--- title: "Systems of Semantics: The Fourth Enterprise Data System" url: "https://atlan.com/know/ai-agent/data-for-ai/systems-of-semantics/" description: "Systems of semantics govern which meaning your AI agents use, for whom, and when. Here is why AI makes this the fourth enterprise data system type." author: "Emily Winks" author_role: "Data Governance Expert" published: "2026-07-17" updated: "2026-07-17T00:00:00.000Z" --- --- Your company runs SAP as its system of record and Snowflake and dbt as its [system of data](https://atlan.com/know/ai-agent/data-for-ai/systems-of-record-data-knowledge/), yet five AI agents each answer "what is our market share" differently: finance uses a volume-based measure, marketing a revenue-based one, sales a definition from a 2021 wiki page. Atlan's context layer is the governed system of semantics that certifies which definition applies, to whom, and delivers that rule to every agent before it answers, the same discipline behind [Gartner's prediction](https://www.gartner.com/en/newsroom/press-releases/2026-05-11-gartner-says-lack-of-semantics-causes-inaccurate-artificial-intelligence-agents-and-wasted-spending) that prioritizing semantics lifts agentic AI accuracy by up to 80% by 2027. --- | Category | The fourth enterprise data system type, beside systems of record, data, and knowledge | | :---- | :---- | | **What it manages** | Business meaning as a governed asset: canonical terms, definitions, ownership, policy, and links to data | | **Core problem it solves** | Multiple valid definitions of the same term need a governed rule for which applies, for whom, and when | | **Why AI forces it** | Agents need the rule in queryable form; scattered documentation repeats as ungoverned choices at scale | | **Adjacent categories** | Semantic layers align metric logic, ontologies define concepts, graphs connect relationships | | **Delivery to agents** | Governed definitions served through APIs and MCP, so every agent queries the same meaning | --- ## What the first three enterprise data systems cannot carry Enterprise data architecture adds a new system type when the existing ones cannot carry the next job. [Systems of record are authoritative sources for core business data](https://www.ibm.com/think/topics/system-of-record): they record orders, payments, contracts, and other facts of the business. Systems of data turn those records into models and metrics; the [metadata layer](https://atlan.com/know/metadata-layer-for-ai/) behind them tells an agent where a number came from, but not which of several valid meanings it should apply. None of the three manages meaning in a form a machine can consume at inference time. The data stack is shifting under AI See the 7 shifts reshaping data infrastructure for an AI-first world, including why governed meaning is now a system requirement. Download the 2026 Report ### Why were human consumers sufficient for the first three systems? Your SAP instance records transactions; Snowflake and dbt model them into tables; your knowledge wikis explain what the numbers mean. An analyst could read all three and supply the missing interpretation before a number reached a decision, the same interpretive gap that [institutional knowledge loss](https://atlan.com/know/data-for-ai/institutional-knowledge-loss/) leaves behind once that analyst moves on, and the same gap [human and agent data discovery](https://atlan.com/know/ai-agents-vs-humans-data-discovery/) close in very different ways. ### What job do none of the first three systems perform? An order can contribute to volume-based market share for finance and revenue-based share for marketing. A table built for one calculation leaves the agent missing the ownership rule, the same [data quality](https://atlan.com/know/data-for-ai/data-quality-for-ai-agent/) gap that produces a confidently wrong answer downstream. Your agents need the selection rule as a queryable record, with scope, owner, and purpose attached, the same [decision trace](https://atlan.com/know/what-are-decision-traces-for-ai-agents/) discipline proving which definition produced which answer. --- ## How does a system of semantics operate as enterprise infrastructure? A system of semantics manages business meaning as enterprise infrastructure: canonical definitions, disambiguation rules, business terms, metric specifications, ownership, and links to the data behind them, updating whenever a business rule changes, the same governed treatment [Atlan's context layer](https://atlan.com/know/atlan-context-layer-enterprise-memory/) gives every other category of enterprise context. The meaning becomes portable through APIs, [MCP](https://atlan.com/know/mcp/why-mcp-matters-for-ai-agents/), and standard interfaces, so every AI system queries the same governed version through the [context infrastructure](https://atlan.com/know/context-infrastructure-for-ai-agents/) layer above it. ### What distinguishes governing a definition from storing it? A knowledge wiki can store the explanation for a market-share definition. A system of semantics gives that definition an owner, validation history, usage policy, and links to the data it describes. Governance turns valid definitions into an authority your agent can use, the same distinction that separates [unstructured data](https://atlan.com/know/data-for-ai/unstructured-data-for-ai/) sitting in a wiki page from a governed, queryable record. --- ## Why does AI make semantics a system problem? Undefined meaning always carries a cost: conflicting dashboards, analyst debates, and slow onboarding for new hires learning unwritten conventions. A person usually resolved the ambiguity before it reached a decision; an analyst who saw two market-share numbers knew to ask which definition each one used. ### What changes when agents repeat ungoverned choices at scale? AI agents remove the human checkpoint at scale. Five agents across five functions each ground their answers in whichever definition they find, and repeat the same ungoverned choice on every query and every workflow that depends on it, one of the quieter paths to [AI agent hallucination](https://atlan.com/know/ai-agent-hallucination/): a confident answer built on an uncertified definition. This is one of the primary drivers behind [multi-agent memory silos](https://atlan.com/know/multi-agent-memory-silos/): each agent builds isolated context from the same ungoverned sources, the same fragmentation [context-aware AI agents](https://atlan.com/know/context-aware-ai-agents/) are built to resolve. Where's your semantic governance gap? Run the Context Gap Calculator to see which business terms your agents are still resolving without a governed rule. Calculate Your Gap --- ## How does a system of semantics differ from a semantic layer? [Semantic layers](https://atlan.com/know/semantic-layer/) are the closest adjacent category. The dbt Semantic Layer centralizes metric definitions in the modeling layer so downstream tools work from the same logic, giving your dashboards consistency by defining a metric once and querying it everywhere. A system of semantics has a wider job: it governs metrics, entities, policies, relationships, and business terms, and ownership travels with each definition as it changes. A semantic layer makes the revenue-based metric compute the same way in every dashboard; a system of semantics governs which meaning applies, for whom, and when, the [semantic layer for AI agents](https://atlan.com/know/ai-agent/semantic-layer-for-ai-agents/) treatment extended to policy and delivery. ### Where do ontologies and graphs stop and a system of semantics begin? [Ontologies and knowledge graphs](https://atlan.com/know/context-graph-vs-knowledge-graph/) are related but solve different jobs. An ontology represents knowledge about things and the relationships between them; a [knowledge graph](https://atlan.com/know/ai-agent/knowledge-graph-for-ai-agents/) adds real entities and relationships to that domain model. Ontology and graph structure help organize the market-share concept, but a system of semantics adds the certified definition, owner, validation history, approved use, policy, and delivery path for each agent to query, check, and apply at inference time. --- ## What does a system of semantics manage? A system of semantics manages the business meaning your agents need before they answer. For market share, the system records the term, its valid definitions, the authority behind each one, the policy for who can use it, the links to the data, and the interface that delivers it. | Semantic object | What the system records | Why your agent needs it | | :---- | :---- | :---- | | **Canonical term** | Market share, with its recognized variants across functions | Disambiguation across finance, sales, and marketing | | **Definitions** | Volume-based share, revenue-based share, share-of-wallet | The governed meaning for each function | | **Authority** | Owner, certifier, validation history | The approved owner and certifier | | **Policy** | Which definition applies by function and purpose | The rule each agent follows | | **Links to data** | SAP records, Snowflake and dbt models, knowledge wiki sources | The path from answer to source | | **Delivery** | API, MCP, semantic package | One governed interface for every agent | ### How does authority attach to a governed definition? Each recognized meaning carries its own owner, certifier, and validation history: finance owns the volume-based definition and certifies it for board reporting, sales owns share-of-wallet and certifies it for pipeline review. The policy turns those signals into a rule the agent applies automatically, without asking a person which version to trust. ### How does delivery make semantic objects portable across agents? [Links to data](https://atlan.com/know/what-are-decision-traces-for-ai-agents/) connect each definition to the systems behind it: the volume-based calculation uses SAP records, Snowflake tables, and dbt models, so an answer traces from the agent's response back through the definition to the source. Through an [MCP interface](https://atlan.com/know/mcp/why-mcp-matters-for-ai-agents/), any of your five agents queries the same governed definitions from one maintained semantic object, the same [context portability](https://atlan.com/know/ai-agent/context-portability/) requirement that keeps an agent's answers consistent across every system it touches. --- ## How Atlan operationalizes systems of semantics Atlan acts as the [governed context layer](https://atlan.com/know/what-is-the-enterprise-context-layer/) for systems of semantics, turning business meaning into versioned, certified context your agents can query. According to Atlan AI Labs research, agents receiving enriched semantic metadata achieve 38%+ higher query accuracy compared with agents querying bare schemas. [Context Engineering Studio](https://atlan.com/context-engineering-studio/) builds from your existing signals: SQL query history, BI dashboards, lineage, and glossary terms identify candidate definitions for review. For market share, the studio can show volume-based, revenue-based, and share-of-wallet definitions used across your teams before an agent selects one. The business glossary holds certified definitions, owners, validation history, and change history, so the agent can read who owns a definition, when it changed, and where it applies. [Context Repos](https://atlan.com/know/enterprise-ai-memory-layer/) make meaning portable. A repo packages definitions, access policies, and lineage as governed context: build the market-share semantic once, and any MCP-compatible agent, LLM, or internal framework draws from the same governed package. When the definition changes, the update reaches every agent using it, the same [talk-to-data](https://atlan.com/know/ai-agent/talk-to-data-agent-blueprint/) pattern any agent grounded in governed semantics depends on. What's the ROI of governed semantics? Run the Context Layer ROI Calculator to size the cost of agents guessing at business meaning instead of querying a governed rule. Calculate the ROI Open Semantic Interchange defines a shared format for semantic models across AI, BI, and data platforms. [Atlan is a launch partner in the vendor-neutral specification](https://www.snowflake.com/en/blog/open-semantic-interchanges-specs-finalized/), alongside Snowflake, dbt Labs, and Salesforce, so the same governed market-share definition can reach BI dashboards, AI agents, and data platform workflows through one open format. --- ## How enterprises made business meaning machine-readable for AI "All of the work that we did to get to a shared language amongst people at Workday can be leveraged by AI via Atlan's MCP server." Joe DosSantos, VP Enterprise Data & Analytics, Workday Watch Now "Atlan is much more than a catalog of catalogs. It's more of a context operating system. Atlan enabled us to easily activate metadata for everything from discovery in the marketplace to AI governance to data quality to an MCP server delivering context to AI models." Sridher Arumugham, Chief Data & Analytics Officer, DigiKey Watch Now Workday spent years building shared language across its teams; [Atlan's MCP server](https://atlan.com/know/mcp/why-mcp-matters-for-ai-agents/) made that semantic work available to AI at inference time, the same pattern behind [how enterprises use AI agents](https://atlan.com/know/ai-agent/how-enterprises-use-ai-agents/) against governed context rather than raw schemas. DigiKey describes the same layer as a context operating system spanning discovery, governance, quality, and agent inference, the same [enterprise memory](https://atlan.com/know/enterprise-ai-memory-layer/) discipline behind every governed definition an agent queries. --- ## Systems of semantics as the fourth enterprise data system type Each enterprise system type appeared when the previous set became insufficient for a new job. Systems of record made transactions reliable, data systems turned records into models and reports, and knowledge systems collected decisions and interpretation for people. AI agents put governed meaning in the execution path: your five agents need one governed market-share record at answer time, carrying the definition, purpose, certifier, policy, and delivery path each agent uses before it responds. [Context-aware agents](https://atlan.com/know/context-aware-ai-agents/) can only be built on governed semantic infrastructure. Market share is one term with three valid definitions across five functions, and the same problem appears wherever a business term carries different meanings by team, metric, region, product line, or decision type. As more agents depend on those meanings, informal coordination breaks down, the same [organizational cold start](https://atlan.com/know/ai-agent-cold-start-problem/) that shows up whenever an agent has no governed rule to inherit. Systems of semantics are the fourth enterprise data system type because governing meaning has crossed the point where your team can carry it informally: the first three systems hold your facts, your analysis, and your documentation, and the fourth governs your meaning in a form your agents can trust, the reason [agents need a shared enterprise context layer](https://atlan.com/know/why-ai-agents-need-an-enterprise-context-layer/) rather than four independent connections, following the same blueprint enterprises use to [implement an enterprise context layer](https://atlan.com/know/how-to-implement-enterprise-context-layer-for-ai/) across their data estate. Data lineage, feature stores, and [metadata management](https://atlan.com/know/ai-agent/data-for-ai/metadata-management-for-ai/) sit downstream of the same problem: an agent that traces a number back to its source still needs to know which definition that source certifies. A system of semantics is the rule that [data lineage for AI](https://atlan.com/know/ai-agent/data-for-ai/data-lineage-for-ai/) and [feature stores](https://atlan.com/know/ai-agent/data-for-ai/feature-store-for-ml/) both assume already exists. Book a Demo --- ## FAQs about systems of semantics ### 1. What is a system of semantics? A system of semantics is infrastructure that governs business meaning so AI agents answer from certified definitions. It records canonical terms, the valid definitions under each, the owner and certifier, the policy for which definition applies to which purpose, and the links to the data behind them. It manages meaning the way systems of record manage transactions and systems of data manage analysis. ### 2. How is a system of semantics different from a semantic layer? A semantic layer centralizes metric definitions so business intelligence tools return consistent results at query time. A system of semantics covers more than metrics, extending to entities, policies, relationships, and business terms. It attaches ownership and certification to each definition and serves them to both people and AI agents across the stack, well beyond BI consumption. ### 3. Is a system of semantics just a knowledge graph or an ontology? No. An ontology defines the concepts and relationships in a domain, and a knowledge graph connects those concepts to real data. Both are components a system of semantics uses. The system adds ownership, certification, validation history, usage policy, and a delivery interface, so an agent knows which definition is authoritative and can use it. ### 4. Why do AI agents need a system of semantics? Agents cannot infer an organization's disambiguation rules from scattered documentation. When five agents each read "market share" differently, each grounds its answer in whichever definition it finds and repeats that ungoverned choice on every query. A system of semantics gives agents one certified rule for which definition applies, for whom, and when, before they answer. ### 5. What does a system of semantics manage? It manages canonical business terms, the valid definitions under each, the authority that certifies them, the policy for which definition applies by function, the links from each definition to its source data, and the interface that delivers all of it to agents. --- ## Sources 1. [Gartner Says Lack of Semantics Causes Inaccurate AI Agents and Wasted Spending, Gartner (2026)](https://www.gartner.com/en/newsroom/press-releases/2026-05-11-gartner-says-lack-of-semantics-causes-inaccurate-artificial-intelligence-agents-and-wasted-spending) 2. [What Is a System of Record?, IBM (2026)](https://www.ibm.com/think/topics/system-of-record) 3. [dbt Semantic Layer, dbt Labs](https://docs.getdbt.com/docs/use-dbt-semantic-layer/dbt-sl) 4. [Web Ontology Language (OWL), W3C](https://www.w3.org/OWL/) 5. [Ontologies vs. Knowledge Graphs, Hedden Information Management (2024)](https://www.hedden-information.com/ontologies-vs-knowledge-graphs/) 6. [Open Semantic Interchange Specification Finalized, Snowflake (2026)](https://www.snowflake.com/en/blog/open-semantic-interchanges-specs-finalized/)