Evolving data management: 3 ways to scale AI for the future

Build AI-ready data by shifting to knowledge repositories, interconnected metadata and a workforce designed for human-agent orchestration.

In brief

• AI at scale needs AI‑ready data: shift from static catalogs to knowledge repositories containing business rules, context and usage guidelines.
• Connect the dots with a semantic graph and serve logic, policy and data to humans and agents with an API control plane.
• Reskill and elevate roles for human-agent orchestration.

The rapid adoption of artificial intelligence (AI), including generative and agentic systems, is exposing the limitations of traditional, static approaches to data management. To remain competitive, organizations must shift from rigid data catalogs to dynamic knowledge repositories — systems that embed business logic, context and value propositions directly into their information product offerings. This evolution is not just technical; it’s a strategic imperative for building trust, agility and business value in an AI-driven world.

Reimagining data management requires foundational changes. Three key concepts include:

• Transitioning to knowledge repositories: moving from point-in-time data catalogs to knowledge repositories that codify business rules and decision logic for information products
• Advanced mechanisms for storing and serving metadata: establishing robust technical mechanisms for storing and serving metadata within the knowledge repository to enable agentic self-service
• Workforce transformation: preparing a workforce that natively leverages AI capabilities so that employees can effectively collaborate with AI systems

Embracing this pivot can enable organizations to realize the potential of their AI investments and enhance their ability to execute on more traditional reporting, analytical and modeling tasks.

Deep dive: How information products work

Storing and serving metadata to enable agentic self-service

Transforming disparate data assets into contextually intelligible building blocks necessitates a mechanism for storing and serving highly interconnected metadata inside the knowledge repository. This “semantic graph”:

• Formalizes a business ontology
• Maps business entities to data elements and objects
• Overlays a policy engine that enforces access compliance
• Operationalizes logic into machine-executable constructs 

The semantic graph is most naturally instantiated through a graph database, but other technologies can be used depending on client-specific use cases, workloads and skill sets. The semantic graph imbues data with axiomatic meaning, enabling agents to discern, for example, that a credit score of 580 implies “high-risk” and in accordance with organization lending guidelines requires VP approval to extend a loan.

With respect to the agentic AI domain, the semantic graph ensures agents consistently apply business logic when evaluating data, dynamically ensures compliance to ensure policies are enforced relative to agentic intent and data context, and provides causal auditability through lineage-traced explanations for autonomous actions. Without this semantic substrate, agentic systems often operationally devolve by misinterpreting siloed data schemas, violating latent regulatory constraints and propagating inconsistent decisions.

The semantic graph rectifies this by establishing a unified framework in which agents can:

• Query knowledge graphs to retrieve contextually grounded rules
• Have policy engines intercept non-compliant actions
• Assess the fitness-for-purpose of data through semantic metadata 

This extends data stewardship by providing infrastructure for trustworthy agentic operations and transforms governance from a passive constraint into an active enabler.

Traditional frameworks designed for human interpretation and batch processing prove inadequate for synthetic agents operating semi- or fully autonomously. Data and metadata must be served through interoperable interfaces that support real-time, cross-system access. These interfaces should be designed to expose metadata, the logic needed to act in business relevant ways and the underlying data itself.

Proper interfaces enable agents to support seamless, autonomous workflows by:

• Querying and acting on appropriate datasets
• Triggering actions
• Completing tasks and generating insights 

The knowledge repository’s control plane is instantiated through an application programming interface (API) suite and allows information products of all forms to be mapped to desirable business functions and consumed via composable services that are governed, compliant and well-documented.

This framework enables discovery and expression of business logic, access constraints and usage patterns in ways that both humans and AI agents can interpret and act upon. It also changes the passive information product consumption model to one that is dynamic and ready to support arbitrary decision-making and automation for a variety of organizational tasks.

The semantic graph and control plane within the knowledge repository facilitate the technical means to leverage information products by agents. As these foundations are laid, organizations must also rethink how teams interact with data and AI agents.

Organizational readiness and workforce of the future

GenAI and agentic AI capabilities radically reshape how organizations function and operate. New expectations around resourcing, decision-making, autonomy and collaboration are being formed because of AI, so a well-formed technical point-of-view is insufficient to fully adapt and optimize ROI. As enterprises evolve, so will their operational and organizational governance.

Anticipated changes to operational and organizational governance

Critically, companies must mitigate risks like skill atrophy, the over-reliance on eroding expertise and agent complacency in which humans automatically approve AI decisions. Success demands continuous reskilling, agent-free critical thinking exercises and ethical guardrails to ensure humans remain accountable for outcomes. The future workforce isn’t human vs. AI, but humans and agents evolving together.

Conclusion

Agentic AI systems require data that is not only accessible but interpretable, not only governed but responsive, and not only documented but contextualized. Meeting these demands means evolving from simple data catalogs to knowledge repositories of rich, machine-consumable metadata that describe the underlying information products. Because of the richness of this metadata, both humans and AI agents can leverage it to write sophisticated queries, perform reasoning and deploy autonomous workflows.

Context should be embedded throughout the data value chain and persisted in the knowledge repository’s semantic graph, aligning business rules and transformations to information products and their business value. The knowledge repository’s control plane helps facilitate data exchange, enabling information products to be used by both humans and machines with clarity and traceability. Finally, the workforce of the future natively leverages humans and synthetic agents working in concert to maximize return on technical investments while supporting employees in pursuing differentiated activities.

The journey toward AI-ready data does not begin with reinvention, but with evolution. The foundational capabilities in which organizations have invested over the past decade, such as modern data platforms, governance frameworks and data catalogs, remain essential as they serve as the scaffolding for trust, discoverability and control. These pillars help guide AI initiatives toward futures that are marked by growth and transformation as opposed to constraint or collapse.

Thank you to Amer Abed Rabbo, Abhishek Ghosh, Yevgeniya Virina, and Aditi Mukherjee for contributing to the development of this article.