Last edited on Oct 26, 2026
Hope Watson, product manager for v2 adapters at dbt Labs, framed this session around an 1800s railway metaphor: when two train systems use different track gauges, nothing is technically broken, but every transfer between them costs time and money. dbt v1 and dbt v2 have been running as two separate codebases for the past year, and for any given feature, teams have had to wait for it to land on whichever engine shipped it first.
Watson's talk covered how dbt Labs is closing that gap without asking users to pay the "cost of transfer" themselves. The core fix is architectural: dbt open-sourced the overlapping pieces of v1 and v2 so the ecosystem is converging on one engine and one binary over time, even though the migration itself will take years.
To make upgrading practical today, dbt ships a v2 parser inside v1.12 that lets teams check compatibility before committing to a switch, plus a dbt autofix deprecations command that resolved 22 flagged issues in a live demo in about five minutes. v2 reached general availability on most adapters as of the day before this talk.
Adapters got their own segment as what Watson called dbt's "secret foundation": the layer that makes wildly different databases (Snowflake uppercases identifiers, Postgres lowercases them; Snowflake has three timestamp types where BigQuery has two) look consistent inside dbt. In dbt v1, each adapter lived in its own repository with its own dependency chain, which meant installing a second adapter could quietly downgrade dbt core or pull in conflicting SDKs.
In dbt v2, adapters ship inside the same binary as the engine on a shared layer built on the ADBC driver standard, so multiple adapters, demoed live with Snowflake and Databricks, install and run together without version conflicts.
The parsing improvements were demonstrated head to head on a synthetic 1,800-model, 72-YAML-file project: v1 parsed in roughly 10 to 13 seconds, v2 in about 3. On error detection, catching a single SQL typo took about 1.5 seconds on the OSS distribution (which has to round-trip to the warehouse) versus 0.06 seconds on dbt's full distribution, which understands SQL directly.
The new linter, demoed catching and auto-fixing a lowercase-formatting rule violation, was described as 50 to 100 times faster than other linters. Docs v2 replaces the single, sprawling manifest.json (nearly half a million lines of JSON on that same 1,800-model project) with small, queryable Parquet files.
In the demo, Watson used Claude to write ad hoc queries against those files, including a column-level lineage query and a "blast radius" query showing every downstream model affected by renaming a single upstream column, the kind of impact analysis that would otherwise mean manually parsing a massive JSON file. The redesigned docs UI also added DAG grouping views.
Watch every keynote and breakout from dbt Summit 2026 on demand. Watch dbt Summit 2026 on demand →
Get started in dbt
Join the analytics engineers building data infrastructure that actually scales.
Install dbt Wizard CLI
Get started with an agent purpose-built for analytics engineering. It knows which tool to call, which context to pull, and checks its own work before surfacing anything to you.
Latest posts
Optimizing your runs for lower compute, fresher data, and faster iteration with dbt State
on Oct 26, 2026
Showing our work: a controlled experiment in context engineering
on Oct 26, 2026
From prompt to PR: how Careem’s AI writes and validates dbt models before a human ever sees them
on Oct 26, 2026
Join the largest community shaping data
The dbt Community is your gateway to best practices, innovation, and direct collaboration with thousands of data leaders and AI practitioners worldwide. Ask questions, share insights, and build better with the experts.