What is an agentic semantic layer?

An agentic semantic layer uses AI agents to build and maintain business definitions, instead of engineers writing every metric by hand. Agents scan source systems, propose how each one maps to a shared concept, and flag when sources change. People still approve the definitions. The result is a semantic layer that keeps up with the systems underneath it.

The term is new and used two ways: a semantic layer built by agents, and a semantic layer that agents read. The useful ones are both.

Traditional vs agentic

• How definitions are created. Traditional: engineers write each metric in code. Agentic: agents propose mappings from what they find in source systems.
• Who maintains it. Traditional: the data team. Agentic: agents do the first pass, and business owners approve.
• How it handles change. Traditional: someone notices a broken report. Agentic: agents flag source changes as they happen.
• Who reads it. Traditional: BI tools and analysts. Agentic: people and AI tools.

What agents should and shouldn't do

Agents are good at the slow work: finding every table that holds deposits or revenue, spotting that two fields mean the same thing, and noticing when a source schema changes.

They shouldn't approve definitions on their own. An agent can infer what a field probably means. Only an owner can say which definition the business stands behind, and that matters most for figures that reach a board or a regulator.

Questions to ask vendors

1. What do agents propose, and what do they decide on their own?
2. Who approves a definition before anything uses it?
3. How is each change recorded?
4. Which systems can the agents read today?
5. Can every figure be traced back to its source?

How Preql helps

Preql's agentic semantic modeling uses a Discovery Agent to scan your data sources and find how each system represents the same business concept, and a Translation Agent to turn those into shared business logic. Your team approves the mappings, every change is version-controlled, and agents monitor sources and surface changes that need attention.

FAQ

Is an agentic semantic layer the same as a context layer?

They overlap. A context layer adds ownership, approval history and cross-system mapping, which is what an agentic semantic layer needs to be trusted.

Does agentic mean no humans are involved?

No. Agents do the discovery and upkeep. People approve what the numbers mean.