Snowflake unveils Cortex Agents, powered by Anthropic’s Claude 3.5 Sonnet

Snowflake has today announced it has expanded its Artificial Intelligence (AI) capabilities with the public preview of Cortex Agents, streamlining agentic app data access across structured and unstructured data. Cortex Agents leverages Anthropic Claude as a key LLM powering agentic text-to-SQL for high-quality structured data retrieval.

AI agents, autonomous systems that perform tasks using AI, can enhance business productivity by handling complex, multi-step operations in minutes. Agents need to access an organization’s ever-growing structured and unstructured data to be effective and reliable. As data connections expand, managing access controls and efficiently retrieving accurate information—while maintaining strict privacy protocols—becomes increasingly complex.

“At Snowflake, we believe that AI agents will soon be essential to the enterprise workforce. They’ll enhance the productivity for many teams such as customer support, analytics, engineering, and they’ll free up employee time to focus on higher value things,” said Baris Gultekin (pictured), Head of AI at Snowflake. “Data agents, which is a specialized category of AI agents, will combine data and tools to deliver accurate grounded insights by effectively selecting the right data sources.”

“At Luminate, we’re revolutionizing how we deliver precise, data-driven insights to our clients through generative AI applications. Snowflake’s unified data and AI platform provides our developer team with scalable processing and retrieval for both structured and unstructured data — the critical building blocks for developing, deploying, and orchestrating data agents powering our application,” said Glenn Walker, Chief Data Officer at Luminate Data. “Using Snowflake Cortex AI brings advanced AI within the same security and governance perimeter as our data and saves us countless development hours, allowing us to unlock the full potential of entertainment industry data with agentic AI.”

Cortex Agents orchestrates across structured and unstructured data sources — whether it be Snowflake tables or PDF files stored in object storage — to deliver insights. They break down complex queries, retrieve relevant data, and generate precise answers using Cortex Search, Cortex Analyst and LLMs. This enables accuracy, efficiency, and governance at every step.

Cortex Agents plan tasks, use tools to execute them, and reflect on results to improve responses. Available as a convenient REST API, Cortex Agents can seamlessly integrate into any application. Agents use Cortex Analyst (structured SQL) and Cortex Search (unstructured data) as tools, along with LLMs, to analyze and generate answers. The workflow involves four key components:

1. Planning: Applications often switch between processing data from structured and unstructured sources. For example, consider a conversational app designed to answer user queries. A business user may first ask for top distributors by revenue (structured) and then switch to inquiring about a contract (unstructured). Cortex Agents can parse a request to orchestrate a plan and arrive at a response:

• Explore options: When the user poses an ambiguous question (e.g., “Tell me about Acme Supplies”), the agent considers different permutations — products, location, or sales personnel — to disambiguate and improve accuracy.
• Split into subtasks: Cortex Agents can split a task or request  (e.g., “What are the differences between contract terms for Acme Supplies and Acme Stationery?”) into multiple parts for a more precise response.
• Route across tools: The agent selects a tool — Cortex Analyst, Cortex Search or SQL generation from natural language — to facilitate governed access and enable compliance with enterprise policies.

2. Tool use: With a plan in place, the agent can retrieve data efficiently. Cortex Search extracts insights from unstructured sources, while Cortex Analyst generates SQL to process structured data. A comprehensive support for tool identification and tool execution enables the delivery of sophisticated applications grounded in enterprise data.

3. Reflection: After each tool use, the agent evaluates results to determine the next steps — asking for clarification, iterating, or generating a final response. This orchestration allows it to handle complex data queries while increasing accuracy and continuing to maintain compliance controls within Snowflake’s secure perimeter.

4. Monitor and iterate: After deployment, customers can track metrics, analyze performance and refine behavior for continuous improvements. On the client application developers can use TruLens to monitor the Agent interaction. By continuously monitoring and refining governance controls, enterprises can confidently scale AI agents while maintaining security and compliance.

Combined with other Snowflake offerings, Cortex Agents now provide an end-to-end solution for retrieving, processing and governing both structured and unstructured data at scale.