Snowflake Expands Agentic AI, Data Interoperability, Governance

SiliconANGLE reports that Snowflake used its Snowflake Summit in San Francisco to present a vision it calls the "agentic enterprise," unveiling product updates meant to help organizations build, govern, and operate AI systems on enterprise data. Snowflake announced general availability of Iceberg v3 support and new Snowflake-managed storage for Iceberg tables. It expanded Horizon Catalog, a governance, security, and discovery service that integrates the Apache Polaris open-source catalog and offers bidirectional, read/write interoperability between Snowflake-managed data and external engines. EVP of product Christian Kleinerman, quoted by SiliconANGLE, said, "Models keep changing, and capabilities keep advancing, but the data is constant."

Snowflake announced an intent to acquire Natoma, an enterprise Model Context Protocol (MCP) platform, on the same day as its Q1 FY2027 earnings. Snowflake says the deal will give it a natively integrated governance and identity layer for AI agents and MCP tool access, making it easier to connect and manage how AI systems reach enterprise applications, databases, APIs, and tools. The company also pointed to a collaboration with Anthropic and said its agent surfaces, Snowflake Intelligence and the CoCo coding agent, are among its fastest-adopted products.

Packaging Iceberg compatibility and managed storage with catalog and governance addresses a persistent practitioner problem: keeping model inputs, feature stores, and production data in sync across runtimes. Bidirectional interoperability between a central service and external engines reduces point-to-point ETL but increases reliance on robust metadata and access controls, and it raises questions about consistency guarantees, metadata latency, and cross-system transaction semantics.

Track uptake of Iceberg v3 in customer architectures, technical documentation or benchmarks on how managed Iceberg storage handles consistency and transactionality across Snowflake and external engines, how third-party model and runtime vendors interoperate with Horizon Catalog APIs, and the close and integration of Natoma alongside the Anthropic collaboration.