“Every agent needs its own identity, access limited to only the data and systems its task requires, and a human owner who is accountable for what it does.”
Sumeet, your career has moved from networking systems at Cisco to AI-powered analytics at ThoughtSpot, and now to product leadership at Teradata. What lesson from those different stages still shapes how you decide which problems technology should solve?
My product philosophy remains consistent across the various roles I’ve had, and I’m a firm believer that customers don’t buy technology for technology’s sake. They buy the ability to do something they couldn’t do before or do it much better than before. So, I always ask three questions. What problem are we solving and for who? Is this a problem worth solving – will it change outcomes for a business? Can we solve it in a way that is significantly better than the existing state?
As a summary, I have built great teams, great products, and great businesses that are inspired by compelling industry-changing missions (examples: secure affordable internet, enterprise analytics for all knowledge workers, AI fused together with enterprise knowledge wherever that knowledge lives).
And once we commit to the mission, I think it’s equally important for people in my role to not just build a great product but also build it fast. It has to be both fast and excellent. And doing things faster in this era means that you have to also treat velocity as a product itself.
At ThoughtSpot, you helped advance AI-powered analytics, where the value of an answer depends on the data and context behind it. What changes when AI moves from generating insights to taking actions that can directly affect business outcomes?
The shift from AI-powered analytics to AI that acts as an autonomous agent fundamentally changes how organizations approach data, governance, and accountability. When AI is used to generate insights, humans still remain at the decision gate; they validate, contextualize, and own the outcome. When an AI agent acts autonomously, that gate dissolves.
This age of autonomous AI agents demands a completely different infrastructure: data quality has to move from “good enough for analysis” to mission-critical; governance must shift from audit trails to active and online guardrails that constrain and correct agent behavior while it is running; and governance must be built into the foundational data architecture itself.
As AI agents gain access to enterprise systems, each agent introduces new questions around identity, permissions and accountability. How should organizations ensure that agents can access the information they need without exposing sensitive data beyond the task they are meant to perform?
Every enterprise wants AI’s speed and cost benefits, but no one can accept more risk to get them. That’s why risk and governance, not compute or talent, are enterprise AI’s real ceiling heading into 2027. The answer starts with treating every agent like a new member of the workforce. Each agent needs its own identity, access limited to only the data and systems its task requires, and a human owner who is accountable for what it does. Every action should be logged so teams can see what an agent accessed and why. As autonomous loops spread across every function, orchestration, observability and control (including a kill switch for every loop) move from nice-to-have to required infrastructure. The enterprises that build this governance into their platform today won’t just move faster. They’ll be the ones trusted to scale autonomous systems while others are still catching up.
Also Read: AITech365 Interview with Ron Zayas, Founder & CEO at Ironwall by Incogni
Gartner predicted in May 2026 that 40% of enterprises would demote or decommission autonomous AI agents by 2027 because of governance failures identified after production incidents. What are organizations most likely to overlook when moving AI agents from controlled pilots into live business environments?
Organizations often overlook how much context AI agents need to make sound decisions. An AI agent only has the knowledge you give it, and our research with Wakefield found that 77% of leaders say less than 20% of their data is contextualized enough for agents to use. Before allowing agents to act autonomously in live environments, organizations need to ensure they have the right context, governed data and guardrails in place, along with mechanisms for continuous learning. Until that foundation is established, agents can’t be trusted to act autonomously, and human accountability must remain central.
As AI spreads across teams and systems, traditional data governance practices face the challenge of keeping pace with how quickly agents access and process information. How can enterprises make privacy and access controls work in real time without slowing down useful AI applications?
Governance can’t be a checkpoint that agents wait in line for. It has to travel with the data. That means access policies, privacy rules and guardrails are attached to the data itself and checked automatically every time an agent makes a request, so the right controls apply in the moment without a manual review. Bringing AI to where data and knowledge already live also helps, because information doesn’t have to be copied into new systems where it’s harder to protect. When agents get consistent, ready-to-use context for each task, including the limits they must stay within, they can move quickly without putting sensitive data at risk. Done right, strong controls stop being a brake on AI and become what lets organizations scale it with confidence.
Your background spans distributed systems and large-scale infrastructure, where resilience has always mattered. As AI becomes part of critical business workflows, what should organizations rethink about data availability, recovery and continuity when an outage or an incorrect agent action can affect an entire chain of operations?
Organizations can no longer rely solely on traditional backup and failover approaches. When an agent error cascades across downstream operations, the blast radius extends beyond a single system. The real challenge: ensuring data governance, lineage tracking, and recovery mechanisms are built for autonomous workflows, not just human-driven ones. This means moving beyond point-in-time snapshots to real-time visibility into data quality and agent actions. It requires guardrails that prevent errors from propagating unchecked, combined with recovery frameworks fast enough to contain and reverse agent-driven changes. The organizations that win are those who embed governance, auditability, and controlled rollback into their data platform before deploying agents at scale. Continuity no longer means uptime alone, it means trustworthy autonomy.
Teradata’s Autonomous Knowledge Platform is designed to bring trusted data, analytics and AI together across cloud, on-premises and hybrid environments. For enterprises with sensitive data and strict operational requirements, what becomes possible when AI can work within the environments and controls their businesses already depend on?
The real opportunity is to move beyond AI pilots and put autonomous AI to work at enterprise scale. That requires more than powerful models: it requires trusted data, business context and governance built into the system from the start. When AI operates within the environments and controls enterprises already depend on, organizations can turn intelligence into action while maintaining control over their data, costs and operations. With the flexibility to run AI across cloud, on-premises and hybrid environments, enterprises can scale AI on their terms, meeting their security and regulatory requirements without compromising performance. That’s how you move from experimentation to reliable, production-grade AI.
You have led product and engineering teams at both an established technology company and a growing SaaS business. When AI capabilities are evolving quickly, how do you decide which security and governance safeguards must be built in from the beginning and where teams can afford to evolve their approach over time?
I start by architecting from the ground up, treating AI agents as a core design priority, not an add-on. Leading teams at both an established technology company and a fast-growing SaaS business taught me that the real question is what’s hard to undo. Anything that would be costly or impossible to fix after the fact needs to be built in from day one: agent identity and access controls, protections for sensitive data, audit trails and the ability to pause or stop an agent. What can evolve is what gets better with use, like the context agents rely on, the workflows they support and how much autonomy they’re given. Teams can start with tighter human oversight and expand autonomy as agents prove themselves. The goal is to govern autonomous action without operational trade-offs while letting AI take on the tasks it’s best built to perform.
As Teradata advances its AI capabilities, what do you believe will separate enterprises that confidently put AI agents into production from those that remain stuck in experimentation? What is the one change you would urge technology leaders to make now to prepare for secure, resilient and well-governed AI?
The enterprises that reach production will be the ones that treat context and governance as part of the architecture, not a project for later. Agents are only as good as the data and knowledge they can use, and only as trustworthy as the controls around them. Many organizations stuck in experimentation have capable models but lack the governed, contextualized data to let those models act safely. The one change I’d urge leaders to make now is to decide what expertise and data they need to protect, and build those protections into their AI foundation today. Legislation will lag far behind the technology, just as it did with data privacy. Leaders who wait for regulators may look back and say, “We should have protected that.” The ones who act now will be ready to scale AI with confidence.
Thanks, Sumeet!
About Sumeet Arora,
Sumeet Arora is the Chief Product Officer at Teradata, where he leads the company’s product organization, overseeing engineering, product management, and innovation strategy. With extensive experience in analytics, artificial intelligence, and cloud technology, he has a proven track record of building enterprise technology products and scaling high-growth businesses. Before joining Teradata, he served as Chief Development Officer at ThoughtSpot and held senior leadership positions at Cisco, including SVP and General Manager of Service Provider Networking.
About Teradata,
Teradata is an enterprise data and AI company that helps organizations transform data into actionable intelligence and business outcomes. Its platform supports enterprise analytics, artificial intelligence, and data management across cloud, on-premises, and hybrid environments. By combining trusted data, advanced analytics, AI capabilities, and enterprise-grade governance, Teradata enables businesses to accelerate decision-making, improve operational efficiency, and scale AI initiatives across their organizations.
