Graphene is a data analytics framework designed for coding agents, combining SQL with semantic layer governance to enable faster data exploration and visualization. It provides token-efficient languages, agent-friendly CLI controls, and supports multiple databases including Snowflake and BigQuery, with business logic stored in version-controlled repositories.
Snowflake open sourced pg_lake, a set of PostgreSQL extensions that enable Postgres to query, manage, and write to Iceberg tables and data lakehouse files using standard SQL. The technology was developed by Crunchy Data over several years and is now available under an Apache license to benefit the broader Postgres community.
This article discusses optimizing SQL queries that invoke large language models (LLMs) for data processing. The authors propose jointly optimizing query plans and LLM inference to achieve up to 14x speedups, addressing the high cost of AI-SQL systems that can generate millions of model calls per query.
A set of eight rules and tools prevent AI coding agents from corrupting data warehouses by controlling SQL operations, masking personally identifiable information, and requiring confirmation before destructive actions. Evaluation showed Claude Haiku 4.5 deleted entire tables and invented data without safeguards, but caused no damage across 27 runs with rules enabled.
Databricks introduced Lakehouse//RT at Data + AI Summit, claiming superior performance for analytic serving workloads directly on lakehouse data without copying. Snowflake and ClickHouse disputed the demo results, with ClickHouse publishing detailed methodology showing 420 QPS on a single node and Snowflake questioning Databricks' configuration, while StarTree demonstrated Apache Pinot can also query external lakehouse tables at scale.