Snowflake Inc. is leading off its Snowflake Summit 2025 user conference in San Francisco today by unveiling a series of new artificial intelligence capabilities intended to simplify the ways users interact with data.

The announcements reflect the company’s push to embed generative AI across its platform and make advanced analytics more broadly accessible.

Snowflake is betting that the future of enterprise analytics lies in seamless, AI-enhanced collaboration across technical and nontechnical teams and wants its platform to be the default interface between people and data. The strategy continues its evolution beyond its data warehousing roots into a unified platform for intelligent data operations.

“The goal is to bring the power of AI to analysts and personas that are typically comfortable with database technology but may not be fully versed in how AI works,” said Christian Kleinerman, Snowflake’s executive vice president of product.

Snowflake said the overall aim of today’s announcements is to reduce friction in data workflows and shorten the time between data ingestion and insight. The broader context is more strategic. With data volumes surging and demand for AI tools growing across industries, Snowflake’s latest offerings position it as not just a data cloud provider but as an AI-native platform.

Conversations with data

Leading the new lineup is Snowflake Intelligence, a conversational data agent that allows users to query enterprise data using natural language. Targeted at business users with limited coding skills, the tool enables them to ask plain-language questions of both structured and unstructured sources.

The functionality is made possible by intelligent data agents that operate within an organization’s Snowflake environment, inheriting all existing security, data masking and governance controls, Snowflake said. Embedding directly within a company’s trusted infrastructure avoids the compliance and security tradeoffs often accompanying external AI tools.

The agents can parse and unify data from various platforms including Google Drive, Workday Inc. applications, Box Inc. storage and Zendesk Inc.’s customer service platform using Snowflake Openflow, an extensible, managed multimodal data integration service for moving data between sources and destinations. The company said users can analyze spreadsheets, images, PDFs and database entries side by side with no custom data engineering required.

Universal Search for External Data enables users to discover data assets in sources like PostgreSQL or MySQL from within Snowflake. This is a small but strategic move that reflects the reality that most enterprise data is spread across a mix of platforms, and any tool that can reduce the friction of finding and using that data adds significant value.

Snowflake Intelligence supports access to third-party sources through Cortex Knowledge Extensions, a new feature in the Snowflake Marketplace that integrates external content from sources such as Associated Press, Stack Overflow, CB Insights and USA Today. That allows users to contextualize findings with current events, market trends and technical resources within the Snowflake platform.

“Our goal is to leverage AI to continue to shrink the effort and time that it takes to migrate data from a number of sources to Snowflake,” Kleinerman said.

On the more technical side, Snowflake is introducing Data Science Agent, a feature now in private preview that helps data scientists automate some of the more repetitive and time-consuming elements of machine learning workflows, such as data preparation, feature engineering and model training. These steps often stall machine learning projects between experimentation and production because of a lack of engineering resources or the complexity of debugging pipelines.

Data Science Agent uses Anthropic PBC’s Claude large language model to dissect machine learning projects into logical steps and deliver executable pipeline components that can be run inside Snowflake Notebooks. Users can iterate on the results by adjusting parameters or adding follow-up prompts, effectively turning the model into a copilot for ML development.

Multi-modal queries in plain SQL

Cortex AISQL, the newest member of Snowflake’s Cortex AI suite, is designed to extend the reach of SQL to unstructured formats like images, audio or long-form text. It effectively turns a familiar SQL environment into a multi-modal data interface for more informed analysis.

Now in public preview, Cortex AISQL lets analysts use standard SQL commands to query across diverse data types. This permits structured sales data, for example, to be merged with social media sentiment or customer service transcripts to be integrated with customer relationship management records. That means analysts can engage with a wider range of enterprise data without needing to learn new programming languages.

The system is backed by large language models from OpenAI, Meta Platforms Inc., Mistral AI SAS and Anthropic, which are integrated directly into Snowflake’s SQL engine. Snowflake said performance optimization features currently in private preview promise up to 70% performance improvements and 60% cost savings, depending on the workload.

AI-powered modernization

Snowflake is addressing the complexity of migrating legacy data systems into the Snowflake ecosystem with SnowConvert AI, a tool that simplifies data migration from older platforms into the Snowflake data warehouse and speeds the process up to threefold, according to the company.

SnowConvert AI leverages agents to automate code conversion, extract/transform/load processes, reconfiguration and report migrations. It also validates results with reduced human intervention. SnowConvert AI isn’t limited to databases; it also supports migrations of some business intelligence tools and ETL workflows, enabling organizations to standardize on Snowflake for a greater quantity of storage and analytics. In the same vein, a new Universal Search for External Data feature, enables users to discover data in sources like PostgreSQL or MySQL from within Snowflake.

Image: theCUBE Research/DALL-E 3