Why Agentic AI Needs a Data Control Plane
In our previous Newsletter editions, we explored why agentic AI systems struggle in production and what it takes to make them work reliably.
We showed that the problem is not reasoning. It is context.
Operational AI requires Precise Operational Context. Context must be scoped to the task, aligned to a specific entity, and governed before it reaches the AI agent. We then examined how entity-centric data products provide the foundation for delivering that context, by unifying, synchronizing, and governing data around the business entity.
This brings us to the final step.
How do you ensure that context is delivered correctly and that actions are executed safely at runtime?
In many implementations today, AI agents are responsible for everything. They determine what data to access, assemble context, apply logic, and execute actions across systems. While this approach may work in controlled environments, it breaks down in production.
When agents manage context and access themselves, behavior becomes inconsistent. Governance is applied unevenly. Security boundaries expand. Each agent effectively becomes its own integration layer, making systems difficult to control, audit, and scale.
This is not a limitation of LLMs.
It is an architectural issue.
Operational AI requires a clear separation between reasoning and control. AI agents should focus on interpreting intent and deciding what should happen. They should not be responsible for assembling context, enforcing policies, or orchestrating system interactions.
That responsibility belongs to a dedicated layer: the data control plane.
The data control plane sits between AI agents and the data foundation. It manages how governed, entity-scoped context is delivered to AI agents and how actions are executed against underlying systems. It evaluates permissions, enforces privacy and security policies, determines what context can be used, and ensures that only approved actions are performed.
Within this layer, data agents handle execution. They retrieve approved context from entity-centric data products, enforce runtime policies, and execute governed read and write actions. Every interaction is controlled and auditable.
Recommended by LinkedIn
This introduces a critical separation.
Entity-centric data products are governed at design time. They define the structure of the entity, embed ownership, lineage, masking, and allowed interfaces. They establish what context exists and how it can be used.
The data control plane enforces governance at runtime. It determines what is allowed for a specific request, at a specific moment, based on policies, permissions, and intent.
Together, they ensure that governance is not left to the AI agent and is not applied inconsistently. It is enforced deterministically before any context is delivered or any action is executed.
This changes how agentic systems behave in production.
Context is delivered consistently. Policies are enforced centrally. Access is constrained. Actions become reliable, correct, and repeatable.
Most importantly, the system becomes scalable.
Without a data control plane, every agent becomes a custom integration layer. With it, agents operate within a governed runtime environment that supports many agents, many tasks, and many operational workflows without losing control.
This is the final piece of the architecture.
Precise Operational Context defines what is required. Entity-centric data products provide the foundation. The data control plane ensures everything operates correctly at runtime.
Together, they enable agentic AI to move from demos to production.
In his latest article, Ronen Schwartz brings this full architecture together and explains why a data control plane is essential for running agentic AI systems at scale.