Everyone wants AI, and they want it now—from the C-suite to the production floor. But are their data environments ready to support a full-bore AI effort? Data industry leaders and experts we spoke with say there has been a lot of progress fulfilling AI needs, but there’s still a lot of work that needs to be done before today’s data platforms are capable of supporting robust, AI-driven organizations.
DEFINING A ‘MODERN’ DATA PLATFORM: WHAT’S NEW?
What constitutes an AI-ready modern data platform? “They share several defining traits: cloud-native architectures that are simple to scale, an open approach to support both lakehouse construction and metadata management, and the ability to unify analytical, streaming, and AI workloads over a single, governed data layer,” said Ed Beauvais, director of product management, AI, and cloud data infrastructure at HPE.
For a truly AI-ready data platform, AI has to be more than an add-on. These platforms need to evolve into knowledge platforms—those “that don’t just store and process data but actively deliver context and meaning to AI systems that need to reason and act,” said Sumeet Arora, chief product officer at Teradata. This means combining structured business data with multimodal unstructured data such as documents, emails, and videos, Arora added.
Such platforms are becoming increasingly autonomous, Arora noted. Just as critically, they require an “ability to support agentic outcomes without requiring data to be moved.” Rather than surfacing raw data, “Modern platforms deliver knowledge and context directly to agents at the point of need, eliminating the cost, latency, and risk of data movement.”
These platforms also need to support data readiness—“particularly for AI, database observability and monitoring, security, automated governance, and multi-platform consolidations,” said Graham McMillan, CTO at Redgate Software. Database platforms are shifting to meet new demands for security and compliance, he said.
MISSING INGREDIENTS
Building and sustaining an AI-ready data platform requires more than simply adding new features—much more, Mark Hiew, senior product marketing manager at SAS, emphasized. “In reality, the fundamental ingredient that is often missing is the ability to modernize and actively govern a sprawling data estate that has grown faster than operational discipline has kept up.”
Historically, data estates have lacked rhyme or reason—they have sprung up organically as collections of data environments through the ages. “Cloud platforms, legacy systems, data lakes, warehouses, and analytics environments now coexist with little shared visibility,” said Hiew. “Many organizations cannot clearly answer basic questions: ‘Where does our data live?’ ‘Which assets are actually used?’ ‘What does it cost to maintain them, and which data can be trusted for analytics and AI?’”
Metadata that helps both people and systems get at the information they require also needs to be part of the picture. Without strong metadata, lineage, and observability, “modernization efforts stall or drift into perpetual migration cycles,” said Hiew.
Industry experts also pointed to a range of ingredients needed to build and sustain an AI-ready data platform:
- Governance: This is perhaps the most essential ingredient needed by modern data platforms. “Trust and governance—as agents take on more autonomous roles, lineage, policy enforcement, and auditability become mission-critical, and most organizations aren’t there yet,” said Arora. Demand for AI adoption “is outpacing the demand for governance,” said McMillan. “In essence, organizations need tools to control who is accessing the data, understanding what data they have, and ensuring they’re meeting their legal obligations.” Much of this governance and related enforcement is still either manual, inconsistent, or siloed by platform, said Hiew. Plus, it needs to “operate as part of day-to-day data operations and not as a downstream compliance exercise if organizations want to scale analytics responsibly.”
- Standardization: Most of the missing ingredients in modern databases are operational in nature, said Ryan McCurdy, VP of marketing at Liquibase. “Teams have powerful platforms, but inconsistent standards around how data is defined, changed, validated, and audited. Many organizations still treat governance as documentation instead of controls embedded in delivery workflows. The result is predictable: ‘Variance at the foundation multiplies at the top.’”
- Cost optimization: “Flexibility without control leads to duplication, idle assets, and runaway consumption,” said Hiew. “Too many organizations modernize infrastructure without modernizing how they prioritize, monitor, and optimize usage across the estate. The result is a more complex environment that is harder, not easier, to manage.”
- Automated testing and deployment: A lack of automated testing and deployment is another missing ingredient, said McMillan. “Many organizations still rely on manual processes to test and deploy database changes, which can cause huge slowdowns in the development process.”
- Data quality: Another gap seen is “the layer that sits between the cloud data infrastructure and the AI systems that consume it,” said Jay Limburn, chief product officer at Ataccama. “This gap keeps appearing, almost universally. The piece missing in between is what actually certifies data before an agent is permitted to act on it.” At issue, Limburn explained, is that “data being available in a cloud environment is not the same thing as data being trustworthy. Source systems accumulate quality problems over time: duplicate identity records across channels and acquisitions, stale reference values, fields populated with placeholder defaults from old migration scripts, schema definitions that have drifted in meaning since they were first established.” When this data lands in a centralized platform, “those problems do not disappear,” he continued. “They become more accessible, and once AI systems start acting on them, they stop being data management issues and start being operational incidents."
- Observability: The gap also presents as fragmentation across the IT lifecycle, said Scott Ward, chief business officer at Compucom. “Many organizations have invested in modern tools, but infrastructure monitoring, endpoint analytics, security telemetry, asset management, and service management platforms often operate in isolation. Without unified observability and lifecycle integration, leadership teams are still forced to make decisions with incomplete context.” Data managers often assume that validation and certification work “is already handled somewhere in their architecture, but it often isn’t,” said Limburn. “They had cataloguing, which tells you where data came from. They had lineage, which traces how it moved. What they were missing was the layer that validates records, resolves identities, enforces business definitions, and produces a signal that says whether a specific dataset meets the threshold for automated execution. That signal needs to exist in a form that orchestration systems can actually read and act on, not just in a dashboard someone checks periodically.” This is also complicated by a dearth of internal best-practice sharing, McMillan added. “Most teams lack a process to consolidate database best practices, leading to human error and security issues.”
- Business context: Things also need to be context-aware. Foremost is “the ability to deliver trusted, use-case-aware context to agents,” Arora explained. “Most organizations have data but lack the infrastructure to make it meaningful to AI at the moment of need,” said Arora. This also relates to a lack of outcome alignment to the business. “Technical dashboards frequently emphasize system metrics—uptime, latency, incident volume—without connecting those indicators to workforce productivity, customer experience, or financial impact,” said Ward. He added, “While companies collect vast amounts of telemetry, many lack embedded workflow orchestration and governance guardrails that allow insights to trigger consistent, policy-driven action. Predictive analytics without integrated execution still leaves organizations reactive.”