle
>> Oh, absolutely. Absolutely.
Savannah Peterson
>> I feel like you would, right? Yeah.
Belinda Runkle
>> I mean so much of like, particularly as a manager of people and organizations, so much of what you're dealing with are just the fundamentals of anthropology. It's like culture, and values, and habits, and even it's like a way of looking at the social construct of just how to get things done, why people do what they do, what they value, what behaviors are rewarded versus punished. So a lot of it's very portable.
Savannah Peterson
>> Yeah.
Belinda Runkle
>> And it's also, it allows me to step back, and not get so over-involved in what's happening. You can step back and look at it a little bit from the sociologist, or anthropologist's eye, and sort of think about it in a fundamental sort of way.
Savannah Peterson
>> Wow. I feel like low-key, you could be teaching us all workshops on how to do that, and have better balance between emotions.
Belinda Runkle
>> There's actually tons of folks. I mean, so for instance, if you look at folks coming out of user experience, a lot of what they're bringing are the things that have been sort of evolved from ethnographic studies. We bring them into UX and design, and now they're just sort of fundamentals to think about design of product. So, I think that's kind of already true, in that there's sort of a community of practice in product development that already appreciates that. I think that's less true in engineering. Engineering folks don't generally grow up seeing the world in those paradigms.
Savannah Peterson
>> Yeah, I think that's an interesting observation in marketing and TV. I don't know, I can ask the production team what terms they see things in, might get the conversation even spicier than we're anticipating today. I'll keep us on track here. So, Lisa, you said something, again, I'm going to definitely steal this and give you credit for it, but the brain to the code, when it comes to agentic. I'd love to actually hear from both of you, not a bigger hyped term right now than agentic AI, and AI agents. How do you define agent? Lisa, I'll start with you, since you kicked that off.
Lisa Shen
>> Sure. Yeah. AI agent, basically you have AI that normally in the past that does everything, but AI agents are those... It's basically a software program that specifically focus on a very specialized task. So, usually you use the large language model, and then what distinguish actually what the AI agent versus a traditional, let's say, chatbot services, is actually the use of a effective use of tools.
Savannah Peterson
>> Yeah.
Lisa Shen
>> So the tools refers to things like external systems, external databases. So, basically the large language models are the brain of an agent, and then the tools is really like the agent's hands reaching out to the digital world. And then so you have agent plus tools, and then actually the large language models plus tools, then actually made up an AI agent to help you accomplish a specific task.
Savannah Peterson
>> Yeah. That gave us all a really nice little visual. I can see the little jazz hands of the tasks that the agents are running around doing. I like that a lot. Is there anything that you would add, Belinda?
Belinda Runkle
>> In tech, of course, we love to give names to things, so that we call an agent-
Savannah Peterson
>> And acronyms.
Belinda Runkle
>> Yes, names and acronyms, all over it. So, the agent is really like an evolution of software. So it's a pattern. And so if you think about what is an agent, it's a piece of software that's able to leverage a model, as Lisa was saying, that can observe its environment, whatever its environment it's been trained to observe, and it can take action. And the way that it can take action is often through tools, as MCP is where you start to see the integration of different tools, APIs as a way it can take action. But it's really about being able to observe its environment and context, reason about it, and then make decisions and do stuff. And I think when we think about agents, like a lot of the sort of magic is really about how does it reason about? Like what is the context that it has access to, and then what kinds of actions is it able to take? And that's also where you see the evolution of multi-agent systems. So it's how do you string together each of these entities that has the ability to reason about some set of data, and take independent action, which might be sort of tailored from thing to thing. Well, we'll get into some, I think examples of that, as we continue to talk.
Savannah Peterson
>> That's where I'm us next.
Belinda Runkle
>> Yeah. Yeah.
Savannah Peterson
>> Actually, why don't we stay there, and I'll let you both answer this as well. What are some of the use cases you're seeing or that you think are best practices for agentic in general? And then also, chiming in on why Cloud Run is so great for our agentic future. We'll say with you, Belinda, just to start us off.
Belinda Runkle
>> Yeah. So I think there's a lot of common ones that we're all sort of aware of. So we've seen the chatbots, chatbot is pretty prolific, as sort of a garden variety kind of agent. Of course, there's a lot of advancements that can be made on top of the chatbot, as we're seeing today. Even if you look at things like autonomous vehicles, self-driving cars, those are systems that have been built around that same agentic thinking. You can take it all the way to the more cutting-edge example. So, you've probably seen the Boston Dynamics Spot dogs, and those are pretty wild. Lots of different variations of that. But, same thing, right? They're able to use sound, video, take in their environment, reason about things, use tools, but it's that... Again, it's that same sort of pattern of thinking and reasoning is what you're going to see across any sort of agentic system. And the more agents evolve, I think the more we're seeing the sophistication, in terms of multiple agents being able to collaborate with one another, with sort of specializations.
Savannah Peterson
>> Yeah.
Lisa Shen
>> Yeah, I can definitely add to that. So, there are so many AI agent use cases now being active, adopted in the field right now in each of those verticals, right? I talked about the code that's going to be able to make a coffee for itself. That might be a use cases for tomorrow, but there are plenty of other use cases that are actually actively being used today. So I'll give you an example. For example, in the academic research area, let's say we have an AI agent for a research assistant. So what this research assistant does is that it will take a foundational paper. Foundational paper usually are those paper that has big ideas, right? It will analyze the idea and understand what this paper is about and then it's going to go ahead to reach out, use this and reach out to the digital world, whether it's via the Google search or some other additional scientific search tools to find out additional paper that have cited this foundational paper. And then it actually uses advanced analytical skill to try to understand, hey, all these ideas are derived from this foundational paper. What is amazing to me is that then this AI agent is going to propose the future research directions based on all the synthesized information. So imagine what this can do to the research scientists, accelerate the whole research discovery process, and then you just basically make the breakthrough much easier. Right?
Savannah Peterson
>> Yeah.
Lisa Shen
>> So if this is abstract enough, but we also have AI agents that touches upon our day-to-day life. For example, the auto insurance agent, that is very obvious. You can have auto insurance AI agent that helps you to fill out a new membership registration, the filing the claim, helps you with, let's say, getting the roadside assistance, getting that tow truck to you as soon as possible. So I'm sure that type of AI agent is already used today. And one of my favorite one in yet a different vertical is in the financial industry area. So, financial advisor AI agent, can you imagine what it does?
Savannah Peterson
>> It's so funny that I... Well, we could be giving you a lot of portfolio advice. It's obviously not a financial advisor if it's a digital piece of code, but I'm excited to hear what you'll say next because I was literally just thinking about this. I was like, "I need to be a little bit savvier investor. I bet I can use these tools to help organize my learning based on what I already feel confident in versus my gaps and whatnot." So yeah, so please tell us, Lisa.
Lisa Shen
>> Yeah, so you can have a financial advisor agent that consists of, let's say, multiple sub-agents. So let's say you have a stock symbol that you're interested to buy. So it has this data analysis sub-agent that actually look into... Take the stock symbol, and it's going to look into the... Pull out the SEC data and go to the, let's say Google or some other website to get all the information, recent publications, recent news related to this stock and actually figure out what is the signal noise, and then actually come up with a data analysis report.
Savannah Peterson
>> That's .
Lisa Shen
>> I know. And this report then gets fed into the next sub-agent, which is let's say trading strategy, trading strategist agent, and where it's going to come up with a number of trading strategies based on the data research from the previous agent. And then you have the execution sub-agent, which will then take this trading strategist agent output to say, "Hey, based on your investment horizon or timeframe based on your risk assessment." And then it's actually recommend whether to buy the stock or short the stock or get some option out of it. And a financial advisor agent is not going to be complete without a risk analysis. So there's a risk analysis agent that actually then take all of above and then I come up with a risk assessment and then say, "Hey, when you do take this action, here is the risk that you might be taking." Right?
Savannah Peterson
>> Yeah.
Lisa Shen
>> So look, this thing just completely changed the way how we live, how we work, and how we play.
Savannah Peterson
>> So do you think Wall Street's nervous or excited about this?
Lisa Shen
>> I think every vertical you're going to see this kind of... It's going to change how we...
Savannah Peterson
>> It's going to take out all the boring stuff that took too long anyway.
Lisa Shen
>> Yeah.
Savannah Peterson
>> It's going to reduce toil. That's for absolute sure. Those were great examples, both of you. Both of you mentioned a lot of examples that are edge use cases. Do you think that we'll see agentic adoption happen faster at the edge, or we'll see more human benefit at the edge? Any thoughts on that?
Lisa Shen
>> The edge? What do you mean by edge?
Savannah Peterson
>> Well, we're talking about devices or researchers doing things on their PC. I'm talking about where the physical action or the physical integration is with that. You might be on your cell phone talking to your risk analyst, or your financial advisor. So I'm just curious, do you think that most of the human interaction will happen on the edge? Are we still going to have a lot of agentic activity that's happening on prem? I don't know, just curious.
Belinda Runkle
>> Yeah, I mean I think we're already seeing it at the edge. So if you think of mobile devices as an edge, people are already embedding agents inside of phones, inside of applications. And so you can imagine, for instance, we've got a lot of customers, for instance, that are working on their shopping experience. How do we enable customers to have a better time, whether they're in a store, whether they're on their way to a store, whether they're shopping online, all of those experiences can be sort of enabled through better leveraging of agents. So for instance, traditional experience. I went to my local big box hardware store, I needed to buy a device. All the devices are in complete lockdown, and the way that it works right now is I have to go and use my phone to summon someone to come and unlock the thing, even though I already knew on my way to that store what I wanted. So, if you imagine an agentic experience that's much more evolved would have a set of agents that's... Like already knows what I'm coming into the store for, already prepared to pull that thing off the shelf, have it ready for me when I get there, and it's not a three-hour wait or a 24-hour wait
Savannah Peterson
>> Yeah. And it's teed up with your loyalty points-
Belinda Runkle
>> Exactly....
Savannah Peterson
>> when you've got that $5 coupon off for your new drill, or whatever it is.
Belinda Runkle
>> Exactly. Probably will suggest a bunch of other stuff that I need to buy with that chainsaw or whatever it is that I was going in to buy.
Savannah Peterson
>> Of course. Or maybe some projects that recent customers just built with their new chainsaw or whatever that might be.
Belinda Runkle
>> Yeah. Yeah, yeah. Exactly. Exactly. So I feel like we've barely scratched the surface. So I think where we see agentic use cases kind of really picking up, it's like people are trying to solve a complex problem that they already have to solve today, which is great. I mean, that's a fantastic way to start.
Savannah Peterson
>> Yeah.
Belinda Runkle
>> I don't think we've yet seen the sort of upside of how can we actually enrich the human experiences that we're already going through today? I think that's a buffet of agentic use cases that we're going to see evolve.
Savannah Peterson
>> Absolutely. I think Jonathan Ross, one of the... used to be on your team and helped to build the TPU, runs Groq now, one of my favorite people to interview. I love it. He always says, "We're going to solve problems we don't even know how to think about yet."
Belinda Runkle
>> Yep.
Savannah Peterson
>> And it's such an interesting thing. We've got this now, now we have this Z axis. It is like when we got computers and we could communicate, we did what we'd have always done, send letters, and we called it email. We're kind of in that phase right now where we're doing things we've always done, and we're calling it agentic, we're calling it AI. Obviously there's a layer of efficiency and effectiveness that comes along with that, that's very powerful, and has great impact. But on the flip side, there's so much to come, and I think it's going to be really awesome. So speaking of that, why is Cloud Run such a wonderful tool for this? I'll start with you Belinda-
Belinda Runkle
>> Yeah....
Savannah Peterson
>> since you've given me the nod.
Belinda Runkle
>> So what's Cloud Run? So Cloud Run is a serverless container runtime. We run functions, we run containers, we can run from source code, and it's very lightweight, no management, fast scale up, scales to zero, pay as you go, and really has been kind of the evolution of how to get sort of ready-made compute in a very low toil sort of way for developers. So we very much have targeted developers, particularly for API services, websites, things like that. So where agents has proven to kind of be a really good fit, and we're actually getting great adoption both from customers, but also inside of Google, and it's because a lot of the same capabilities are needed by agents. So if you imagine in those kind of workflow systems, where you have multiple agents, typically what's happening is something needs to kind of be spun off, like I'm going to assign a task, a thinking task, a context task, going and taking action, integrating with an MCP server, an API, and each of those is going to basically spin up a little unit of work that needs to go get done. And so that ability to do fast fan out, fast scale up, scale to zero when it's done, those things are already built into Cloud Run. And then the payment model, that pay as you go, is really efficient. We've also got a lot of great fundamentals that are already in place, which our enterprise users and customers love. So, strong security boundaries, it's already a strong isolation, multi-tenancy. So, the sort of hardcore security is already taken care of. And so, for the people that are using Cloud Run in those use cases, they can really focus on the sort of business logic of building their agentic system, and not worry so much about how do I manage the compute, and infrastructure layer. So, we're basically solving that problem for those agentic developers, so they can focus on the agentic aspects of the problem that they're solving.
Savannah Peterson
>> Yeah, they can go do the thing, it's like those fancy meal food boxes that show up, and everything's set for you to have the right amount of protein, have everything, and then you can just go do whatever creative thing that you're trying to do. It's that kind of plug and play. So, I mean, every time we have these conversations with your team about Cloud Run, it makes so much sense, and the agility and the flexibility, particularly as the needs of the different enterprises and companies ebb and flow throughout both their given day of workloads, as well as just whatever they might be building right now. What are some of the best use cases or most common use cases of how your community is using serverless GPUs? Lisa, I'll start with you.
Lisa Shen
>> Oh, the serverless GPU? Yeah. So serverless GPU is basically... We actually recently, I think a few months ago, it's becoming generally available on Cloud Run. So not only it's a great place to serve, like Belinda said, to serving and the orchestration... Performing the serving and the orchestration for the AI agents, with the serverless GPU, you actually can now actually host the lightweight open models. So basically you can host the Gemma stream model, DeepSeek r1:7b model. So we support NVIDIA L4 GPU today, and then that's actually great for you to do some cost-effective, low latency AI inference type of use cases. So, there are plenty of benefits for using the serverless GPU, right? Like Belinda mentioned a lot of those benefits already. It is on demand, and this is also super important for the GPU case, because a lot of times you actually have to go reserve, but with the serverless GPU it's on demand, and it scaled down to zero. Obviously, that's a huge advantage when you actually have a sporadic traffic pattern, and it saves you the cost.
Savannah Peterson
>> And most people do.
Lisa Shen
>> Yeah. Most people do.
Savannah Peterson
>> Especially right now.
Lisa Shen
>> Yeah.
Savannah Peterson
>> Yeah.
Lisa Shen
>> Yeah. And then also with the current GPU, you get to actually one of two key benefits that I like most is it really has really fast scale. So, a faster start time. What does it mean, is that, for example, literally from zero to one means that you actually, from zero, you then spin up the container instance, get the driver pre-loaded, it only literally takes five seconds and then takes some additional, maybe 10, 20, 30 seconds to load your model in a framework. The reason this is a-
Savannah Peterson
>> It's pretty quick.
Lisa Shen
>> It is, because this compares to, usually it takes like five, 10 minutes to actually get a model loaded.
Savannah Peterson
>> Yeah. You're like getting up and going, getting a cup of coffee.
Lisa Shen
>> Yeah.
Savannah Peterson
>> .
Lisa Shen
>> I know. Yeah. Why this is important? Because this actually, then imagine you are a developer, you need to iterate upon .
Savannah Peterson
>> And the power you're using.
Lisa Shen
>> Exactly. And it's fast iteration that you get out of this, right? So it gives you the great developer velocity to a large degree for that as well.
Savannah Peterson
>> Well, and who doesn't want that?
Lisa Shen
>> I know. I know.
Savannah Peterson
>> Always, but especially right now, such a unique landscape.
Lisa Shen
>> Yeah.
Belinda Runkle
>> And I would say for our watchers, maybe we should probably distinguish a little bit between when do we use Cloud Run GPUs, and how is that relevant to an agentic system?
Savannah Peterson
>> Yeah.
Belinda Runkle
>> So, an agent, for example, doesn't necessarily need a GPU.
Savannah Peterson
>> Correct.
Lisa Shen
>> Yeah.
Belinda Runkle
>> Where the GPU is used is when we're going to run some sort of inference.
Lisa Shen
>> Absolutely.
Belinda Runkle
>> So for instance, if that agent needs to basically reach... Attach to a model, if it needs to generate some creative set of cat pictures from the prompts that were generated-
Savannah Peterson
>> Yeah, inference is all inference is all the stuff that makes the AI experience feel real.
Lisa Shen
>> Yeah.
Belinda Runkle
>> Yeah, exactly.
Savannah Peterson
>> And magical.
Belinda Runkle
>> Exactly.
Savannah Peterson
>> Yeah.
Belinda Runkle
>> Yeah, those magical bits where they're CPU intensive, memory intensive, need a lot of compute power, and are reaching to the model, that's where the GPUs fit into place.
Lisa Shen
>> That's actually such a good point, Belinda. So, let me just give you a few quick examples of how customer use in the serverless GPU, right? And then for example, we have a customer, it's a global phone maker, and then they actually use the GPU in their mobile app. For example, there's an app called a Magic Eraser. It actually allows you to actually edit your phone. So the photo, edit your photo image, and then let's say you want to remove some unwanted thing from your photo, you'll be able to actually... It's actually in the back end, it calls the serverless GPU to do that, right? And we have-
Savannah Peterson
>> I always wondered how that worked.
Lisa Shen
>> Yeah.
Savannah Peterson
>> I've seen people use the Magic... Oh, that's cool.
Lisa Shen
>> That's how it works. The serverless GPU in the back end. And then we have another customer who actually use the serverless GPU to host a embedding model. So, basically let's say we have this customer called a Wayfair. It's a furnishing company. They have a vast amount of furnishing catalog, and then you want to turn them into a vector in the vector database. So, vector database allow you to do some semantic search. So basically it's fast and it's also, you'll be able to search based on the context, not only on the word itself. So, then they actually use the serverless GPU to host the embedding model, turn the catalog into a vector and then store them in the vector database. So, that's another use of the serverless GPU. It's a little bit than the AI agents itself, because this is only one part of the AI agents, the use of the AI model part, right? AI agents is more than that. We talked about AI agents has tools, AI models.
Savannah Peterson
>> Yeah.
Lisa Shen
>> So yeah, it's great call out by Belinda.
Savannah Peterson
>> Yes. I love that differentiation. I think that's important. Belinda, how do y'all drink your own champagne when it comes to using Cloud Run inside Google?
Belinda Runkle
>> Oh, so there's some history here. So, it is probably not well known, but Cloud Run is actually the evolution of the infrastructure that was started by Google App Engine. And so, Google App Engine was actually sort of the pre-cloud cloud that Google took to market. And if you look at the genesis of App Engine, it used to have a code name internally, it was Project Prometheus. And they called it Prometheus because the idea was that they were going to bring fire to the humans, and that fire was actually Borg, because inside of all of Google is Borg as sort of the compute infrastructure, sort of the predecessor of Kubernetes.
And so that bringing of fire, the infrastructure side of that, which is things like our data planes, some of our autoscaler, a lot of those pieces are actually still present today, and have been evolved, and are essentially what's under the hood for Cloud Run. So when we look at what we're doing with Cloud Run, inside of Google, it is actually the part of Google Cloud that is the most like Borg, right? And so, since Googlers are already used to using Borg, again, great developer experience, super low toil, don't have to think about infrastructure management, has been very attractive inside of Google. So, we're seeing a lot of adoption, everything from internal IT workloads, there's actually a project called Corp Run, right? Which is sort of a wrapper around Cloud Run and it Enables things like HR apps, IT apps, things like that. And then actually we have a lot of products that are also essentially using Cloud Run as the underlying compute infrastructure. Actually, a number of the AI products that you've probably done other sessions with use Cloud Run in that capacity, some of the API products as well. And then we've been collaborating a lot with the DeepMind folks, just as we talked about, given everything we described around agents and the sort of compute capabilities and the fast startup and cost-effectiveness, we're getting a lot of traction, particularly around AI workloads that want to take advantage of Cloud Run. And it makes a lot of sense. I mean, there's no reason to kind of go and have to build multiple sort of competing compute infrastructures when we've got something so great, and just battle tested, like Cloud Run that's able to solve that for folks. And I think for us, the benefit is, as much as we have an incredible enterprise adoption with public sector, and a lot of financial services, which is wonderful to see, our biggest customer is still Google, and Google customers, they will come to you with their feature requests, their complaints, but they'll also provide pull requests, and actually help us get things out the door more quickly. And so, it creates a pretty dynamic environment to be able to build cutting edge solutions for Googlers as they're bringing stuff to market, and then be able to turn around and provide those same benefits to our outside customers.
Savannah Peterson
>> It's a pretty sweet feedback loop too-
Belinda Runkle
>> Yeah, yeah....
Savannah Peterson
>> with that internal and external community. It allows you to iterate in a way that really serves the future, which is what everyone's trying to throw a dart at right now. I mean always, but particularly right now. What's one thing you wish more people knew about Cloud Run? Lisa, I'll start with you.
Lisa Shen
>> Sure, yeah, I just want everybody knows that Cloud Run is such a great platform not only for the developers, but also for the enterprise. So Cloud Run itself, a lot of times people mention Cloud Run, they think, "Oh, this is great for prototyping, great for experimentation." But Cloud Run itself is such a great enterprise grade platform that it has a robust set of security compliance and networking features, whether you want to do, for example, integrate with Identity-Aware Proxy allows you to do the user authentication, authorization. We actually have secure software supply chain to support our enterprise customers, for example, only authorized image gets deployed to the production. We are binary authorization, things like that. Those features are super, super important to enterprise customers.
Savannah Peterson
>> And super expensive if you're building each one of those things out yourself internally.
Lisa Shen
>> Absolutely.
Savannah Peterson
>> Particularly if you're trying to show something at scale. I mean, each one of those tasks require their own teams to be compliant-
Lisa Shen
>> Yeah, absolutely....
Savannah Peterson
>> to have the privacy and security that you need. I feel like nerds are kind of having a moment right now, and I feel like some of the traditionally less sexy aspects of our jobs are also having a moment right now, because if we don't put the right bumpers up, we can't score the strike on the bowling lane, and do awesome things. I think that's so important. What about you, Belinda? What do you wish everyone knew?
Belinda Runkle
>> So, when I meet with customers, I go to conferences, there's really sort of two big myth-busting areas that I always run into and it just cracks me every time. So the first one is the notion of serverless, which has become kind of a little bit of an overloaded, confused word. But for many folks when they think serverless, they think functions, right? Because Lambda really sort of defined it with their functions offering. And so when they think Cloud Run, they assume it's a functions offering. Now, we do have functions in Cloud Run. We can accept functions as source code, but it is really a full-featured compute platform. And while it's serverless in nature, it is not at all defined to functions. Actually most of what customers bring to us are containers. So it's really more of a container orchestration that has a lot of the same simplicity and magic of what you would think of when you think of serverless. T. He other myth-buster is people assume that it's Kubernetes under the hood, and that's a fun one to sort of detangle. And some of that's self-inflicted, because we actually have a Kubernetes API. So we built Cloud Run with the notion that you would have Knative compatibility, so that when you build a workload and you wanted portability between Kubernetes and Cloud Run, you would get that by virtue of Knative. But folks have read that to mean, "Oh, it must be Kubernetes inside." Right? Oh, .
Savannah Peterson
>> I've had to dispel that myth actually a few times, especially as a part of doing this series. It's like, "No, no, no. No."
Lisa Shen
>> Yeah.
Savannah Peterson
>> Two separate entities.
Lisa Shen
>> Yeah.
Belinda Runkle
>> Yeah.
Savannah Peterson
>> Two different engines. Two different tools.
Lisa Shen
>> Absolutely. And I just want also-
Savannah Peterson
>> .
Lisa Shen
>> Yeah, add one more thing to what Belinda said. Actually, since we talked a lot about AI agents, I feel like I have to say that Cloud Run is extra supports a diverse type of workload. Agentic AI app, AI agents, we talked about serverless GPU, but AI agents, Cloud Run is such an ideal platform for AI agents. For example, we talked about the different components of AI agent, I talked about a large language model as the brain, and the tools is more like agent's hands reaching out into the digital world. But then you also have the, let's say, the orchestration layer that pieces everything together. It knows when to call what tools, or when to call the large language model. It knows the goal of this AI agent. And then so for this to host all this orchestration layer, you need a runtime. The runtime is literally where you bring the... It's the engine where you bring the AI agent into life. And this is where the Cloud Run plays really a key role. And also, for example, for the large language model, needs to talk to the tools. And then we have this... Anthropic came up with this MCP, Model Context Protocol. I'm sure you heard about it. And this actually basically allows... It gives a standard way for a large language model to talk to all these different type of services out there. So it's like, it's more like a USB-C, you have the laptop and then connecting to all the different peripherals via the USB-C. And by the way, actually, this is more of a client and server architecture, and then Cloud Run is perfect to host MCP server. So MCP server is usually, it's just a thin layer of function that wrapped around your existing API services. So, the client, which are the AI agents, can call those servers. They don't have to worry about any of the API changes. So, for example, you have the weather server, can call the Weather.com API, then we'll be able to actually... Or you can call the Yahoo stock website. All you need to do is just use the MCP protocol and you don't have to worry about all the underlying APIs. So, Cloud Run-
Savannah Peterson
>> That's pretty nice....
Lisa Shen
>> plays a very key role there.
Savannah Peterson
>> Yeah, we're getting to this much more simplified, unified experience when it comes to interoperability that's hopefully going to be a little bit less messy and clunky as some of our last technological era shifts. All right. Final question for you both. What do you hope that our current technological revolution, it doesn't have to be agentic or AI or even containers, just what do you hope that the stuff that we're working on now does for the people that you love outside of our little technology bubble? Belinda, I'll start with you.
Belinda Runkle
>> So I love where AI is headed. Even in the last six months, some of the things that we've seen evolve have really surprised me. So I'll give an example. AI Studio, and the stuff that it's launched, and seeing the integrations that we've done with Cloud Run, we can now see non-programmers basically go and describe an idea that they have, see it implemented in code, and get deployed to the cloud. And while certainly having some programming skills and understanding what's happening under the hood can help, if they want to edit it and make changes, a lot of people don't have that. So there's a lot of folks who have limited, or no technical background that are now developers, which, that's kind of mind-blowing. I mean, you think about 20 years ago, there was no equivalent. So, I think seeing this kind of democratization of the notion of what it means to be a developer, and how to build stuff that people can use, and find valuable, and have it be in the cloud, it's really kind of changing a lot of our assumptions about what we think of when we talk about what is a developer. When we thought about who is a developer with Cloud Run, we never had a specific kind of developer in mind, like a Python, Node, Java, whatever. It was always meant to be we want to be most welcoming and inclusive of the kinds of workloads, independent of the type of workload or type of code or type of framework. And now if you think about where we are today, that even the notion of code, and languages, and all that is really being abstracted away. So, for me, the evolution of that becomes much more and more how can we make that which we call cloud, much more self-driving, much more prescriptive, not just adaptive, but actually be able to kind of anticipate what it is that people are going to need to manifest the ideas that they have? Going back to that art and anthropology background, the notion that building software becomes more of a creative practice than an engineering practice is sort of mind-blowing, and I think it's awesome, and I can't wait for it to be the case.
Savannah Peterson
>> That's exciting.
Belinda Runkle
>> Yeah.
Savannah Peterson
>> It really does. And way to bring us full circle there, Belinda. Love, love, love that circle back to the anthropology. Lisa, what about you? What do you hope this era brings into your world?
Lisa Shen
>> Yeah, let me actually start with this wipe coding, right? So I'm sure you heard about wipe coding, right?
Savannah Peterson
>> Oh, yeah.
Lisa Shen
>> So this is really a emerging paradigm for all the software development. So wipe coding is all about how you turn the ideas into code with minimum friction. And then Cloud Run seems, to me, is a natural fit, and the next step, where then you turn your source code into a live, functional app, with no time, and with minimum infrastructure to manage. And then wipe coding is also all about this instantaneous feedback loop, and then Cloud Run basically gives you all these faster iterations, and then you can get to do the experimentations, prototypes. And whenever you're ready to push it into production, you'll be able to actually, Cloud Run scales up for you instantaneously. But the future innovation, I do think it's pushing this paradigm even further. The wipe is broken if you have to worry about when you turn your code into production, you have to worry about how do I configure this concurrency settings in my container instance? Or how do I have to... Do I have to worry about the security vulnerability patches or OS updates? Right?
Savannah Peterson
>> It's a lot to think about.
Lisa Shen
>> It's a lot to think about. Right?
Savannah Peterson
>> Yeah.
Lisa Shen
>> So the wipe is broken. So the goal is to make the serverless infrastructure be so adaptive and intelligent enough, it becomes totally invisible to developers. And then, so you get to stay in this wipe, where you have the concept, take the concept to prompt, to code, to deploy it all the way to production in a single frictionless workflow. And the reason I mentioned that, how does that affect us? Because we are all going to be the wipe coders, or we're already wipe coders. We're part of this wipe coding community, because AI is democratizing whole software programming. Right?
Savannah Peterson
>> Yeah.
Lisa Shen
>> So we actually then get to enjoy really the... Focus on the joy of creative building. And then this is also when I think the code not only has a brain, but it's also going to make a coffee for itself. That time is coming.
Savannah Peterson
>> Also bringing it full circle. Well done, Lisa. Yeah, I think it's really, it's about vibe creating, right? That we were blocked by not being able to speak the same language, whether that be NLP, or code, or anything, quite frankly, from a truly universal collaboration experience. And I agree with both of you. The future is bright. It's going to be great. Thank you both so much for taking the time today. This has been absolutely fantastic.
Belinda Runkle
>> Thanks so much. It's been a pleasure.
Lisa Shen
>> Yeah, thank you.
Savannah Peterson
>> Yes, and I hope you've all had as much fun as we've had here on episode nine of our exclusive series with Google Cloud: Passport to Containers. My name is Savannah Peterson. You're watching theCUBE, the leading source for enterprise tech news.
