We’re finding out that context is everything when it comes to successful enterprise AI deployments. Removing ambiguity, and working around agreed definitions and vocabularies are essential as agentic AI starts to become more autonomous. At their recent data and analytics summit, Gartner predicted that by 2030, USLs will be treated as critical infrastructure alongside data platforms and cybersecurity, and that 44% of data and analytics leaders have already implemented semantic layers, with a further 48% planning to by 2027.
But CIOs have been here before. Data fabric, data mesh, lakehouses, and active metadata were each promoted as critical enterprise infrastructure on previous Gartner cycles, and each has since been absorbed into adjacent products, or quietly forgotten. The semantic layer is a more concrete capability than any of those, but before treating predictions as a buying signal, it’s worth pressure-testing the claim.
What a true semantic layer looks like
Ben Clinch, leading AI, data, and architecture community advocate at the EDM Association and DAMA-UK, sees the strategic value of trusted semantic layers.
“Well-crafted semantic layers bring a level of context and intention that empower AI, processes, and people,” he says. “Enterprises of a significant scale often have a diverse set of data stores and systems that can’t easily interact or share consistent meaning without considerable complexity and expenditure. True semantic layers unify these data stores and their meaning in a powerful network effect.”
The key here is Ben’s emphasis on well-crafted and true, which set a high bar that most enterprise implementations have historically struggled to clear. Technology is rarely the hardest hurdle. Organizations often have conflicting definitions as to what counts as revenue or a customer. Cross-functional alignment is essential and is where most semantic layer initiatives stall, long before the architecture is drawn up.
The vendor land-grab
Every major data platform has spent the past 18 months reframing itself around a semantic layer for AI. Microsoft’s Fabric IQ, launched at Ignite in November 2025, is positioned as the semantic foundation for enterprise AI. Databricks shipped Unity Catalog Metric Views and wired its Genie agent directly into them. Snowflake’s Cortex Analyst sits on native Semantic Views, Salesforce launched Tableau Semantics to feed Agentforce, and dbt Labs open-sourced MetricFlow at Coalesce 2025 to power agentic workflows.
But through their experience specifying and deploying large data platforms, CIOs will understand that universal is doing a lot of work in the marketing of these vendors’ offerings. Most of these products are BI-era semantic models with agentic veneers, and originally built to feed human-readable dashboards, not provide the dynamic, real-time context that autonomous agents demand. A semantic layer designed to power a clean Tableau dashboard will likely buckle under the unpredictable, non-linear requests of an agent stack.
The Open Semantic Interchange (OSI), launched in September 2025 by Snowflake, Salesforce, dbt, BlackRock, Alation, and others has the potential to address these issues. Yet OSI is still in early development, and major players including Microsoft and SAP haven’t signed up. The standards war is a long way from being won.
Why pilots don’t scale
McKinsey’s 2026 research on agentic AI tells an increasingly familiar story about AI deployment roadblocks. Nearly two-thirds of enterprises have piloted agents, but fewer than 10% have scaled them, and around 80% cite data limitations as the core problem.
The constraint is not raw data availability, as large enterprises have more data than they can use. It’s the absence of shared meaning across it, which is exactly the gap a semantic layer can solve. Enterprises are rightly nervous about exposing that context to autonomous agents without semantic-aware guardrails. This is the problem with the BI-era semantic layer underneath most pilots, as it was designed for a different job entirely.
What CIOs should demand
Should CIOs buy the universal pitch? Not yet, but they should invest in metric governance and one well-defined semantic model where AI is touching customers or revenue. The main problem is overcoming vested interests within the enterprise, and focusing on the customer, the most important element for any business, will help build a strong foundation for the future.
Also, demand OSI alignment from vendors. Portability is the only thing that turns a feature into infrastructure, and it’s the easiest thing to lose if you don’t ask for it up front.
Then apply Clinch’s test to whatever product you’re considering. Does it actually unify meaning across your diverse data estate, or does it only generate network effects within a single platform?
Gartner’s 2030 prediction may well come true, but only if the OSI effort succeeds, the hyperscaler land-grab is constrained, and enterprises do the unglamorous organizational work of agreeing what their data means. The test, in the meantime, is the one Clinch sets, which asks if the layer in front of you carries context and intention across the estate. So treat universal as a forecast, not yet a fact.