In this interview from Google Cloud Next 2026, Yasmeen Ahmad, managing director of product management, data and AI cloud at Google Cloud, joins theCUBE's Dave Vellante to discuss how AI agents are displacing humans as the primary users of enterprise data platforms — and why that makes the modern data stack the new legacy. Ahmad explains that agent-native architectures require a fundamental rethink: where traditional stacks centered on SQL engines optimized for human queries, agents need vector search, embedded AI reasoning and graph capabilities. She also highlights how the shift from developer APIs to tools and skills unlocks new scale — enabling thousands of modular capabilities without the brittle tech debt of versioned API management.
The conversation explores how Google Cloud frames this evolution as a move from systems of intelligence to systems of action, with the knowledge catalog emerging as the critical missing layer. Ahmad reveals that early generative AI deployments on raw data platforms topped out at around 50% accuracy — and that a context layer providing aggregation, enrichment and hybrid search is what closes the gap. Partnerships with Salesforce, Workday, SAP and ServiceNow reflect the need to aggregate context across enterprise SaaS platforms alongside structured and unstructured data sources. She also unpacks the newly launched data agent kit, which consolidates data engineering and data science agents into modular tools compatible with Cloud Code, VS Code and Gemini CLI. From Apache Iceberg enabling a true cross-cloud lakehouse without forcing enterprises to anchor to a single cloud provider, to Shopify's leaders reframing themselves as managers of agent swarms, Ahmad provides a roadmap for organizations ready to move from AI-assisted workflows to fully AI-native operations.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
Google Cloud Next 2026. 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 the link to automatically sign into the site.
Register for Google Cloud Next 2026
Please fill out the information below. You will receive 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 Next 2026.
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 Next 2026. 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 the link to automatically sign into the site.
Sign in to gain access to Google Cloud Next 2026
Please sign in with LinkedIn to continue to Google Cloud Next 2026. 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
Yasmeen Ahmad, Google Cloud
In this interview from Google Cloud Next 2026, Yasmeen Ahmad, managing director of product management, data and AI cloud at Google Cloud, joins theCUBE's Dave Vellante to discuss how AI agents are displacing humans as the primary users of enterprise data platforms — and why that makes the modern data stack the new legacy. Ahmad explains that agent-native architectures require a fundamental rethink: where traditional stacks centered on SQL engines optimized for human queries, agents need vector search, embedded AI reasoning and graph capabilities. She also highlights how the shift from developer APIs to tools and skills unlocks new scale — enabling thousands of modular capabilities without the brittle tech debt of versioned API management.
The conversation explores how Google Cloud frames this evolution as a move from systems of intelligence to systems of action, with the knowledge catalog emerging as the critical missing layer. Ahmad reveals that early generative AI deployments on raw data platforms topped out at around 50% accuracy — and that a context layer providing aggregation, enrichment and hybrid search is what closes the gap. Partnerships with Salesforce, Workday, SAP and ServiceNow reflect the need to aggregate context across enterprise SaaS platforms alongside structured and unstructured data sources. She also unpacks the newly launched data agent kit, which consolidates data engineering and data science agents into modular tools compatible with Cloud Code, VS Code and Gemini CLI. From Apache Iceberg enabling a true cross-cloud lakehouse without forcing enterprises to anchor to a single cloud provider, to Shopify's leaders reframing themselves as managers of agent swarms, Ahmad provides a roadmap for organizations ready to move from AI-assisted workflows to fully AI-native operations.
In this interview from Google Cloud Next 2026, Yasmeen Ahmad, managing director of product management, data and AI cloud at Google Cloud, joins theCUBE's Dave Vellante to discuss how AI agents are displacing humans as the primary users of enterprise data platforms — and why that makes the modern data stack the new legacy. Ahmad explains that agent-native architectures require a fundamental rethink: where traditional stacks centered on SQL engines optimized for human queries, agents need vector search, embedded AI reasoning and graph capabilities. She also hig...Read more
In this interview from Google Cloud Next 2026, Yasmeen Ahmad, managing director of product management, data and AI cloud at Google Cloud, joins theCUBE's Dave Vellante to discuss how AI agents are displacing humans as the primary users of enterprise data platforms — and why that makes the modern data stack the new legacy. Ahmad explains that agent-native architectures require a fundamental rethink: where traditional stacks centered on SQL engines optimized for human queries, agents need vector search, embedded AI reasoning and graph capabilities. She also hig...Read more
exploreKeep Exploring
Why is the modern data stack not appropriate for the AI (agentic) era?add
How should we evolve system and data-platform architecture to support agents (including swarms of agents) — specifically addressing increased API call volume, latency/speed requirements, and integration from the data layer through AI models and hybrid computing?add
What role does a large language model (LLM) play in an enterprise AI architecture, and how should it be integrated with context/knowledge catalogs and the action/transaction layers to produce reliable, agentic outcomes?add
How are customers reacting to the availability of powerful AI models, and what tools or products are you offering to help enterprises use AI for data engineering and data science?add
How is the role of humans evolving as organizations adopt and orchestrate AI agents?add
>> Hi everybody. Welcome to this special Cube presentation. We're here at the Analyst Summit at Google Cloud Next 2026. My name is Dave Vellante, I'm here with Yasmeen Ahmad, who's the managing director of Google Data Cloud. Yasmeen, good to see you again. Last year we were on The Cube on the show floor, here, it's kind of intimate, quiet. We get to have a nice conversation, so I really appreciate you spending some time with me.
Yasmeen Ahmad
>> It's great to have you back here at Next, and great to be having this conversation.
Dave Vellante
>> As you know, I'm usually on the set. It was really a pleasure being part of the analyst program, and you and I had some conversations upfront. You gave me a little preview as to what you guys were doing. It's actually quite remarkable what's happening with the so called modern data stack. It's like the new legacy now, isn't it?
Yasmeen Ahmad
>> Yes.
Dave Vellante
>> So we separated compute from storage, we made it cloud elastic, all that wonderful stuff, BigQuery. You guys were right there with Snowflake, and some others creating that world, but everything changed. It feels like overnight. Why is the modern data stack not appropriate for this AI agentic era?
Yasmeen Ahmad
>> So you're absolutely right to say the modern data stack is kind of looking a little bit legacy. Now that's not to say that all of that innovation is gone, it's just there's a new user in town. That user is the agent. So even when we sit down to do strategy planning, we're no longer thinking about human personas like data engineers, data scientists. We're putting agent as the persona, because we really believe agents are going to be operating data platforms. And there will be humans, but humans are taking a much higher level orchestration role, not the doer role. But when it comes to the stack, think about what an agent needs. What an agent needs is very different to what a human user would need. Just a few examples. When we think about agents, agents don't do querying. The traditional stack was built around this SQL engine at the heart of it, and it was about actually optimizing this engine to be as fast and efficient at these SQL queries. Well, this new agent in town is reasoning. In fact, it's actually not super great at writing SQL, but it's great at doing search over unstructured data, reasoning over that data, this multi-reasoning loops that Gemini can do now. So fundamentally we had to rethink, is it a SQL engine anymore? Do we even have a SQL engine? We still have the SQL engine, but what's critical is the engine has more parts to it. It has to include vector embeddings, vector search. It has to have AI embedded reasoning. It has to have graph capability. These are things that we never imagined for that engine to have before. That's just one example. Another one I would say is, APIs. As a product engineering team, we always built APIs on our data platforms. So developers could come and build applications. Well, frankly, agents don't work with APIs. They work with tools and skills. And it actually opens up a whole new world for us because before we would release five APIs, 10, 50. Every API was tech debt that you had to manage. Because if you change an API, a developer's application was going to break. Now with tools, you can have thousands of tools, and that's part of what we've been launching here.
Dave Vellante
>> Well, you made the point today too, because we've all experienced this. Maybe a lot of people don't know this, but when you have this wonderful dashboard, and it works beautifully, and then all of a sudden it doesn't because you haven't been there in maybe a few months, and it's because the API broke, it's brittle. You said that the agents had no problem. They'll just fix it in real time. Okay. So that's interesting and unique. But how do you guys approach it? So how do you go from the legacy, the new legacy, which is the modern data stack, which you guys kind of helped invent. How do you go from there to this new world? And what happens to all that sort of technical debt that you guys built up? Do you break it down, tear it down and rebuild it? Or what's your approach to this problem?
Yasmeen Ahmad
>> So we see it as moving from systems of intelligence and evolving to systems of action. So it's not that all of that stack goes away, but it's got to evolve to meet this new user, the agent. So when I think about the system of action, we're kind of thinking really hard about, how do we architect this new data platform to meet what agents are doing? And it's not just one agent, we're talking about swarms of agents now operating on top of the stack. In fact, in a session earlier today, I was talking about how, in terms of web traffic, some of the API gateways are seeing massive spikes in incoming queries. So in the time that a human user would normally do one click, an agent coming in is doing 10 to 20 calls. And so suddenly how you architect this stack, and the latency and the speeds and how it operates right through from the data layer to the AI model layer right down to the AI hybrid computer, you have to look at that optimization that you're doing.
Dave Vellante
>> So there's some new terms here. Hypercomputer, that's a new term. I like that term. That's pretty cool. You use the term system of intelligence, we use the term system of intelligence, but I think we use it differently. So to us, the system intelligence is a fundamental requirement for the system, what you call the system of action. So it does the data harmonization, what you might call the semantics, and it provides the inputs and takes the outputs of the LLM and provides inputs to them. So it's an intelligence layer, which is this cognitive surface, which is new, and then it feeds the system of action. So help us understand how you see this new stack emerging. You've got the Hypercomputer. I've got a picture of it here. I'm going to steal it. You've got the Hypercomputer down at the bottom of the stack. We're going to talk about this, the cross-cloud Lakehouse. That's super exciting because now we have this virtual data store, essentially. You don't have to move the data all around. The knowledge catalog is sort of what we would normally call the system of intelligence. So it's a knowledge graph and that's where the context is, and the semantics. And then that feeds the system of action, but it's necessary because you have to trust it. So take us through how you guys are thinking about this new era, this new stack.
Yasmeen Ahmad
>> So I love how you framed this. You have that system of intelligence, it's creating knowledge, it's creating intelligence. And then we are layering on that system of action. I think about it in terms of the agentic harness. At the center of things, you have these amazingly powerful now multimodal, super creative models. So Gemini three, it can do multi-step reasoning. It does thinking mode, all of these cool things. But you need that system of intelligence, that intelligence layer, the context layer, you need the action layer around it. Because that harness around it is what makes these models useful in an enterprise context. So that harness is having that access to enterprise memory. Be it analytical data or operational data. It's having access to the context, that knowledge catalog, the context layer. And then for action, it's having access to tools, to skills that mean as those agents generate great intelligence, how can we have those agents reach into an ERP system, and actually book an order? How can it reach out into a ledger? Can it reach out into the CRM system and activate a new campaign? So for us, the system of action takes it from being a great intelligence generation to actually an agentic engine that can go take action, and drive outcome.
Dave Vellante
>> Okay. So the LLM is the reasoning component of this.
Yasmeen Ahmad
>> Yes.
Dave Vellante
>> It's the probabilistic piece. We always talk about that. And then what we would look at is the transaction layer, the database, the applications. Those are the deterministic components, and somehow they have to work together. So how do you guys think about the role of the LLM? I mean, Gemini is amazing. By the way, Nano Banana is unbelievable. I live off of Nano Banana because I'm a terrible artist. But anyway, aside. But what role does the LLM play? How do you see it evolving? People talk about small language models and specialized language models. I feel like the LLMs themselves, a frontier model like Gemini, become so capable of doing so many things. I think it's going to play a bigger and bigger role in that stack. Do you agree? What role does it play? How does it fit?
Yasmeen Ahmad
>> This is a great question. And actually, we have learned a lot over the last three years as we've been bringing now generative AI together with data platforms. I'll be honest, three years ago, we were focused on cleaning the data, making sure AI-ready data meant good, clean, structured, well-lineaged data. Well, frankly, as we focused on that, it only got us to 50% accuracy. 50, 60% accuracy with the models because you have this really powerful world model. It understands the web of data. You're trying to marry it with this data platform with basic RAG pipelines, and actually the connection was not working very well. The trust and accuracy was not there. What we discovered is the missing 50% is actually the knowledge catalog. It's the context layer. So rather than have an agent or an AI directly connect to a raw data platform, it actually needs a context layer as the translation layer in between. And that context layer is the business intuition, the meanings of data. It's the representation of even what data is there, what quality that data is, and that knowledge catalog acts as that translation layer to help the AI model, or the agent get to the right data with the right context at the right time.
Dave Vellante
>> So you met George the other day, George Gilbert on the phone, on the call. So I want to give him credit because it was his research and thinking that got us here that... He calls it the 4D map of the enterprise. We talked about this a little bit. Where it's normally we store things in strings that databases understand. And now when you talk about the knowledge catalog, or what we say the knowledge graph, it's people, places, things and activities, which could be processes. These processes might live in applications. There's workflows that you need access to. Giving the LLM direct access to that is messy. It doesn't work. And so you're bringing that. Your architecture, if I understand it correctly, is bringing that together with context, and of course with governance as well. So you've got things like AlloyDB, you've got BigQuery, you've got Spanner. So you've got transactions and analytics coming together. And it's that layer, that new emergent layer. This is very high value real estate people in this new emerging stack. It's that layer that actually completes the picture. Is that right?
Yasmeen Ahmad
>> That is correct. And with the knowledge catalog, we talk about three foundational capabilities, aggregation, enrichment, and search and retrieval. Key is aggregation. Like you mentioned, there's multiple data systems there. There's analytical platforms like BigQuery. You've got operational databases like your Postgres, AlloyDB, Spanner, also your SaaS applications, which have a lot of business logic encoded in them. That's why we announced the partnerships with Palantir, Salesforce, Workday, ServiceNow, SAP. We need to aggregate actually context from all of those sources. So the first thing the knowledge catalog does is understand data across multiple clouds, across application layers, and analytical operational platforms. Next is the enrichment piece because it's great having data and structured data comes actually with a lot of schema and metadata information baked in. Unstructured data doesn't have much schema. It doesn't have much description. You have a thousand PDFs. There's no reference to what entities, is it people, places? How are they connected? What's the hierarchies? So the enrichment part of knowledge catalog is now auto generating that context for unstructured data. We're auto generating and inferring relationships across data sets, doing disambiguation. But the third piece, which you mentioned, I think super critical is the search and retrieval. Because in search and retrieval, first you got to be aware of the access policies and governance around who can access what context and data. But next is, we know agents actually degrade in performance if you give them too much context. So part of the knowledge catalog is it doing the right search and retrieval, ranking. We actually use the hybrid search stack we built for Google search, and we've brought that in so that we can actually serve up the right context.
Dave Vellante
>> Interesting. I want to go back to something you said before about some of the learnings that you had. Because early on there was kind of conventional wisdom. You got to get your data house in order before you go spending resources on AI. What we found, and I did some research with five really leading advanced firms. I mean, it's not a secret. It was like JP Morgan type of firms, Lilly, firms like that, that really had the resources to go out and bring in AI expertise. And they said almost to a firm, "Yeah, we thought the same thing." And what we found was, didn't work. What we needed to do was let the AI help us clean the data. And then we got on the AI curve and it created this flywheel. So the earlier we got on, the better it was. The first project might have taken whatever, 15 months or nine months, whatever it was. The second one was faster and faster and faster. And then we were able to redeploy staff to do higher value work. We were able to use the AI to improve the data and we were able to put down a governance structure that was repeatable. And now we're like on the curve, and we're seeing real productivity gains. It sounds like you've gone through a similar... I mean, you Google obviously advanced use for customer zero. Are your customers finding the same thing? Are they understanding the importance of getting on that AI curve sooner rather than later?
Yasmeen Ahmad
>> I think customers across the board are understanding the need to get onto the AI curve because actually these powerful models are uniformly available to everybody out there. Whether you're a three person startup, or you're the world's largest multinational enterprise, everybody has in their hands these powerful capable models to go invent the future. So as an enterprise, if you're not adapting, and flexing and getting onto that AI curve now, somebody out there is retransforming, redesigning your entire industry. So I think that urgency, definitely see that from customers to say, "Hey, it's not about trying to figure out the entire data stack to begin with." We actually have to get going with those initial use cases. But to what you mentioned, we actually have to use AI to help us with the data work. And that's been a big unlock as we look at now intent driven engineering. The models have gotten so good that if you give them tools and skills, like data engineering tools, data science tools, the models can do a lot of hard, heavy lifting for you. And that's actually our data agent kit that we launched today. We actually took our data engineering agent that we launched last year, the data science agent we launched, all of the agents we'd launched, we flattened them, broken them down into tools and skills, packaged them into a data agent kit. That data agent kit can plug into Cloud Code, VS Code, Codex, Gemini, CLI, and suddenly that environment can help you build an entire medallion architecture and data pipelines and do all of your modeling.
Dave Vellante
>> You know, Yasmeen, just speaking with you, and thanks again for taking some time with me. We were saying before about SQL, I feel like if you are somebody who's really good at SQL... And I was talking to some of your customers today, and what they were saying is, we need people who know how to build and deploy and manage and govern agents. So that's the pivot that you need to do with your career because it's... Maybe it's not that... If you can do that great SQL work, it's not a huge pivot to actually learn how to do this agentic work. Is that what you've found? And that there's actually a reasonable adjacency there.
Yasmeen Ahmad
>> There is definitely an adjacency there. And we talk about the idea of the role of the human evolving. Humans are becoming orchestrators of agents rather than doers of tasks, which I think is frankly very liberating for humans. It's no longer, I have to get these three tasks done. It's focusing on what's the business outcome or goal. And by the way, you have a swarm of agents that you are managing and orchestrating. And we see customers moving in that direction. I actually spoke earlier in a session with Farhan, VP of engineering over at Shopify, and he said their leaders are actually now reframing, and thinking of themselves as managers of agents. And they're actually hitting delete on some workflows that were human-centric before because these agents can do them in a different way and much more optimized and faster. So I think there's a big cultural shift there for enterprises as we look at what is the role of the human now?
Dave Vellante
>> Well, and I think the message there is don't pave the cow path. Don't apply agents to a process that could be improved dramatically. One last question I have is, we talk a lot about Iceberg and open table formats. How do you make this not a walled garden? What's Google's philosophy there? Convince me that it's not going to be a big lock-in.
Yasmeen Ahmad
>> So when I think about how the industry and my customers have been talking to me for the last five years, everybody is multi-cloud. I find it hard to recall a customer who didn't have some data that lived in another cloud. So it's a reality of today. But actually as we've looked at the landscape, a lot of vendors spoke about multi-cloud offerings, but you still had to choose a cloud to deploy the technology. So fundamentally, it was still forcing a choice of which cloud were you going to gravitate around. So when we took a look at the space, we really wanted to bring a cross-cloud approach. Don't move all your data to one place. How can we enable you to connect the data regardless of where it resides? But actually, fundamentally, the technology didn't exist before. A year ago, if you asked me this, I don't think it would have been possible. You mentioned Apache Iceberg. Iceberg has become a defect to open standard. That is enabling both Google and our partner ecosystem to share data in ways, and with ease that we just couldn't before. It would have taken us two years to build a unique federation with a unique partner. Now with Iceberg, and the Iceberg rest catalog, you can immediately connect. And then the other big, of course, thing is cross-cloud interconnect. Suddenly latencies, egress, aren't a concern anymore. And so that paves the way for a true cross-cloud solution that just wasn't there before.
Dave Vellante
>> Well, you guys are moving at the speed of AI, Yasmeen, and of course data is at the heart of it. So thanks so much for spending some time with us.
Yasmeen Ahmad
>> Thank you.
Dave Vellante
>> All right. And thank you for watching. This is Dave Vellante reporting from the Analyst Summit here at Google Cloud Next, 2026. You're watching, The Cube.
Dave Vellante