Semantic Layer for Enterprise AI in 2026: What Production Use Requires

Quick Answer: 
Enterprise teams need three things from a semantic layer for AI in 2026: it has to be tool-agnostic, it has to enforce governance automatically across every connected tool and AI agent, and it has to be portable across clouds. Strategy Mosaic, the universal semantic layer, has been running this pattern in Fortune 500 production deployments, including Pfizer, Hilton, and GUESS, for years before agentic AI became its own category. 

Strategy Software, an enterprise software company built around a semantic layer that gives AI the business context it needs, has spent 35 years doing exactly this at Fortune 500 scale. The AtScale 2026 Semantic Layer Summit on May 20 shows the growing recognition of the importance of semantics and context as enterprises embrace AI. AtScale Co-Founder and CTO David Mariani recently framed the central question for the event this way: "the biggest hurdle for enterprise AI in 2026 is not the AI, it is the meaning." That diagnosis is correct. The open question this post takes on is what production deployment actually looks like once teams accept the diagnosis. Here are five production checks every enterprise team should run before committing to a semantic layer for AI workloads and why Strategy Mosaic closes the gap. 

Mariani is right that meaning and definitions are the hurdle. AI engines built directly on raw data sources cannot enforce business definitions consistently, do not understand fiscal calendars or multi-currency rollups, and have no governance model that survives moving between tools. The semantic layer is the right architectural answer to this problem. AtScale has been a real contributor to that case, including participating in the Open Semantic Interchange (OSI) standard, where Strategy is a co-founding member alongside Snowflake. 

A production-grade enterprise semantic layer is a governed, tool-agnostic abstraction layer that enforces a single set of business definitions across every BI tool, AI agent, and application, without moving data and without locking into any single cloud or vendor. Mosaic has been operating that pattern in large enterprise production deployments for over a decade, well before the term "agentic semantic layer" entered the category vocabulary. 

What separates conference framing from production reality is not the diagnosis. The difference shows up in the depth of the answer.

Use these five checks to evaluate any platform pitched against that framing. 

1. Multi-cloud portability. Does the semantic layer move with your data, or is it tied to a single cloud's compute layer? If your CFO consolidates spend onto a different cloud next year, does your semantic foundation come with you, or do you rebuild every metric definition? Mosaic runs across AWS, Azure, and GCP for the same enterprise customers, and metric definitions move without rewriting.
2. Tool-agnostic governance enforcement. Do governance policies apply automatically across every BI tool, AI agent, and application, or do they need to be configured per tool? The configuration-per-tool pattern is where governance silently breaks row-level security applied in Power BI does not automatically extend to a Tableau dashboard or a Claude agent reading the same data. Mosaic enforces row-level and column-level security at the semantic layer, so one policy set applies automatically across every connected tool.
3. Production AI agent depth. Has this been deployed against AI agents in production at enterprise scale, or only in demos? A platform that can answer one question reliably in a controlled walkthrough is not the same as one that has been governing thousands of agent queries per day for two years. Mosaic supports two-plus years of production AI agent deployments at customers including Pfizer, Hilton, and Lotte. 
4. Governance observability. Can your team see, in real time, who or what queried which metric, and detect PII exposure or anomalies? Mosaic Sentinel adds three real-time governance modules: risk management, audit and compliance, and usage insights, observable across every connected tool without requiring data movement.
5. Standards alignment. Is the semantic layer aligned with open standards, or is it a closed format? Mosaic supports the Model Context Protocol (MCP), Agent to Agent (A2A) and has a robust suite of APIs, so AI agents on Claude, Copilot, ChatGPT, Gemini, and AWS query the same governed semantics. Mosaic is part of the Open Semantic Interchange (OSI) co-founding group alongside Snowflake, Salesforce and Databricks.

Mosaic was built around the idea that a semantic layer should be the consistent foundation of every analytics and AI experience in the enterprise, not a per-tool feature inside one BI vendor's stack. That positioning has practical consequences worth naming directly. 

Mosaic connects to over 200 data sources natively, from ArcGIS through to Power BI DAX, without moving or replicating data. Mosaic Sentinel adds the governance layer on top, with real-time risk alerts, comprehensive audit visibility, and clear usage insights observable across every connected tool. Mosaic Sentinel does this without requiring data movement and without creating vendor lock-in. Mosaic supports MCP, which lets AI agents query the unified semantic layer directly without going through an intermediate agent, and AI Sync for Linked Models, which auto-validates and re-enables AI experiences when underlying semantic models update. 

Mosaic does not treat governance as a separate workflow that gets bolted onto AI later. Governance lives in the same layer that defines the metrics, and the same definitions that power dashboards power AI agents through MCP. That is what bridges the gap between pilot and production. 

See how Mosaic Sentinel monitors governance across every connected tool in real time. Request a Mosaic walkthrough scoped to your current cloud and BI footprint at strategy.com/software/strategymosaic.