Fivetran and dbt Labs package state, context and agents for the post-merger data stack

Fivetran and dbt Labs used their first dbt Summit since merging to frame an “agent-ready” data stack, but the September 16 announcement describes products at sharply different stages. dbt v2 and dbt State are generally available. Fivetran Context Layer is in private beta; dbt Wizard in the dbt platform and Wizard Explore Mode are in public preview; Wizard CLI is in public beta; and Wizard Desktop is in private beta.

That availability split is the practical headline. dbt v2 is a Rust rewrite of the dbt engine. Fivetran says it can parse a 10,000-model project up to ten times faster than v1 and can surface errors, column checks and lineage before execution. dbt State compares warehouse metadata with model SQL, then decides whether each run should build, skip, clone or defer work. Fivetran positions that mechanism as a way to simplify orchestration and avoid unnecessary warehouse compute.

The primary source also supplies two customer results, both of which should be read as vendor-selected examples rather than independent benchmarks. Virgin Media O2 reports 25% savings in both job runtime and BigQuery compute costs. RxBenefits reports a 59% reduction in warehouse costs on scheduled dbt-platform jobs during its first 60 days, with 716,000 model executions reused instead of rebuilt.

The context layer is not generally available

Fivetran Context Layer is a private-beta service that combines dbt’s structured project context with unstructured material such as documentation and Slack threads. The announcement says it uses Agents Schema, an open-source format stored in the data warehouse, and makes context available through MCP or AI tools. That is a narrower and more inspectable claim than treating the context layer as a finished universal interface.

Wizard’s availability is similarly segmented. The dbt-platform experience is in public preview, while its terminal and desktop variants are respectively in public beta and private beta. The announcement says Wizard connects natively to a dbt project, retrieves project context and validates changes before they ship.

A separate dbt Wizard product post describes the context it can load: models, descriptions, lineage, contracts, tests, semantic definitions and development-schema configuration. It also lists validation behaviors including compile checks, dbt build commands, testing and parsing. The post says users can ask project questions or delegate changes such as documentation and model creation.

Explore Mode, now in public preview, extends that project grounding to conversational analytics. The launch announcement says business users can ask plain-language questions against the same dbt project maintained by the data team, and that weak answers can expose gaps for the team to improve. It does not establish general availability or promise an autonomous end-to-end analytics workflow.

For buyers, the summit package is therefore two things at once: a generally available engine and state-aware execution mechanism, plus a set of agent and context products still moving through beta and preview. Teams can evaluate dbt v2 and State as current products, but should treat Context Layer and the Wizard variants according to their stated availability rather than as one deployable GA stack.