From dbt Core to dbt v2: how CarGurus unlocked dbt State and AI at scale
CarGurus moved an 80-developer, three-project dbt v1 shop onto dbt platform and dbt v2 without breaking production, then used that foundation to stand up a governed, in-house agent platform.
Overview
CarGurus migrated 80 developers from dbt Core to the dbt platform and upgraded to dbt v2. dbt State cut Snowflake costs 25% in five months and AI tools reclaimed 24,000 annual work hours. Covers infrastructure modernization and governed AI integration across Snowflake, Atlassian, Omni, Glean, and Slack.
What you'll take away
- 01
CarGurus staged its migration as four sequential bets, orchestration, a clean parse on dbt v2, project-by-project cutover, then dbt State, rather than one big-bang cutover.
- 02
dbt's autofix tooling resolved about 70% of v2 deprecations automatically, reaching a clean parse across all three projects in roughly three weeks.
- 03
With governed data in place, CarGurus built three live AI agents, a PR review agent, a job triage agent, and a conversational analytics agent, all bounded by the dbt MCP server as the single boundary between agents and dbt.
Session speakers
Jordan Morgan
CarGurusStephen Thibeault
dbt Labs