Cortex Agents by Snowflake Makes Powerful, Scalable Agentic AI a Reality

Snowflake is unveiling Cortex Agents, a fully managed service that simplifies integration, retrieval, and processing of structured and unstructured data, whether they exist in Snowflake tables or as PDF files stored in object storage. These AI data agents are designed to deliver accurate, grounded insights at scale, powered by Cortex Search, Cortex Analyst, and large language models (LLMs).

Snowflake’s Cortex Agents are a response to the growing interest—and challenges—of agentic AI. With “challenges proportional to its potential,” enterprises implementing AI agents struggle to maintain accuracy, security and trust, and governed data access.

To help organizations with this agentic juggling act, Cortex Agents help Snowflake customers build high-quality agents at scale atop a unified governance framework and efficient processing engine for both structured and unstructured data. Cortex Agents are capable of planning tasks, executing them with various tools, and reflecting on results for continuous improvement, powered by Claude 3.5 Sonnet—Anthropic’s most intelligent LLM, according to Snowflake.

Breaking down these capabilities further, Cortex Agents can:

• Plan: Cortex Agents can parse a request—even if an ambiguous question is posed—to orchestrate a plan and arrive at a response. Cortex Agents can split tasks into multiple parts for improved accuracy, as well as route across tools to facilitate governed access and enable policy compliance.
• Leverage tools: Once a plan has been developed, Cortex Agents can efficiently retrieve data and extract insights from unstructured and structured sources with other Snowflake tools.
• Reflection: After tool usage, Cortex Agents evaluates the results to guide its next steps—from requesting clarification to iterating or generating a final response. This unlocks the ability to orchestrate complex data queries while maintaining accuracy and compliance.
• Monitor and iterate: After deployment, teams can track metrics, analyze performance, and refine agent behavior for confident agent scaling. Additionally, with TruLens, customers can monitor agent interactions.

Delivered as a REST API, Cortex Agents can be easily integrated into any application, according to Snowflake.

Cortex Agents can use three key Snowflake technologies: Cortex Analyst, Cortex Search, and Cortex AI Observability.

Cortex Analyst generates SQL to process structured data, based on business terms that it has mapped to underlying data. Cortex Analyst now features improved handling of increased schema complexity; semantic model generation and monitoring; and customization for business-specific logic.

Cortex Search is the basis for which Cortex Agents retrieves unstructured data, featuring a combination of vector and lexical (keyword) search. Some improvements to Cortex Search include enhanced scale and affordability, improved customizability, and new preview features.

Finally, Cortex AI Observability, powered by TruLens, brings reliability, performance, and trust to generative AI applications. With end-to-end evaluation, evaluation comparisons, and comprehensive tracing, Cortex AI Observability helps make Cortex Agents more transparent and explainable.

To learn more about Cortex Agents, please visit https://www.snowflake.com/en/.