Fivetran + dbt Labs at dbt Summit 2026 — dbt v2 and dbt State GA; Fivetran Context Layer, dbt Wizard, Lake Compute and dbt Charts introduced (press release + recap, 16 Sep 2026)

Fivetran + dbt Labs at dbt Summit 2026 — dbt v2 and dbt State GA; Fivetran Context Layer, dbt Wizard, Lake Compute and dbt Charts introduced (press release + recap)

Tag: S-2026-09-16-dbt-summit-2026-announcements Type: article (vendor press release, BusinessWire-syndicated, plus the vendor’s own product recap blog) Author(s): Fivetran + dbt Labs — press release bylined Elaine Green (quotes Anjan Kundavaram, CPO); recap by Corinne Hallander; customer quotes from Virgin Media O2 (Gordon Curzon), RxBenefits (Chris Shepherd), Obie, Paylocity, RMIT University Date of source: 2026-09-16 Date ingested: 2026-09-17 Authority weight: medium — dated primary announcement with an explicit availability-status table; low for every performance and cost figure (vendor/customer-testimonial claims, no independent corroboration retrieved) Raw file: S-2026-09-16-dbt-summit-2026-announcements.md

What it claims

At dbt Summit (Las Vegas, 16 September 2026; 2,000 attendees claimed) the merged Fivetran + dbt Labs announced a portfolio it frames as “Open Data Infrastructure” — a vendor-neutral stack in which storage, compute, movement, transformation and visualisation can each be chosen independently, so that AI agents “work across platforms using data and context the enterprise owns”.

Engine. dbt v2.0 is GA: the Rust “Fusion” engine becomes the single dbt engine (dbt Core, the Python implementation, is renamed dbt v1 and stays Apache 2.0). v2 ships in two distributions — a free superset dbt with SQL-comprehension features and optional paid add-ons, and an Apache-2-only dbt-oss subset. Claimed: parses a 10,000-model project up to 10x faster and surfaces “errors, column checks, and lineage before anything runs”. Adapters GA for BigQuery, Databricks, DuckDB, Redshift and Snowflake; ClickHouse/Spark beta; Athena, Fabric and Postgres “coming soon”.

dbt State (GA, paid). Inspects warehouse metadata and model SQL to build, skip, clone or defer each model per run. Freshness moves “from the job to the model”: each model carries a lag_tolerance in code, so “the guardrails now sit in the infrastructure instead of with whoever, or whichever agent, issues the command”. Claimed average 15–30%+ warehouse-compute reduction; Virgin Media O2 cites 25% savings on run time and BigQuery cost; RxBenefits 59% on scheduled jobs. Runs on dbt v1.7–v2.0 across Snowflake, BigQuery, Databricks and Redshift.

Context for agents. The existing dbt Semantic Layer (GA), Agents Schema (open-source context standard) and dbt MCP Server (GA) are joined by out-of-the-box Anthropic and ChatGPT integrations (GA) and by Fivetran Context Layer (private beta), which adds unstructured sources (docs, tickets, Slack threads) to dbt’s structured context and maintains the result “as data moves and transforms”.

Agents for the data team. dbt Wizard (public preview in the dbt platform; CLI public beta; Desktop private beta) is a project-grounded agent that “validates proactively: checking upstream and downstream impact, compiling and building the change before anyone sees the diff”, reviewable against the full DAG; Wizard Explore Mode (public preview) gives business users natural-language answers grounded in the same project.

Storage/compute. Managed Data Lake Service (GA) writes Apache Iceberg tables to customer-owned storage; Lake Compute (private beta) is a DuckDB-based single-node engine that runs tagged dbt models directly against Iceberg, with refs working across engines.

BI. dbt Charts (public beta) defines dashboards as YAML “version-controlled right next to the models they depend on, living in the same repo, the same pull request, the same CI as the SQL underneath it” — “a shared language” humans and agents can read, write and review.

Notable quotes

“…gives teams and their agents accurate real-time feedback, surfacing errors, column checks, and lineage before anything runs.” (press release, “A faster, more efficient engine”)

“…the guardrails now sit in the infrastructure instead of with whoever, or whichever agent, issues the command.” (recap, dbt State section)

“Instead of governance living in a separate, closed tool, they are defined as YAML and version-controlled alongside the dbt models they reference…” (press release, dbt Charts paragraph)

“Agents are only as trustworthy as the context they can access.” (press release, “An open standard for agent context”)

What’s speculative vs. asserted

• Asserted (vendor, verifiable in principle): GA/preview/beta statuses per the recap table; the v1/v2 renaming and licensing split; adapter list; the design of dbt State (per-model lag_tolerance), Wizard (pre-ship impact validation), Charts (YAML in repo/PR/CI) and Context Layer (Agents Schema, MCP delivery).
• Asserted but unverified: 10x parse speed; 15–30%+ compute reduction; all customer percentages (testimonials selected by the vendor).
• Speculative / forward-looking: Lake Compute and Fivetran Context Layer are pre-GA; the “open lakehouse vision” and “bidirectional” multi-engine future are positioning.
• Not addressed: any regulator, regulation, evidence-retention model, certification, or EU/UK financial-services customer (Virgin Media O2 is a UK telco). The governance read-across on Fivetran + dbt Labs and Agentic Data Access Governance is this wiki’s inference and is marked as such.

Topics this feeds

• Fivetran + dbt Labs — new company page (second and third vault sources naming dbt after S-2026-09-09-ataccama-apache-ossie-converter, S-2026-08-24-qlik-idc-marketscape and S-2026-09-04-alation-idc-marketscape).
• Agentic Data Access Governance — the transformation layer now claims the governed-context-for-agents role (MCP Server, Agents Schema, Context Layer) and adds pre-run validation and dashboards-as-code as evidence primitives.

Open questions raised

• What policy, masking or certification metadata travels with a dbt MCP / Context Layer call, and is each agent call logged with retention? Not stated (the same gap recorded for every MCP-context claim on the topic page).
• Does dbt Wizard’s pre-ship validation produce a retained record (impact analysis, test results) that could serve as change-control evidence, or only an in-session diff? Not stated.
• Is the lag_tolerance freshness rule reportable as a timeliness control (BCBS 239 Principle 5) — i.e. can a firm evidence which models breached tolerance and when? Not stated [inference].
• Which EU/UK regulated firms run dbt v2 / dbt State in production? No FS customer named.