Semantics Is Not a Layer. It’s the Engine.

At the Gartner Data & Analytics Summit in London on 11 May 2026, Rita Sallam, Distinguished VP Analyst at Gartner, delivered a message the industry has been circling for years: AI agents fail not because the models are wrong, but because the data they operate on lacks meaning.

Gartner’s prediction is direct: by 2027, organizations that prioritize semantics in AI-ready data will improve agentic AI accuracy by up to 80% and reduce costs by up to 60%. Organizations that neglect it face wasted spending, hallucinations, bias, and governance failures.

“Without context — a clear understanding of the specific relationships and rules within an organization’s data — AI agents cannot operate accurately and are far more likely to hallucinate, introduce bias and produce unreliable results.”
— Rita Sallam, Distinguished VP Analyst, Gartner

We agree. Completely.

But we want to add something important: a “semantic layer” bolted onto existing infrastructure is not the answer.

What Gartner Got Right

The core insight is correct and timely. AI agents fail when they lack business context. When an agent is asked “is this customer high-risk under our policy?”, the answer depends on a precise set of rules specific to your organization. No general-purpose LLM carries that knowledge. It cannot be embedded in a system prompt. It lives in your data, your logic, your policies — and it is almost never written down in any form a model can reliably interpret.

The problem compounds across every step of an agentic workflow. Each step where context is absent or ambiguous is a step where the model guesses. And in regulated, complex, or high-stakes domains — finance, pharma, insurance, manufacturing — guesses are not acceptable.

Gartner’s call for organizations to treat semantics as a core infrastructure component, not an afterthought, is exactly right. The 60% cost reduction prediction follows directly from this logic: every deterministic step — one that does not require a model to interpret or infer — is a step that does not incur token costs. Semantics is a cost-control strategy, not just a quality strategy.

Where the Picture Is Incomplete

The challenge with the “semantic layer” framing is that it positions semantics as a translation service sitting on top of existing infrastructure. A semantic layer can enrich a query. It can map a business concept to a database column. It can provide a glossary.

What it cannot do is execute.

At runtime, the operational problems remain. Legacy infrastructure was not built to reason. A semantic mapping does not change the fact that your data lives across incompatible schemas in six different systems. A glossary does not prevent an agent from hallucinating when it encounters edge cases that fall outside the documented definitions.

Semantics as a layer treats the symptom. It does not treat the cause.

Business logic needs to be executable by both humans and AI. It needs native infrastructure where semantics are the engine, not a post-hoc addition.

The Operational Ontology: Semantics That Runs

An operational ontology is the architectural answer to Gartner’s challenge.

Where a semantic layer sits above your data and translates between business concepts and queries, an operational ontology processes data directly through the lens of business logic. It does not describe what “customer” means — it executes the rules that determine whether a specific customer, under your specific policy, meets a specific condition. Against live data. Deterministically. With full lineage.

This is what makes the Gartner cost savings actually achievable:

• Deterministic steps replace probabilistic ones. When an AI agent can execute a policy rule directly — rather than asking a model to interpret and approximate it — the result is exact, auditable, and costs a fraction of what an LLM inference would.
• Data stays where it lives. The Prometheux Engine processes data across warehouses, databases, and cloud platforms without migration or duplication. The semantic logic travels to the data, not the other way around.
• Lineage is native, not reconstructed. Every conclusion the engine draws carries a complete trace: the inputs considered, the rules applied, and the path to the output. In regulated environments, this is not a nice-to-have — it is the difference between a decision that can be explained and one that cannot.

Natively Neurosymbolic

Prometheux is natively neurosymbolic. This matters because the challenge Gartner describes is precisely the gap between two modes of intelligence that most AI stacks only partially address:

• The PX Engine handles deterministic logic and operational reasoning — the rules, relationships, and inferences that define your business.
• LLMs handle language and intuition — natural language understanding, summarization, generation, and the interpretation of unstructured inputs.

Agents need both. Most stacks only provide one. Prometheux is natively both.

When an enterprise asks whether a client relationship creates systemic risk, whether a manufacturing batch violates a compliance rule, or whether a patient should be flagged for adverse event review — these are not questions for a language model alone. They are questions that require deterministic execution of business logic over real data. The LLM defines the question and presents the answer. The PX Engine executes the reasoning.

The Cost Savings Are Real — But Only with the Right Architecture

Gartner’s 60% cost reduction figure reflects a genuine architectural opportunity. Enterprises today route almost everything through LLMs because they have no other mechanism for applying business logic at runtime. Every step becomes a probabilistic inference. Every inference has a token cost. Every token cost compounds across millions of agent invocations.

Operational ontologies change this. Enterprises can decide exactly which steps require the interpretive power of an LLM — and which steps can run autonomously on deterministic, fast, and cheap compute. This split is modular and can be adjusted as priorities and costs evolve.

Full spending control. Full governance. Without sacrificing capability.

Semantics as Operating Model

Gartner is right that “context with semantic coherence will become a cost-control and trust strategy, not a nice-to-have.”

But context and semantics cannot be a layer. They need to be embedded into the operating model — as operational ontologies that define, execute, and audit business logic natively, wherever your data lives.

That is the enterprise of the future. One that runs autonomously. On operational ontologies. On Prometheux.

Read the full Gartner press release →

Book a Demo →