Data Platforms Race to Embed Jev-Style Decisions Inside SQL Pipelines
VentureBeat reports Databricks, Snowflake, MotherDuck, and Astronomer are adding SQL decision primitives, with MotherDuck labeling 100,000 rows in 40 seconds for 50 cents.
Just one month after Jev’s debut, cloud data warehouses and analytics engines are racing to place System One decision models directly inside SQL queries, VentureBeat’s Sean Michael Kerner reported on October 19.
MotherDuck’s 100,000-row benchmark
MotherDuck, the cloud analytics platform built on DuckDB, introduced a native prompt_jev() SQL function for paid workspaces so analysts can classify text columns—such as extracting the primary complaint from support transcripts—directly in SQL. In a benchmark over 100,000 news articles from a public four-topic dataset, MotherDuck reported that Jev classified every row in 40 seconds for $0.50 at 89% accuracy. By comparison, an OpenAI generative model reached 88% accuracy in nearly 32 minutes at a cost of $37.58.
A new hybrid SQL pattern across warehouses
VentureBeat documented similar moves across the entire data stack: Databricks shipped native SQL decision functions and serverless open-weight options, Snowflake launched governed decision models inside Cortex, Astronomer published orchestration walkthroughs, and open-source extensions brought jev and jev_choice functions directly into DuckDB and PostgreSQL.
The shared architecture is emerging quickly across data engineering teams: deterministic SQL handles joins, thresholds, and exact filters; a fast decision model resolves high-volume messy judgment columns with confidence scores; and only low-confidence rows ever touch a heavyweight LLM.