The semantic layer is dead. Long live the semantic layer!
dbt Labs' semantic layer product lead argues the first wave of semantic layers stalled because it only delivered the status quo, and that agents are now creating the value that makes a second wave stick, if the trust problem gets solved.
Overview
Explores how semantic layers evolved from BI infrastructure into core AI strategy via "context engineering"—defining what AI systems know and can trust. Covers MetricFlow and the dbt Semantic Layer's direction, the dbt MCP server turning governed metrics into agent-usable context, and fewer hallucinations, lower token costs, and consistent BI.
What you'll take away
- 01
The first semantic layer wave stalled because it only delivered the status quo; the second wave, driven by agents, offers something new: "a tireless analyst at the fingertips of anyone in the business."
- 02
Even a 92%-accurate agent degrades fast across a multi-step chain — the same math behind a 92.3% free-throw shooter's odds dropping to 45% on ten straight makes.
- 03
Four conditions for the second wave to succeed: close the BI integration gap with "semantic SQL," push semantics upstream into transformation, make the semantic layer the spine of a broader context layer, and scope agent access with full observability.
Session speakers
Zach Mandell
dbt LabsToby Mao
Fivetran