Generative AI is no longer just an experimental topic: It’s become a core focus of enterprise strategy. Behind the scenes, companies are moving beyond proofs of concept and into the more challenging work of scaling infrastructure, establishing governance and delivering measurable outcomes. As the excitement settles, the center of gravity is shifting from innovation theater to operational reality.

That shift requires less fanfare and more friction management, according to Bruno Aziza (pictured), group vice president for data, business intelligence and artificial intelligence at IBM Corp.

IBM’s Bruno Aziza talks with theCUBE Research’s Dave Vellante about what’s possible when agents work across apps, clouds and workflows to drive real value.

“While there’s excitement, we now need to think about a true agentic strategy, which includes cost and governance,” he said. “We’re not talking about this as much because it’s not sexy, but I guarantee you the next six months, probably the next five years in fact, are going to be about that.”

Aziza, along with executives and partners from companies including PepsiCo Inc., Heineken N.V., United Services Automobile Association, IBM and leading analysts and data strategists, spoke with theCUBE’s Dave Vellante at IBM Think, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed IBM’s agentic strategy, the challenges of scaling generative AI in enterprise environments and the operational frameworks needed to make AI work in production. (* Disclosure below.)

Here are three key insights you may have missed from theCUBE’s coverage of IBM Think:

Insight #1: Enterprises are scaling generative AI from architecture to return on investment with deliberate intent.

PepsiCo Inc. has embedded generative AI across procurement, supply chains and customer engagement. That shift didn’t begin with flashy prototypes: It started with architecture. By building a unified data lakehouse and composable foundation, PepsiCo created the conditions for scalable AI impact, according to Magesh Bagavathi, senior VP, global head of data, analytics and artificial intelligence at PepsiCo, and  Shobhit Varshney, head of data and artificial intelligence, VP and senior partner at IBM.

USAA’s Ramnik Bajaj talks with theCUBE about how IBM’s technologies are helping bridge structured and unstructured data to support AI use cases.

“If you really want to make yourself a data and AI company, you have focused attention on the priorities, you put money behind it, and … you have one foundation, what we call an EDF, enterprise data foundation,” Bagavathi said during the event. “Then you start seeing the value unlock … wins in Turkey, in France. We started seeing wins in many markets. Automatically, we started seeing the population and the convergence. You’ve got to see wins.”

Underpinning those wins was a deep investment in governance and consumption frameworks. Golden key performance indicators, consistent reporting and contextualized insights helped shift PepsiCo’s data culture from fragmented silos to decision intelligence, according to Bagavathi and Varshney. With IBM’s support, the team optimized for scale, turning localized efforts into global systems.

“If you have a seller in Texas who says … ‘Give me sales for Paris,’ we know that the context is that the seller has access to only Texas territory, hence we’re talking about Paris, Texas,” Varshney told theCUBE. “Bringing in that context to improve the accuracy of data is amazing.”

United Services Automobile Association’s approach builds on PepsiCo’s, emphasizing operationalization and value extraction from AI in live production. The company embeds generative AI directly into core processes, supported by centralized architecture and strict governance standards. It’s not just about speed; it’s about trust, according to Ramnik Bajaj, SVP, chief data, analytics and AI at USAA.

“I call it first-class data now that it is possible to do those types of analytics,” he told theCUBE. “Having all of those in one place in the enterprise … really supercharges our ability to move fast and move responsibly. With a technology like gen AI, it’s important to also learn our way into the right architectural patterns that can give us not just speed to market, but also reliability [and] accuracy, preserving trust and … privacy. All of those core data management functions tie very well into that AI genre.”

USAA’s collaboration with IBM is a key piece of this strategy. By combining IBM watsonx with the company’s Granite models, the company is bridging unstructured and structured data to extract actionable insights from documents such as claims and underwriting files, according to Bajaj. The ability to translate real-world inputs into an AI-ready structure is a critical enabler for automation at scale.

“We’ve run some very successful pilots with the watsonx products, and we are looking at it from two angles,” Bajaj said. “One is the problem of tackling unstructured and structured data together. Our documents, claims or any of the underwriting documents are English-language documents, but … there can only be six options. To be able to call out those entities and … to use them as structured elements, that’s something IBM is providing.”

Here’s the complete video interview with Magesh Bagavathi and Shobhit Varshney:

Insight #2: Data architecture and modular flexibility power AI everywhere.

Global enterprises sometimes trade agility for control when scaling AI, but that tradeoff is starting to shift. Heineken N.V. is rewriting the equation with an adaptable data stack designed to keep the nuance of local markets while driving connected intelligence globally. With IBM’s help, the company built a flexible platform that allows for experimentation, faster decision-making and secure AI deployment across decentralized operations, according to Ronald den Elzen, chief digital and technology officer​ of Heineken N.V

“We want to move from a very decentralized, fragmented approach to a much more centralized approach, but modular, composable in different pieces because we need to be agile in the world of today,” he told theCUBE. “I need to build in some flexibility on the tech stack … so that we can modify to changes in innovation. The world is going so fast that we need to be flexible.”

Distributed data architecture plays a critical role in making AI usable across environments, according to Ed Calvesbert, VP of product management for IBM watsonx.data at IBM, and Scott Brokaw, VP of product and data integration at IBM. The company’s approach combines cloud, on-prem and mixed environment performance without forcing unnecessary data movement. That design unlocks cost savings and operational efficiencies while enabling more advanced generative AI use cases across structured and unstructured data.

IBM’s Ed Calvesbert and Scott Brokaw talk with theCUBE about the company’s strategy around data, hybrid infrastructure and governance.

“Data workloads belong where data actually is,” Brokaw said during the event. “Why are you going to move data just to be able to do transformation or processing or analytics? Locality costs data, prominence costs, being able to actually run where the data is … should be in every client’s first order of purge.”

Heineken’s journey reflects the broader shift in enterprise priorities: From data collection to connected intelligence. AI can only be effective when fed by trusted, harmonized data, which is why the company worked with IBM to modernize its data backbone and build AI-ready systems that balance performance with governance, according to den Elzen. This approach lets Heineken scale insights across markets while staying grounded in local context.

“IBM helps us in many of these elements … on helping to clean and harmonize our data management on the digital backbone,” he told theCUBE. “They help us to build the future on the tech stack. They help us on data, and they helped us on decommissioning. Now, they’re helping us with their AI and their gen AI … to do things in a much smarter, faster way and cheaper way.”

That drive for modularity isn’t unique to Heineken. Across the board, enterprises are looking to balance local nuance with architectural consistency, and that’s where IBM’s strategy shows its depth. Through watsonx.data, IBM emphasizes a distributed data architecture designed to support real-world AI deployment across diverse environments, according to Brokaw and Calvesbert.

“You get a lot better accuracy, which means you can do more use cases,” Calvesbert said during the event. “It’s not just about information retrieval now. It’s analytical. It’s operational. That’s the next generation of agents and apps. That’s number one. Number two, you unlock all that unstructured data for traditional lakehouse workloads, business intelligence, data engineering [and] machine learning.”

Here’s the complete video interview with Ronald den Elzen:

Insight #3: IBM’s hybrid integration strategy positions it as a leading orchestrator in the enterprise AI stack.

While the generative AI race has spurred a frenzy of scale-first strategies, IBM is leaning into a subtler, more sustainable path: Agentic orchestration across hybrid environments. That means bridging clouds, tools and teams under a unified execution layer that balances openness with governance, according to Aziza.

“We’re … learning what really matters is to think … about the idea of being more than just agents,” Aziza told theCUBE. “I think while everyone’s thinking about the opportunity of engineering new content and getting better engagement, you also have to think about everything that matters in production. How are you going to manage costs? How are you going to manage governance?”

IBM’s agentic strategy blends automation and orchestration by embedding AI agents directly into existing enterprise systems. Rather than a rip-and-replace approach, IBM enables customers to scale into orchestration gradually, layering agentic capabilities across legacy workflows, accelerators and third-party platforms, according to Ritika Gunnar, general manager for data and artificial intelligence at IBM.

SanjMo’s Sanjeev Mohan talks with theCUBE about IBM’s evolving strategy and data platform aspirations.

“Being able to execute across agents that you’ve built … that exist in your third-party systems, across stuff that maybe you have done from accelerators that we have given you and your fixed systems, is important,” she said during the event. “Your existing estate is really critical in that discussion, and bringing your clients in this journey so they can see that as quickly as possible is important.”

The combined emphasis on orchestration, visibility and integration — from Aziza’s conductor analogy to Gunnar’s focus on agentic layering — resonates with analysts tracking how IBM is building a distinct moat in the enterprise AI market. The company isn’t chasing hyperscalers on compute economics; it’s focused on cohesive integration across Red Hat Inc., consulting and open tooling. That strategy enables IBM to serve as a modular layer atop fragmented data environments, delivering orchestrated business value, according to Sanjeev Mohan, principal analyst at SanjMo.

“They are driving this application story now with AI agents, with IBM Consulting,” he said during the event. “They’re coming from the top down, [with] data being spread across hybrid and cloud and on mainframe gives them the differentiation.”

IBM’s recent acquisition of DataStax Inc. underscores that vision, according to Mohan. The move wasn’t about owning Cassandra, the open-source database, but about acquiring Langflow, a tool designed for business developers to orchestrate AI workflows without heavy technical friction. Coupled with the watsonx portfolio and orchestration layer, IBM is assembling a full-stack architecture that unifies data, governance and generative AI.

“Although there are a lot of new products … I think with watsonx.data, watsonx.governance [and] watsonx.ai, they’ve got the three pillars,” Mohan said. “They’ve got data, AI and governance. Watsonx Orchestrate is the layer they’re putting on top to orchestrate different pieces and help business users develop AI pieces.”

Here’s the complete video interview with Bruno Aziza:

To watch more of theCUBE’s coverage of IBM Think, here’s our complete event video playlist:

(* Disclosure: TheCUBE is a paid media partner for IBM Think. Neither IBM, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo:  SiliconANGLE