dbt’s semantic manifest is the best AI context nobody is using

What the semantic manifest actually is

Every dbt build produces artifacts, and the semantic manifest is the crown jewel: a JSON document describing your metrics, dimensions, entities, and their definitions — precisely, in your team’s own words, version-controlled and reviewed in pull requests. “Monthly recurring revenue,” “active customer,” “gross margin”: defined once, by the people who own them, in a format machines can read natively.

Compare that to what AI-on-data usually gets: raw table names and a prayer. The LG study we keep citing — 93% on simple questions, 4% on arithmetic reasoning — is what the prayer produces. Business-knowledge injection (giving the model your real definitions) is worth double-digit accuracy points by itself. dbt teams have that knowledge sitting in a file, already written.

The modern data stack’s strange blind spot

Here is the odd part. dbt teams obsess over metric consistency — one definition, everywhere — and then their company’s AI answers “what’s our MRR?” from raw tables with a definition it invented on the spot. The semantic layer solved consistency for dashboards and notebooks years ago; the AI channel is just late to the party. When the AI reads the same manifest everyone else reads, consistency extends to the one channel that was missing.

The pattern: manifest in, certified answers out

Read the manifest: metrics and dimensions flow in as candidate definitions — no rewriting, no re-derivation. Certify: the analytics team signs the ones that matter (this is a review of work they already did, not new work). Answer: business questions route through those certified definitions, with filters shown and an audit line attached. The raw warehouse stays where it belongs: behind the boundary, touched only by governed, scoped queries.

For dbt-native teams this is the fastest possible path to trustworthy AI on data, because the hard part — agreeing on what the numbers mean — was finished the day they merged their first metric PR.

DEBO connector status: P1, on the roadmap

Honest badge: the dbt connector (semantic manifest → certified definitions) is in the first connector batch. It is arguably the cleanest of them all — the manifest is already JSON, already reviewed, already versioned. dbt-native teams: early conversations shape it.