Dominic Jainy stands at the forefront of the modern enterprise revolution, bringing decades of deep-seated expertise in artificial intelligence, machine learning, and the burgeoning world of decentralized ledger technologies. As an IT visionary, he has navigated the complex transition from simple data storage to the era of agentic intelligence, where autonomous systems are no longer a laboratory curiosity but a primary driver of corporate strategy. With a unique perspective that bridges the gap between raw compute power and the nuanced needs of business logic, Dominic has become a leading voice in explaining how organizations can finally move past the “pilot phase” of AI. In this conversation, we explore the tectonic shifts occurring in the data landscape as platforms evolve into autonomous control planes, looking specifically at the strategies required to build trust, scale workflows, and maintain governance in an increasingly automated world.
How does implementing a centralized business ontology prevent conflicting metrics between different AI agents, and why is this specifically the “missing link” for enterprise-grade intelligence?
The implementation of a centralized business ontology acts as a North Star for every autonomous agent operating within a corporate ecosystem, effectively silencing the internal noise that often leads to catastrophic decision-making. In the past, we suffered from the “which number is right” syndrome, where a sales agent might report revenue based on bookings while a finance agent reported it based on cash receipts, leading to a paralyzing lack of credibility. By establishing a single, preapproved, and traceable definition for every business term, we move away from the dangerous world of probabilistic model guesses where an AI essentially “hallucinates” a metric based on patterns rather than facts. This structural anchor ensures that when an agent queries a database, it isn’t just looking for a column header that looks similar; it is adhering to a human-approved semantic layer that provides an identical context across the entire organization. When we talk about moving agents into production, this consistency is the foundation of trust, as it allows leadership to trace every AI output back to a verified business rule, eliminating the friction of conflicting reports.
What does the transition from a passive data warehouse to an autonomous execution platform actually look like for a company’s daily operations?
This architectural transition represents a fundamental shift from a “storage-first” mindset to an “action-first” reality, effectively turning the data warehouse into the brain of the enterprise rather than just its memory. For years, teams have been bogged down by the heavy tax of extract, transform, and load—or ETL—maintenance, which consumed countless hours and introduced significant data latency that rendered insights obsolete by the time they reached a dashboard. By moving toward an autonomous execution platform, we are seeing the emergence of a control plane where agents can schedule repeated workflows and build their own data pipelines through tools like Pipeline Builder without constant human handholding. The sensory experience of this shift is palpable; you no longer have rooms full of data engineers firefighting broken connections, but rather a streamlined environment where the infrastructure itself governs model access and controls costs in real-time. It is the difference between having a library full of books you have to read yourself and having a staff of scholars who can instantly synthesize that knowledge and act on it to solve a problem.
In an era where every major vendor is offering a semantic layer, how does the openness of a catalog like Snowflake Horizon truly differentiate a company’s ability to avoid vendor lock-in?
The true battleground for enterprise lock-in has moved from who provides the compute power to who owns the customer’s metadata, security rules, and business context. While some competitors operate within closed ecosystems that treat foreign catalogs as read-only endpoints, an open, multi-engine catalog architecture built on standards like Apache Polaris and the Apache Iceberg REST Catalog API creates a much more fluid environment. This openness allows an organization to maintain a single, governed copy of their data that can be accessed by various engines, preventing the data from being “trapped” in a proprietary format that requires expensive egress or translation. When you look at the architecture, having the ability to leverage open-source capabilities means you aren’t beholden to a single vendor’s roadmap or pricing whims; you are building on a foundation that is designed for interoperability from the ground up. This flexibility is vital in 2026, as companies are increasingly hesitant to commit their most valuable intellectual property—their business logic and security rules—to a black box that doesn’t play well with others.
With statistics showing that 28% of organizations are seeing revenue gains from AI—up from 20% in January—what specific governance features are finally helping the remaining 72% bridge the trust gap?
The move from 20% to 28% in just a few months is a significant jump, yet the fact that nearly three-quarters of organizations are still struggling highlights a persistent trust gap that can only be closed through rigorous governance as infrastructure. Features like the Cortex AI Gateway are instrumental here because they treat security not as an afterthought or a separate layer, but as a core part of the agentic lifecycle. By implementing Agent Abuse Detection, companies can finally sleep at night knowing that an autonomous tool won’t overstep its permissions or access sensitive data it wasn’t meant to see. We are also seeing the rise of sandboxes, which provide a secure, isolated space to run custom code right next to proprietary data without the risk of a breach or uncontrolled cost overruns. This combination of granular control and cost transparency is what allows a skeptical CFO to sign off on a production rollout, moving the project out of the “theoretical” pilot phase and into the revenue-generating reality.
How does the introduction of bots that inherit employee access rights change the security profile of automated workflows, and how are “cloud sandboxes” mitigating that risk?
The ability for a bot to inherit an employee’s access rights is a double-edged sword; it provides the frictionless agility needed to act on a human’s behalf, but it also creates a massive security surface area if not managed correctly. To solve this, we are utilizing isolated cloud environments—essentially secure sandboxes—where agents like the CoCo coding agent can execute complex tasks without endangering the broader network. These sandboxes ensure that even if an agent performs an unexpected action, the impact is contained within a “bubble” that has no path to the core production systems unless specifically authorized. Furthermore, by letting users schedule these repeated workflows through automations, we move away from ad-hoc, manual bot triggering which is often where security protocols are bypassed in a rush to get results. The goal is to create a seamless experience where the bot feels like a natural extension of the team’s capability, while the underlying architecture provides a rigid, invisible safety net that prevents any single point of failure from becoming a catastrophe.
Despite the rapid progress we’ve seen this year, what are the most critical operational gaps that still prevent full, “lights-out” autonomy in mission-critical business processes?
While we have made incredible strides, we are still missing a few “holy grail” components, specifically automated feedback loops that can continuously retrain and refine agents based on real-world outcomes. Right now, many systems still require a “human in the loop” to verify that an agent’s reasoning remains sound over time, especially as market conditions shift or internal data evolves. Additionally, there is a persistent need for more robust master data management to resolve deep-seated issues within the source systems themselves before that data ever reaches the AI layer. If the source data is messy or contradictory at its origin, even the best business ontology in the world is just putting a fresh coat of paint on a crumbling foundation. Until we can achieve a state where agents can identify their own errors, suggest fixes to the underlying data, and learn from their mistakes without manual intervention, we will still need significant human handholding for the most vital workflows.
What is your forecast for the evolution of these agentic frameworks over the next few years?
I predict that by the end of 2028, we will stop talking about “AI agents” as a separate category of software and instead view them as the standard interface for all enterprise data interaction. We will see the “trust gap” narrow significantly as 50% or more of organizations report direct revenue gains, driven largely by the maturity of autonomous execution platforms that handle the “boring” work of data engineering and cost governance. The shift toward open standards like Apache Iceberg will become so absolute that proprietary data silos will be viewed as a legacy liability, much like on-premise servers were viewed a decade ago. Ultimately, the winners in this space will be the companies that treat their business definitions as their most valuable asset, using a centralized ontology to create a “digital twin” of their entire operations that an AI can navigate with the same precision as a human expert. We are moving toward a future where the data doesn’t just tell you what happened, but actively manages the processes to ensure the next result is exactly what you planned for.