Lesson 6 of 10 · Building the context
Business definitions without waiting for a perfect semantic layer
Short answer
Agents need the same definitions people use. You can model them in a semantic layer, extract them from where they already live, or combine both.
By The Lineage Desk · Best Data Catalogs · Published
Desk research from public vendor pages. Editorial assessment.
2 min read
Ask an agent for last quarter's revenue and it will find a revenue column. Whether it matches the board deck depends on definitions: which accounts count, which currency, which date. This is where many agent pilots stall.
What does a semantic layer give you?
A semantic layer defines each metric once and serves it to every tool. Cube serves people and agents from the same governed model and includes an MCP server in its free plan. AtScale serves certified metrics by name to agents including Claude, ChatGPT and Microsoft Copilot. Honeydew offers one shared model for BI and AI analysts and processes data inside your warehouse. The platforms are adding their own: Databricks describes Unity Catalog Semantics as an open semantic layer with metric views, and Snowflake describes Horizon Context as a governed semantic layer.
Source: Cube pricing
Source: AtScale: semantic context for AI via MCP
Source: Honeydew homepage
Source: Databricks: Unity Catalog business semantics (read as a search summary)
Source: Snowflake: Horizon Context (read as a search summary)
What is the catch?
A semantic layer is only as complete as the model someone builds. Business logic that already lives in BI calculated fields, LookML or spreadsheets stays outside until it is migrated. AtScale describes a 30-day workshop from kickoff to results, which is a realistic sense of the effort for a governed model.
What is the extraction route?
The other route starts from where definitions already are. Euno says agents do not need to wait for the semantic layer: it extends AI access to business logic outside curated semantics, deep in BI workbooks, through lineage and SQL analysis. DataHub's Context Intelligence extracts metric definitions and join patterns from query logs, dbt projects and BI dashboards. Atlan's Metrics Generator and Semantic Views generate definitions from metadata. Extraction gets agents working sooner; the trade-off is that extracted definitions reflect what the stack does today, including its inconsistencies, until someone resolves them.
Source: Euno blog: AlphaSense x Euno at dbt Summit
Source: DataHub Cloud context platform
Source: Atlan homepage
Can you combine both?
Yes, and many teams will. Alation's Semantic Model Mastering ingests semantic models from Databricks and Snowflake, governs them and syncs them back. A context platform can point agents at certified semantic-layer metrics where they exist and at extracted definitions elsewhere, with labels that say which is which.
Source: Alation press release: AIOS expansion with six products
How do we score it?
Our business definitions criterion rewards platforms that hold or extract definitions agents can ask for by name. Semantic layers score highest; extraction-based platforms score well but lower. If consistent metrics are your main gap, raise this weight in the weight-your-own tool and watch the order change. Whichever route you take, test it with the metric your teams argue about most. Ask the agent for it three ways, by its business name, by a synonym and by the underlying column, and check that all three paths land on the same definition.
Key points
- Agents fail on definitions more often than on syntax.
- Semantic layers give governed metrics but need building; extraction starts from what exists.
- Combining both, with labels that say which is which, is a common end state.