As companies push AI into production workflows, data access is becoming a more significant architectural constraint. 

According to the World Economic Forum, 82% of executives plan to adopt AI agents within the next 1-2 years. McKinsey found that 23% of organizations are already scaling an agentic AI system in at least one function, and another 39% are experimenting with AI agents. That points to a near-term shift in how enterprise systems consume data: from periodic, human-driven analysis to continuous, machine-driven interaction.

This changes the requirements for the data layer.

[CTA_MODULE]

Why traditional data architectures fall short

Most enterprise data architectures today were designed around analytics pipelines, batch processing, dashboards, and BI queries. In that model, latency is often tolerable, duplication is manageable, and access constraints can be worked around. Agentic AI introduces a different pattern. These systems require repeated access to current and historical data across operational and analytical environments. They depend on interoperability, consistent semantics, and reliable access paths across a growing set of tools. 

That makes data access more than a connectivity issue. The challenge is no longer simply whether an organization owns its data. The challenge is whether data can be accessed, moved, governed, and reused across systems without being constrained by platform-specific APIs, pricing models, closed storage formats, or tightly coupled compute layers. 

The growing constraints of platform-controlled data access

CIO recently reported that increased fees in Salesforce’s Connector program were beginning to affect software vendors, potentially raising integration costs and complicating customers’ AI plans. The article also notes concerns that tighter commercial and technical controls could narrow how customers choose to move and access their own data.

For a while, teams could work around those limits. They added more tools, copied data into more places, and built custom fixes to keep things moving. But agentic AI will put much more pressure on those weak points. 

A rate limit that is acceptable for dashboards can become a bottleneck for automated systems that need frequent retrieval or state synchronization. A platform-specific storage layer may work for a constrained analytics workflow, but create friction when teams need to support multiple engines, vector pipelines, retrieval systems, transactional applications, or model-serving environments. Repeatedly copying data across proprietary systems may preserve short-term access, but it increases storage costs, creates semantic drift, complicates lineage, and makes policy enforcement harder.

What agentic AI actually requires from data

These are not edge cases. Consider a support workflow where an AI agent needs product usage data, CRM history, and ticket context to resolve an issue in real time. Or a finance team using AI to monitor spend and flag anomalies across systems. Or an operations team relying on automation to trigger next steps based on inventory, customer activity, or supply chain changes. In each case, the technical requirement is not just data access. It’s access to the right data, at the right time, with the right controls in place.

This is where architectural control starts to matter in a much more practical way.

• Can data be accessed across systems without rebuilding pipelines for every new workload? 
• Can storage and compute be scaled independently? 
• Can multiple engines operate on the same governed data without forcing replication into proprietary environments? 
• Can policy enforcement, lineage, and business definitions remain consistent across analytics, operations, and AI systems? 
• Can cost and performance be managed as access patterns shift from batch consumption to continuous machine interaction?

If the answer is no, the issue is not just complexity. It’s a loss of flexibility. And that is the problem Open Data Infrastructure is meant to solve.

Introducing Open Data Infrastructure

Open Data Infrastructure is an architectural approach built on open standards, interoperable storage layers, and decoupled system design. Its goal is to keep data portable, addressable, and governable across tools and environments, rather than binding access to a single platform’s control plane. In practical terms, that means organizations can design for shared access patterns across analytical, operational, and AI workloads without making every new use case dependent on proprietary interfaces or redundant data movement.

In practice, that means data can be stored once and used in more than one place. Storage and compute do not have to be locked together. Teams can choose the right tools for the job instead of forcing every workload through the same path. And shared definitions, governance policies, and access controls can carry across systems, which becomes much more important when automated systems are acting on data, not just reading it.

Why this matters for agentic AI at scale

This matters because agentic AI increases the number of systems that need direct or near-direct interaction with trusted data. As those access patterns scale, tightly coupled architectures become harder to maintain. Every duplicated dataset introduces another synchronization problem. Every proprietary boundary introduces another control surface. Every disconnected semantic layer increases the risk that automated systems operate on inconsistent or stale information.

Open Data Infrastructure does not eliminate complexity. Distributed data systems still require strong governance, operational discipline, observability, and performance management. Open architectures also demand clearer ownership models and better metadata practices. But they offer a more durable foundation for change. They make it easier to introduce new processing engines, support new AI frameworks, and expand into new workloads without redesigning the stack around the assumptions of a single vendor or interface.

That flexibility is becoming more important as agentic AI matures. 

The future of data access and control

The organizations best prepared for the next phase of data and AI will not simply be the ones with more models in production. They will have a trusted data foundation — one that gives them the freedom to move faster without giving up control. They will be able to support new workloads without locking themselves into a single path. They will be better equipped to keep data trustworthy, definitions consistent, and access governed as AI becomes more deeply embedded in day-to-day work.

That is why Open Data Infrastructure matters for data access and control. Because as AI systems move closer to execution, orchestration, and decision support, it will not be enough to say your organization owns its data. The architecture will need to prove that data can actually be used — consistently, efficiently, and under the controls the business requires.

[CTA_MODULE]