Snowflake has become the most recent Cloud Wars Top 10 organization to unveil its agentic AI strategy. The company has launched into public preview Cortex Agents — a fully managed service designed to simplify the integration, recovery, and processing of both structured and unstructured data for enterprise agentic AI development.
The announcement of new agentic AI tools coincided with recent news about the company’s ongoing partnership with Anthropic: Anthropic’s Claude LLM is a critical component that powers the agentic text-to-SQL capabilities for high-quality structured data retrieval in Cortex Analyst, which is one of the two core technologies in Cortex Agents.
Orchestration Agents
Cortex Agents empower Snowflake users to orchestrate operations across both structured and unstructured data sources. By utilizing a combination of Cortex Analyst, Cortex Search, and LLMs, these agents simplify complex queries, delivering relevant data with accuracy, efficiency, and built-in governance.
Cortex Agents utilize existing Snowflake tools, specifically Cortex Analyst and Cortex Search, while also leveraging LLMs for additional resources. They can be integrated into any application via a REST API. These agents can be deployed to plan tasks, utilize specialized tools for task completion, and evaluate results to improve future responses.
Cortex Analyst specializes in generating SQL queries from text, focusing on structured data. Recent enhancements for agentic AI in this tool include the ability to handle multi-table queries without sacrificing accuracy, generating and monitoring semantic models, and inputting custom instructions for specific business use cases.
Cortex Search allows Cortex Agents to retrieve unstructured data effectively. Recent improvements in the technology enable the indexing of hundreds of millions of rows while reducing service costs by 30%. The tool now features enhanced customization options, including the ability to choose a vector embedding model for semantic search, and new preview features.
Snowflake will also be introducing Cortex AI Observability. Delivered through TruLens, the LLM validation tool, Cortex AI Observability will enable users to evaluate the performance of AI agents and applications, compare the quality and accuracy of configurations, and accurately trace workflows.
Anthropic Effect
Building on the integration of Anthropic’s most advanced LLM, Claude 3.5 Sonnet, into Snowflake Cortex AI, this powerful model is now powering Cortex Analyst and, consequently, can be used to deliver more accurate structured data retrieval for Cortex Agents.
“Anthropic’s Claude models are raising the industry bar for intelligence,” says Arun Agarwal, lead product marketing manager for Cortex AI at Snowflake. “They outperform most competitor models across a wide range of evaluations on both reasoning and coding skills.”
Snowflake has established a distinction between its own AI agents and those from other companies. The company’s self-proclaimed data agents “combine data and tools to deliver more accurate, grounded insights by effectively selecting the right data sources and tools for retrieval.” As is the case with all AI agents and GenAI tools, access to accurate data is critical for their effectiveness.
Snowflake is leveraging its strengths by focusing on data, particularly through the use of two purpose-built tools designed to orchestrate both structured and unstructured data. The accurate retrieval of data, regardless of its source, is essential for the reliability of AI agents. With its pinpoint focus on data agents, Snowflake is both utilizing and enabling this functionality.
The launch of Cortex Agents, coupled with the integration of Claude 3.5, positions Snowflake as a provider of a solid and, more importantly, unique offering in the AI market.
