Enterprise data layer AI gap shows agent ambitions outpace deployment

Enterprise data layer AI readiness is emerging as a constraint on scaling agents beyond isolated use cases, according to a Google Cloud and MIT Technology Review Insights report covered by CIO Dive. The survey of 300 global CIOs, CTOs and IT and data executives found that more than two-thirds expect to deploy AI agents within two years, yet only 10% said agentic AI was already in wide use across their businesses.

The gap is less about announced intent than whether an agent can reach enough reliable company information to make or support decisions. Surveyed organizations said AI currently accesses 45% of enterprise data on average. Only half said they trusted agent decisions to be relevant and accurate, a result the report linked to their organizations' data readiness.

What does an enterprise data layer need for AI agents?

For this purpose, the data layer is the access and control system between an agent and the information required for a workflow. The Google Cloud and MIT report said agents need relevant structured and unstructured information, business context, and access to operational systems, including supply-chain, point-of-sale and human-resources systems.

That framing makes the problem broader than adding documents to an AI search index. An agent acting across a workflow needs to distinguish among company-specific definitions, understand relationships among records, and operate under limits on what it can see or change. Metadata and provenance can help teams establish where information came from and how it relates to other data, although the survey does not test a particular architecture or product.

Governance is part of the production requirement, not a later compliance exercise. CIO Dive reported that scaling agents also requires changes to processes, workforce practices and governance policies. For teams handling access across operational systems, that work overlaps with an enterprise security program, which combines access controls, policies and operational processes rather than relying on one tool.

Why does the data layer become more important in production?

In a pilot, an AI output can be reviewed by a person and kept away from live systems. In sponsored best-practice commentary, Solutions Review and Denodo argued that production use raises the stakes because outputs may inform transactions or automated actions. They said source lineage, consistent business definitions, enforceable controls and predictable infrastructure costs become more consequential in that setting.

Vendor proposals differ on the label and implementation. Scale advocates an “AI-native data layer” that brings together structured, unstructured and multimodal data while retaining definitions, relationships and provenance. ArangoAI promotes a “contextual data layer” for managing shared business context. Those are product-adjacent viewpoints, not comparative evidence that either design is required.

Cloudera-sponsored research published through Forbes points in the same direction on governance pressure, while carrying important limits. The paid BrandVoice article said 42% of respondents identified security, governance and compliance as the leading reason for data-architecture changes, and 95% said they had delayed or canceled AI projects over governance, compliance or regulatory complexity. The excerpt does not include the methodology behind those figures.

For operators, the immediate readiness questions are measurable: how much relevant data an agent can access, whether it receives consistent context and provenance, whether policies apply across systems, and whether it can safely connect to the tools required to act. The available research supports the importance of those questions, but not a single prescribed data stack.

This story draws on original reporting from CIO Dive.