Dominic Jainy is a seasoned IT expert at the intersection of artificial intelligence and data sovereignty. With deep roots in machine learning and distributed systems, he has spent his career dismantling the silos that prevent large-scale enterprise automation. In this discussion, we dive into the shift from human-paced data tools to machine-speed agentic management, exploring how modern platforms allow businesses to build AI directly where their data lives—whether that is in a legacy ERP or a modern cloud warehouse—without compromising security or control.
Many enterprises struggle to scale AI because regulated data cannot be moved into a central repository. How does bringing compute and governed access directly to the data change the development lifecycle, and what specific steps are taken to treat fragmented hybrid environments as a single estate?
Moving petabytes of regulated data is a logistical nightmare and a massive compliance risk that often stalls projects before they even begin. By bringing compute and governed access directly to the data, we effectively collapse the traditional development lifecycle; you are no longer waiting months for ETL pipelines to be approved, built, and audited. We treat these fragmented environments—on-premises, cloud, and even edge locations—as a single estate by using a unified runtime that bridges the infrastructure gaps. This allows an AI agent to query a local Hadoop cluster and a cloud-based instance simultaneously while the developer sees one cohesive workspace. It removes the friction of data gravity, ensuring that the information stays put while the intelligence travels to it.
Traditional tools for data quality and lineage were built for human-paced work, yet agentic AI requires machine-speed access. What are the primary technical hurdles when consolidating these governance layers into a single surface, and how does this integration improve the reliability of autonomous agents?
The fundamental friction in legacy systems comes from the “human-in-the-loop” bottleneck where data quality checks and lineage tracking are historically treated as batch processes. When an autonomous agent is operating at machine speed, it needs to verify the reliability of a data stream in milliseconds, not hours, to make accurate decisions. The primary technical hurdle is consolidating these disparate governance layers—catalogs, quality tools, and orchestrators—into a single governed surface that can respond at the same speed as the AI. By integrating these layers, we provide agents with a real-time “trust signal,” which drastically reduces hallucinations and improves reliability. It is the difference between an agent guessing if a data point is accurate and the agent knowing its entire history before using it.
Organizations often rely on a mix of legacy ERP systems, modern cloud warehouses like Snowflake, and open engines like Apache Spark. How do pre-built connectors ensure that data sovereignty rules remain enforced at runtime, and what metrics should teams track to verify governed access?
Enterprises are often paralyzed by the sheer diversity of their technology stack, which might range from decades-old ERP systems to modern platforms like Snowflake and Databricks. We utilize over 100 pre-built, governed connectors that act as secure, intelligent bridges, ensuring that data sovereignty is not just a policy written in a document but a technical reality enforced at the moment of execution. These connectors allow applications to draw on existing systems natively through engines like Apache Spark, Trino, or Kafka without ever moving the underlying records. Teams should track metrics like entitlement bypass attempts, lineage latency, and runtime policy violations to verify that access is truly governed. This creates a transparent audit trail that satisfies even the most stringent regulatory requirements while keeping the data firmly in its home territory.
Large companies in regulated industries frequently need to switch between different AI models based on cost or performance. How does a non-usage-metered cost model impact long-term scalability, and what is the process for maintaining workload consistency when moving agents between private and sovereign clouds?
Many organizations experience a form of “bill shock” when they attempt to scale AI because traditional usage-metered models essentially penalize a company for its own success. By adopting a non-usage-metered cost model, companies can finally forecast their AI spending with precision, which is a game-changer for long-term scalability across private and sovereign clouds. This financial predictability allows teams to focus on performance rather than penny-pinching, making it easier to maintain workload consistency when moving agents between different environments. We ensure consistency by using a single runtime across all infrastructures, allowing multiple, interchangeable AI models to run without rewriting the core application logic. It shifts the internal conversation from “can we afford to run this agent?” to “how much value can this agent generate today?”
Building AI applications from plain business requests requires high levels of trust in the underlying data infrastructure. Can you walk through the testing and deployment workflow for a new agent, and how do you ensure that lineage and entitlements are automatically applied without manual intervention?
The journey from a plain business request to a functional AI agent should feel seamless, but it requires a very rugged infrastructure to be successful. In our workflow, an agent is built and tested within a private software factory where data quality, lineage, and entitlements are baked into the execution logic from the very first line of code. During the testing phase, the system automatically simulates runtime conditions to ensure that the agent respects every localized data law and user permission without a developer having to manually code those restrictions. Once deployed, the runtime environment continues to enforce these rules automatically, providing a layer of protection that catches potential leaks before they happen. Seeing a complex agent go from a simple prompt to a fully governed deployment in a fraction of the traditional time provides a level of confidence that was simply impossible a few years ago.
What is your forecast for agentic data management?
I believe we are entering an era where data management is no longer a “back-office” function but the very nervous system of the enterprise. We will soon see a shift where autonomous agents become the primary consumers of data, outnumbering human users by a factor of ten to one. This means data infrastructure will evolve to be entirely self-healing and hyper-localized, where “sovereign-by-design” is the only acceptable standard. The organizations that thrive will be those that treat their data as a living, distributed asset rather than a static hoard locked in a central vault. In the coming years, the ability to deploy intelligence directly to the data source will be the primary differentiator between market leaders and those left behind by the complexity of their own silos.