Google’s Agentic Data Cloud extends AI data management

Beth Pariseau: From Informa TechTarget, I'm Beth Pariseau, and this is IT Ops Query.

This podcast distills the signal from the noise about enterprise software development and platform engineering. Each week, we'll talk to expert guests about the latest tech industry news and trends that engineering and IT leaders need to know.

Don't forget to subscribe to IT Ops Query for more conversations on AI and the future of the enterprise digital workspace.

Greetings from fabulous Las Vegas. I'm here with Andi Gutmans, who is VP and GM of Data Cloud at Google.

We are here at Google Cloud Next, where Google has previewed some new features with its Agentic Data Cloud, including a new Knowledge Catalog to ground agents in business context, Data Agent Kit, and a cross-cloud data lakehouse that provides zero-copy access to data in Apache Iceberg formats in multiple applications, operating systems and AI systems, including AWS and Azure.

Thank you very much for taking time out.

Andi Gutmans: Yeah, no, thank you so much for having me.

Pariseau: So, I always want to know what the takeaway is of these updates for IT buyers and enterprise IT leaders. What will they be able to do with this Agentic Data Cloud that they couldn't?

Gutmans: Yeah, no, that's a great question. And, you know, to take a step back, we announced Agentic Data Cloud today, and really the premise here is we've gotten to the point now where agents are not just assisting, but they're really completely changing how we get work done and are able to be proactive on the business's behalf.

So, what we've been talking about is we're actually moving from this system of intelligence, which was very much, you know, reporting the news, reporting the past, maybe doing a bit of forecasting, to what we call the system of action -- meaning we have agents now that can reason about the business and that can actually also take action on our behalf.

And our view is that that requires a fundamentally different data platform. And so, what we've done -- and, you know, as part of these announcements play into that -- is basically retool and rebuild our data cloud in a way that is ready for the system of action era, which you really think about kind of three key changes.

One is we're moving from human scale to agent scale. And so, how do we think about a data platform that kind of goes and is able to scale that way, not just operationally, but also you're gonna have millions of agents, right? How do you think about that, and how do you govern it and how do you enable it?

The second one is we are moving from these reactive agents. Think about the chatbots. It's kind of, you ask a question, you get an answer. Why do you even have to ask the question? Right? You should wake up in the morning, and it should tell you, you know, based on your priorities, what are the things you should be thinking about or focusing on today. So, we think about this as becoming proactive.

Then lastly, moving from focusing on data to knowledge because the thing you really need to expose to these agents is real strong context. And that means not just data access, but really understanding the semantics of the data and being able to expose it as context.

Pariseau: Right. Of course, I'm sure I don't need to tell you that autonomous AI agents guided by data context is a common theme in the industry over the last year. And there are many, many competitive offerings in this space. So, what does Agentic Data Cloud have that enterprises can't get?

Gutmans: I think, you know, it goes back to our fundamental differentiation is Google Cloud, which is we're really the only hyperscaler that has both AI infrastructure, the model innovation, the differentiated data platform, you know.

My other part of my job is actually to run some of these data platforms for Google, so Search, YouTube and so on are all using, you know, systems like Spanner. And that gives us an ability to vertically integrate the stack and really reimagine what a data platform looks like for the agentic era.

So, we're really leveraging that differentiation to bring together the infrastructure, the models, the data platform. For example, with BigQuery, we have integrated Gemini deeply into BigQuery in a way that is way more efficient than others can do who are renting the models versus have their own models in-house and their own infrastructure in-house.

So, a lot of the, I would say, pieces we announced today are differentiated because of how we have uniquely built our infrastructure and AI at Google. That is, you know, very different from what everyone else is.

Pariseau: OK. And also, there is the new Gemini Enterprise Agent Platform more broadly. So, how does the Agentic Data Cloud slot in with that?

Gutmans: So, we work super closely with the Gemini Enterprise team. And so, for example, they announced the Deep Research Agent today -- that's actually using Knowledge Catalog under the hood. So, if you think about the Deep Research Agent, it may, you know, need to reason and do research around data that is both external and internal. When they go to the internal data, they actually go through Knowledge Catalog to understand what data to access. So, it's a very strong partnership.

Now, of course, the business user doesn't have to know they're using Knowledge Catalog, right? It's just under the hood. But we are powering a lot of those capabilities in Gemini Enterprise.

Pariseau: Right. And also, Google Cloud Storage has the Smart Storage. So, how does that feed into the Knowledge Catalog to your point about the infrastructure?

Gutmans: Yes, if you think about, you know, where does most of this unstructured data actually live, right? There's just a huge amount that's sitting on blob storage or, you know, object store, right? Images, videos, PDFs, you name it. And it's not always cost-efficient to bring that data into a data processing engine, you know, use a model to extract metadata.

So, what we've done is this is again part of our differentiation. We've actually pushed down the processing into GCS, it works very closely with our TPUs and Gemini to actually extract that metadata when it's actually sitting on the GCS platform. And so today we announced that we'll automatically do that for images, PDFs coming soon, and we're going to do way more formats over the coming months.

Pariseau: OK. So, also there have been security updates this week with agentic defense. So how does the data management portfolio tap into that?

Gutmans: Yeah, so that's a great question. I think this is also where, you know, having the fully integrated stack is more important now than ever because you want to have a consistent security model. Right? And so everything we've built -- whether it's our MCPs, APIs on GCP, whether it's how our Knowledge Catalog manages permissioning -- it's all built on the same consistent permissioning model of GCP.

And so that is, I would say, another ingredient in this really being a good choice for customers because being able to vertically integrate the stack with the same security and governance is very critical.

Pariseau: So, what's next? What's the next frontier in data management for agentic AI?

Gutmans: You know, I think we're just getting started. I mean, even if I think of Knowledge Catalog, I mean, we're seeing great success.

But the appetite and what we think is possible by using agents to really reason around the data, understand the relationships, understand the business meaning, personalize for the user -- we have pretty big aspirations, you know? We're at our V1, 1.0, right now when it comes to like that gen 2 architecture. You know, I think our vision is strong, but there's a lot of opportunity that we have to kind of, you know, continue to drive down that.

And we also think that as we're kind of breaking down the walled gardens around data access, you know, you mentioned the cross-cloud lakehouse, I think there's even more opportunity for us to just simplify the data state, bring the transactional analytical databases together and make sure that customers can really focus on business outcomes as opposed to having to do all this undifferentiated heavy lifting of, you know, stitching systems together.

This is also a big benefit of, really, the full stack where this Agentic Data Cloud is increasingly pre-integrated for customers, so they can just focus on outcomes.

Pariseau: So, to bring it back up to a higher level, can you give me an example of a workflow or use case and how it will change with the new data cloud as opposed to what people could do before?

Gutmans: Yeah. So, you know, for example, today, you know, you kind of -- let's say you see an anomaly in revenue. Right? And so you see an anomaly in revenue in a dashboard, and then you're trying to root cause it, and you're trying to understand, OK, what are the factors? But why should you even have to notice and go and ask the system?

So, some of the work that we're doing is actually have proactive agents -- they could be firing, you know, whenever -- look at your revenue, and if they see an anomaly, they're actually starting to investigate on your behalf.

You know, this ends up turning around the whole workflow where you're actually getting a message or an email or something that tells you, 'Hey, Beth. Revenue has gone down, taken a look, you know, it's maybe in this market, there's a weather problem.' Right? And so on and so forth. 'And as a result, the store traffic has gone down.' Right? In some cases, you may even have a remediation, right? Maybe you put more of your ads revenue onto other locations or onto online buying as opposed to in-store.

So, there's just a ton of opportunity for us with this system of action to move these things to being very proactive for the practitioners, as opposed to practitioners having to prompt the system.

Pariseau: OK. What do you think is the biggest challenge as far as customers adopting all of this, adopting agentic AI?

Gutmans: I think the biggest challenge is trust. You know? You have to kind of really figure out as a business, you know, where can you allow the autonomous agent? Where does it make sense? Where do you still want a human in the loop? You know? Kind of thinking through that.

So, I'll give you an example. Customer support today is already very much agentified, right? You have other situations, like replenishing inventory -- you know, to a certain level of spend, of course, you can automate, right? There may be other business decisions where you say, you know what, I still want a human in the loop, but the agent can do a lot of work to make that decisioning much, much easier and more effective.

So, I think for customers really thinking through that and thinking, like, how to deploy that -- and that's really where a lot of our focus is, is how do we build the right trust with the model? How do we build the trust with Knowledge Catalog? How do we build trust, you know, on the data security and compliance side? And so that's, that's a big part of our focus.

Pariseau: So, I'm sure there's plenty. What am I not thinking to ask you about that you might want to mention or is on your mind?

Gutmans: You know, I was just sharing with someone that we, you know, we kind of talked about three big announcements, you know, in Thomas' keynote. But Data at Google actually had 80 announcements this week. Right? So, there's like a huge long list -- like, for example, we made Spanner available outside of Google Cloud. It can run on AWS and Azure now or on-premises, even in disconnected mode, like that's a big deal. Like, two years ago, we thought that was impossible and we figured it out.

There's a lot of innovation across all the database services. And so, I think folks are really interested in data. We published a bunch of blogs, bunch of demos, like there's just a lot of interesting and exciting news, you know. I think I mentioned four. There's another 76.

Pariseau: My goodness. But I noticed that across different products, Google is revving up that multi-cloud 'coopetition,' so to speak. And, you know, even five or six years ago, that would have been pretty unthinkable from a marketing and business standpoint. So, what's motivating that change?

Gutmans: You know, we want to meet customers where they're at, and we realized that many customers are going to be multi-cloud for all sorts of reasons -- for example, by acquisition, you know, is a good reason why a customer will get to multi-cloud. And so our goal is really to meet them where they're at, make them successful.

We do see in fullness of time, customers usually are very excited about what they get in our data and AI platform, and we see an increasing amount of their workload coming to Google, you know, typically. But ultimately, we think we need to have an open ecosystem.

Customers really value the fact that we don't just say we're open. You know, we're actually making progress on being more and more open. Iceberg is an example. Spanner Omni is an example. Three years ago, we launched AlloyDB Omni, which brings Postgres to other clouds. So, we're very serious of making sure we meet our customers where they are and we make them successful in a very open way.

Pariseau: What about on-premises? How do you tap into data?

Gutmans: Yeah, so on-premises is also, you know, very important to many of our customers. By the way, I mentioned AlloyDB and Spanner Omni -- both of those can run on-premises. We also deliver the Google Distributed Cloud, which is an offering that can go on-premises.

So, we do a lot of work, both with connected on-premises and disconnected on-premises, so we do have solutions for all the different archetypes.

When it's connected on-premises, there's, of course, a lot of opportunity to just make those systems more of an extension of the cloud and, for example, you have some transactional databases you need to have on-premises -- let's say it's still an Oracle Exadata there -- you know, we can do change data capture, bring that into BigQuery for analytics and AI.

So, a ton of our focus has been also to enable these both connected and disconnected use cases.

Pariseau: Great. Well, again, it's a crazy hectic show -- like you said, 80 announcements. So, I appreciate you taking the time to chat.

Gutmans: Yeah, no, thank you so much for having me. Really appreciate it.

Pariseau: Thank you for tuning in to IT Ops Query. To learn more about enterprise software development and platform engineering, explore our content on Informa TechTarget sites. Find us on YouTube at our channel, Eye on Tech. Subscribe to our podcast to receive the latest episodes as they drop. And if you liked what you heard today, give us a rating and review on Apple, Spotify or wherever you're listening. Thank you for joining us.