dbt Labs and Snowflake customers shared Iceberg, agent and cost-saving patterns at a dbt-centered October

October was dominated by dbt-centered customer stories about running Snowflake with less spend and more automation. Guitar Center reported cutting Snowflake spend by 40% after rebuilding its platform around trust and control [6]. Espresso AI and Authentic Insurance argued that most dbt models sit off the critical path and can run slower and cheaper, which turns scheduling slack into saved credits [4]. EF Education First built a Snowflake Cortex Agent that scores dbt models on cost and value, and used it to cut runtimes on some pipelines by 80% [13].

Open table formats and catalogs were a second theme. A new catalogs.yml v2 spec separated dbt's Iceberg configuration from Snowflake-specific catalog integrations, and a dbt Mesh demo showed teams on Snowflake and Databricks sharing one set of Iceberg tables through Horizon and Unity catalogs without copying data [12]. Posts described Iceberg on Snowflake as simple enough for a small team to set up in an afternoon [10], and described moving Snowflake-native tables to Iceberg in place while keeping dependent dbt models and dashboards working [11].

AI agents and migration also featured. Okta moved agent context out of the Snowflake UI and into dbt, so Claude over MCP and Snowflake Cortex read the same governed definitions [15]. Under Armour shipped Snowflake agents as dbt models using custom materializations [8]. WebstaurantStore onboarded more than 700 sources in 90 days using Cortex LLM functions for dbt staging [17]. Ecolab used an AI migration agent to shrink about 84,000 lines of legacy Snowflake SQL by 6x [5], and a SQL agentic code converter reached general availability [2].

In the community, a benchmark compared Snowflake with Databricks, BigQuery, Redshift, Fabric, ClickHouse and MotherDuck across 244 queries [3]. Discussion of Databricks versus Snowflake continued, with one thread asking whether people compared features instead of foundations [19].