Teradata empowers enterprises to turn intelligence into autonomous action, grounding AI agents in deep business context and trusted data. Teradata’s push into AI has been deliberate and layered. Sumeet Arora, Chief Product Officer at Teradata, discusses fragmented data silos, how to use AI agents and how the workforce is changing.
Tell me more about Teradata’s origin and how it has grown.
Teradata was established in 1979 from a collaboration between researchers at Caltech and Citibank’s Advanced Technology Group. From day one, the mission was clear: to help organisations bust functional silos and make sense of large volumes of data at a scale that no one else could match.
But what is most fascinating about our origin is how relevant it is today, as we are now in the era of the autonomous enterprise. Teradata was built for efficiency and high performance, not just elasticity in the cloud. This matters now more than ever because AI agents do not switch off. They run continuously, placing constant demands on data, and our platform – engineered to be the most efficient with massive amounts of data and a need for robust performance – is exactly what enterprises need for this always-on world.
Teradata’s push into AI has been deliberate and layered. From execution environments and governance controls to pre-built agent templates, so enterprises can move from experimentation to production at scale. And if your agents are knowledge or data starved, Teradata is enabling the combination of structured data and unstructured information to be available to agents without moving data and increasing risk.
The thread across all of it is trust, context and scale. Enterprises aren’t casual data users – they need AI that understands their data, their industry and their governance requirements. That’s exactly where Teradata is uniquely positioned.
Why do companies find themselves with fragmented data silos?
Fragmented data silos are really a product of how enterprise technology evolved over the decades. We are moving away from having hundreds of siloed applications with rigid, workflow-centric interfaces. Each application served a specific function, but none of them were designed to talk to one another cohesively.
The result is that enterprise knowledge, which lives in your data, your processes and how you make decisions, has become scattered and disconnected. And as unstructured data has exploded in growth, traditional systems simply were not equipped to handle that variety in one unified place.
Many enterprises face significant barriers to scaling AI, with fragmented data being one of the biggest culprits. These constraints prevent organisations from realising the full potential of Agentic AI. That is precisely why our vision is to bring everything together into what we call a Knowledge Fabric, where all silos are busted, and enterprise knowledge is unified, governed and ready for AI.
What scalability issues do businesses face when deploying AI agents?
One of the biggest challenges is moving from experimentation to true enterprise scale. While many organisations are piloting AI agents, far fewer are able to deploy them reliably across the business.
A key issue is infrastructure. AI agents rely on continuous access to large volumes of both structured and unstructured data, often in real-time. Many existing systems were not designed for this level of concurrency, complexity or data diversity, which can lead to inconsistent outputs.
There is also the challenge of orchestration and governance. As agents scale, businesses need to manage how they access data, make decisions and interact with other systems, while maintaining accuracy, security and compliance.
Ultimately, scaling AI agents is not just a technology problem, it is an operational and an economic one as well where progress must be made while minimising risk. Organisations need platforms that can handle high volumes of queries, unify data across environments and support reliable, governed execution, otherwise, agents remain stuck in pilot mode rather than delivering enterprise-wide impact. This vigorous use of enterprise data and knowledge by AI must be done with economic efficiency without introducing data leak risks associated with data having moved to serve AI. We must deliver AI without moving data.
What’s the biggest benefit of AI agents? Any downsides?
The biggest benefit of AI agents is both transformative and deeply practical. Imagine an agent that tracks your flights, predicts delays, reroutes you, reschedules your meetings and simply tells you when to leave your house. That single example captures what agents fundamentally change, shifting us from a world where technology merely informs us to one where it actually performs on our behalf.
For enterprises, that means AI agents can autonomously orchestrate workflows, make decisions within defined governance frameworks and uncover insights across every data type, all without manual intervention.
The downside, however, cannot be ignored. Research shows that 95% of AI projects are failing, largely due to bad data. If the knowledge foundation is not AI-ready, agents will make poor decisions at speed and scale. Governance, security and data quality are not optional extras, they are prerequisites. This is why we are laser-focused on ensuring knowledge is AI-ready, so organisations can actually achieve measurable business outcomes.
Why is there a lack of unified access to structured and unstructured data?
The lack of unified access really comes down to how structured and unstructured data were historically treated as entirely separate problems. Structured data has been the backbone of enterprise analytics for decades, while unstructured data, such as documents, PDFs, images and audio, was largely an afterthought managed by separate systems with no common governance layer.
Now, unstructured data is growing at three times the rate of structured data, yet most platforms were not built to handle both together. The consequence is that AI agents cannot access a comprehensive enterprise context. Think about a financial services agent trying to answer a question about loyalty discount eligibility. It needs to combine unstructured policy documents with structured business data simultaneously, and SQL alone simply cannot address that. Without unified access, enterprises are left with incomplete answers and agents that cannot do their job properly.
What is your favourite feature of Enterprise Vector Store?
Teradata’s ability to concurrently deliver structured data and vector data processing with the same engine not only implies efficiency and ease of use, but it also means that enterprises can co-locate their data and knowledge with their AI tools eliminating risks that can arise from movement of data. Teradata Enterprise Vector Store is multi-modal with built-in ETL and feature enrichment capabilities for documents, audio, images and soon videos. Lastly, Teradata’s compute choices enable a choice across always-on efficiency with mixed workloads as well as ephemeral-elastic capabilities for workloads that fit that pattern.
How do you see the workforce changing over the coming year?
The workforce transformation we are entering can be exciting when approached with the right mindset. We are building for a world where humans, agents and data all work together, because this is not about replacement, but collaboration.
The shift I see most clearly is the move from workflow-centric work to outcome-centric work. Today, knowledge workers spend large amounts of time navigating applications and manually connecting dots across siloed systems. AI agents will absorb much of that friction. When you work with an intelligent platform in the future, it will feel like working with a human. You ask a complex inventory question, and agents instantly pull the context, run the analytics and give you prescriptive, explainable strategies to act on.
Every engineering team inside Teradata is already building agents to make our own stack more intelligent and easier to use. The workforce will increasingly be defined not by who can operate software, but by who can collaborate effectively with agents.
What’s next for Teradata?
We’re focused on one central mission: making enterprise Agentic AI real, not just in pilots, but in production, at scale, with full governance. The launch of our Enterprise AgentStack brings together AgentBuilder, our MCP infrastructure, AgentEngine, and AgentOps into a single integrated toolkit that takes organisations from experimentation to deployment.
Every engineering team inside Teradata is working to make our stack more agentic, intelligent and easier to use, and we are moving toward a chat-first, natural-language-driven interface, where working with our platform will feel like working with a human. I always tell my team that distribution is everything in innovation. We may have built a Ferrari, but if it stays in the garage, nobody benefits. The real challenge is turning that potential into something people can use with confidence.
What differentiates our approach is that we’re not asking customers to choose between capability and control. Teradata is the knowledge fabric that binds enterprise AI together. We are a governed, high-performance data foundation that makes every model, every agent and every workflow more effective and more trustworthy, whether deployed on-premises or in the cloud. As AI agents become a standard feature of enterprise operations, the question won’t be which LLM you choose; it’ll be whether your data platform can support them at scale, with integrity. That’s where we’re building, and that’s where we’re winning.
