Lesson 4 of 10 · Building the context
Lineage from the warehouse to the agent
Short answer
Agents need lineage that runs past the warehouse into transformation and BI, ideally at column level, because much business logic lives in dashboards and models.
By The Lineage Desk · Best Data Catalogs · Published
Desk research from public vendor pages. Editorial assessment.
3 min read
Lineage started as an impact-analysis tool: change this table, and see what breaks. For agents it does a second job. It lets an agent trace a number back to its sources and decide whether to trust it.
Why is table-level lineage not enough?
Table-level lineage tells you that a dashboard reads from a table. Column-level lineage tells you which source fields feed the field the agent is about to use. When an agent answers a question about churn, the difference matters: one column in a wide table may come from a certified model and the next from a spreadsheet someone loaded once.
Why does lineage need to reach BI?
A lot of business logic never reaches the warehouse. It sits in BI calculated fields, LookML and dashboard filters. If lineage stops at the warehouse, the agent sees raw columns and misses the definition the business actually uses. Euno documents column-level lineage from sources through BI dashboards to AI agents and says it extracts definitions that live in BI workbooks. Atlan lists BI tools such as Tableau, Power BI, Looker and Sigma among its connectors. DataHub builds lineage and impact analysis from query logs, dbt projects and BI dashboards. Select Star documents cross-platform column-level lineage.
Source: Euno: Context for AI Agents
Source: Atlan connectors
Source: DataHub Cloud context platform
Source: Select Star homepage
What does lineage look like in practice?
Euno publishes a case from AlphaSense in which a column-level trace from a workbook back to the warehouse took seconds, and a 287-field legacy reporting table with 12 dashboards and 68 dbt models was rebuilt as a 13-model star schema in about 15 prompts. They show the kind of task lineage enables for an agent: refactoring with knowledge of every downstream dependency.
How do you check a vendor's lineage claims?
Pick one metric your business argues about. Ask the vendor to show its lineage from the source columns through every transformation to the dashboard field and to what an agent receives. Note where the trace stops, where it falls back to table level, and whether BI calculated fields appear. Then rename a column in a development branch and see how long the change takes to show up.
What does our scoring reward?
The lineage criterion rewards lineage that is documented at column level across warehouse, transformation and BI, and that reaches the agent. A product that does not set out to map lineage, such as a semantic layer, scores low here by design, and can still be the right purchase if definitions are your main gap.
One practical warning: lineage quality depends on access. A platform that cannot read your BI tool's metadata API or your warehouse query history will show gaps that look like product weaknesses. Grant the access the vendor documents before you judge the result.
Key points
- Column-level lineage lets an agent trace a number back to trusted sources.
- Lineage must reach BI, because much business logic lives there.
- Test lineage on one disputed metric, end to end, before you buy.
Platforms in this lesson
Next lesson
5. Governance rules agents can read