Dominic Jainy is a prominent figure in the evolving landscape of enterprise technology, recognized for his deep expertise in integrating artificial intelligence, machine learning, and blockchain into robust business frameworks. As organizations struggle to move AI from experimental stages into reliable, governed production, Jainy’s insights into data intelligence have become essential for leaders navigating the complexities of 2026. This conversation explores the shift toward Intelligence Operating Systems and the critical need for explicit context in a world increasingly run by agentic AI. We explore why the industry is currently facing a massive “governance gap,” the mechanics of cross-platform registries, and why the future of AI hinges on treating data governance as a systemic foundation rather than an afterthought.
A recent survey of 950 C-suite leaders revealed that 78 percent lack the confidence to pass an independent AI governance audit within the next 90 days. Why is the industry struggling so much with this, and what is the real cost of failing to provide a governed foundation for AI?
The primary struggle isn’t actually about the sophistication of the models themselves; it is fundamentally a context crisis. When you look at the leaders surveyed, the common thread is that the governed data and definitions that these models rely on simply do not exist in a usable format. Without documented lineage or a trusted source of truth, AI agents are essentially left to guess what the data means, which leads to “confidently wrong” outputs that can devastate a finance team during a month-end close. For a compliance function facing an audit, an answer built on bad data or inferred context is the most expensive kind of failure possible because it invites regulatory scrutiny and erodes organizational trust. We have to move past treating governance as a “bolt-on” feature and start building it directly into the way data is discovered and utilized by every agent in the system.
We often hear that AI success is a “capabilities problem,” but many experts argue it’s actually a “systems problem.” How should developers rethink the design of agentic applications to ensure they aren’t operating in a vacuum?
The idea that you can be “one and done” with agentic AI is a dangerous myth that many organizations are finally unlearning as they hit the limits of their initial deployments. To get AI right, you must view it as a living system that requires constant observation and monitoring to learn from errors, human feedback, and new knowledge. Developers need to design applications—whether they are simple dashboards or complex automations—to constantly refine their context through systemic learning loops involving both humans and agents. By building governance into the foundation, organizations can gather context faster and more accurately, ensuring that every action an agent takes is compliant and safe. This shifts the focus from the compute layer to a more holistic view where business meaning is explicitly defined and constantly updated as the business evolves.
With the rise of multi-platform environments involving various providers, how can a centralized registry manage the regulatory risk of agents consuming live data across these different ecosystems?
Managing risk across a fragmented ecosystem is virtually impossible without a cross-platform registry that connects an agent’s compliance posture to the live quality of its data. Currently, we are seeing the implementation of native connectors for six major platforms, including Snowflake Cortex, Databricks MLflow, and Amazon Bedrock, which allow for a unified view of the entire AI landscape. This allows for precise agent lineage tracing, giving teams visibility into exactly which agent touched which data under what specific policy. It is a massive leap forward to be able to see the live status of underlying data and how it impacts the regulatory risk of an agent in real-time. This level of transparency is what finally allows AI programs to move from experimental stages to full-scale production without stalling due to compliance fears.
Large amounts of critical business context live in unstructured formats like SharePoint documents or S3 buckets. How can we transform these “dark” data sources into governed objects that agents can consume reliably?
This is where governed collections become essential, as they allow us to take unstructured context—like compliance policies, business glossaries, and standard operating procedures—and turn them into catalog objects. By referencing documents in SharePoint, S3, or Confluence, agents can ensure they are always reading the most up-to-date version of a policy without any manual intervention. When a source document changes, the agent reads the current version automatically, which prevents the “stale data” problem that often leads to incorrect decision-making. This process takes the business meaning that was previously hidden in PDFs or internal wikis and makes it an active, governed part of the AI’s runtime environment. It effectively bridges the gap between static documentation and active execution, ensuring the agent remains anchored in reality.
One of the biggest hurdles for AI is understanding the “business meaning” behind data, such as a consistent definition for an “active customer.” How do ontologies and semantic model mastering help agents avoid making incorrect inferences?
Agents are only as reliable as their understanding of the business, and without explicit rules, they will try to infer constraints—often getting them wrong in very subtle, damaging ways. Ontologies provide a machine-readable model of how a business actually works, mapped directly to the data catalog and reviewed by subject-matter experts to ensure total precision. For example, by mastering semantic models from platforms like Snowflake or Databricks, we can resolve a term like “active customer” to a single, enriched definition that is synced back across all platforms automatically. Every downstream consumer and agent then runs on that same high-quality definition, ensuring that nobody has to re-enter data or question the source of a metric. This moves us from a world of inferred, “confidently wrong” guesses to one where the business logic is explicitly mapped and reviewed at the source.
What is your forecast for enterprise AI governance?
I expect that by the end of 2026, the era of unmonitored AI agents will be largely over, replaced by a new standard where governance is an automated, runtime requirement rather than a manual audit trail. Organizations will stop asking if they have the right models and start asking if they have a robust Intelligence Operating System to manage the interplay between humans, agents, and data. We will see a massive consolidation of data intelligence, where tools like Intelligent Feeds find the users and provide the exact narrative and visualizations needed for recurring decisions without the user ever having to go searching. Ultimately, the winners will be the companies that treat data context as their most valuable asset, ensuring that every agent in their fleet is anchored to a single, governed version of the truth that is shared across the entire enterprise.