Fivetran and dbt Labs introduce a context layer alongside dbt v2 and dbt State GA

What Fivetran and dbt Labs announced

Fivetran + dbt Labs announced at dbt Summit on September 16 that dbt v2 and dbt State are generally available. The company also introduced Fivetran Context Layer, dbt Charts, new dbt Wizard experiences and an open-lakehouse direction. The release frames these capabilities around an Open Data Infrastructure model in which storage, compute, data movement, transformation and visualization can remain independently selectable.

The distinction between generally available and introduced matters. The press release explicitly describes dbt v2 and dbt State as generally available while describing Context Layer and the other new experiences as debuts or introductions. Teams should verify current product availability and account access before treating every announced capability as a production dependency. The announcement is a primary vendor source, not independent evidence of business outcomes.

Sources: Fivetran + dbt Labs press release, September 16, 2026 · Fivetran Context Layer announcement

A context layer is only as strong as its definitions

For RevOps, the interesting idea is a governed layer that can expose business meaning to analytics and agents without asking every consumer to reinterpret raw tables. That can reduce repeated metric logic, but it also concentrates responsibility. If pipeline coverage, active customer, renewal date or expansion opportunity has several competing definitions, putting one definition in a context layer does not resolve the disagreement unless the business owner and change process are explicit.

Start with a small set of high-consequence entities and metrics. For each one, record the semantic name, business owner, source models, grain, filters, time basis, currency or unit rules, freshness expectation and known exclusions. An agent should receive the same contract a human analyst would need to reproduce the measure. Natural-language access should make governed definitions easier to consume, not make them less visible.

Lineage must survive the agent interface

A useful answer should preserve where the underlying facts came from and when they were valid. If an agent reports a renewal amount, forecast category or customer status, RevOps should be able to trace that statement to the governed model and ultimately to the source system that authored the commercial fact. Without that path, the context layer becomes another copy whose outputs are difficult to correct when the source changes.

Test lineage with deliberately conflicting records. Give the semantic layer a CRM opportunity, a billing record and a contract record that disagree on a date or amount. The system should either use the documented authority rule or expose the conflict; it should not silently choose the easiest value. The test is especially important before agent outputs are allowed to trigger CRM or customer-facing actions.

State-aware builds change the release conversation

dbt State is also relevant to RevOps because analytics changes are rarely isolated. A modified model can alter downstream dashboards, reverse-ETL audiences, forecast reviews or agent context even when the source application did not change. State-aware execution can help teams focus work on changed dependencies, but the operating process still needs to identify which business outputs are material enough to require review.

For a revenue model change, keep a release packet with the changed nodes, affected downstream metrics, representative before-and-after rows and the person who owns the business definition. Re-run critical tests on account identity, monetary fields, period boundaries and null handling. A faster build is useful; a faster unreviewed semantic change can still create a faster path to a wrong revenue decision.

Keep action authority outside the metric layer

A semantic or context layer can answer what a governed metric means and return a value. That does not automatically authorize a business action. A churn-risk measure may justify review without authorizing an offer. A renewal date may determine a queue without authorizing contract changes. Forecast context may help a manager investigate a deal without permitting an agent to rewrite stage or amount.

Define read, recommend and write authority separately from metric access. If an agent consumes context and then calls another system, preserve the context version, source evidence, proposed action and execution identity. This separation lets the data layer remain reusable across many consumers while action policy stays close to the operational system that owns the consequence.

What to validate before operational use

Pick three revenue questions that already have accepted definitions, such as open pipeline by region, upcoming renewals by authoritative date and active customers by billing status. Reproduce the existing answer through the new layer, compare row-level differences and document every expected discrepancy. Then test one changed definition and confirm downstream consumers can identify which version produced their result.

Finally, test failure: stale source data, an unavailable model, an ambiguous entity join and a deprecated field. The agent or application should expose missing confidence rather than continue with a polished answer that hides the gap. The Context Layer announcement is important because it moves business meaning closer to AI consumption. The operational value will depend on whether that meaning remains versioned, attributable and bounded by explicit action authority.