Snowflake Intelligence will provide an interface to design ‘data agents’ to help generate insights from enterprise data and take action.
As enterprises look to automate more processes with the help of AI-based agents that don’t require human intervention, Snowflake has showcased a new offering at its ongoing developer conference, Build 2024, which it claims is its agentic AI proposition for enterprises.
Named the Snowflake Intelligence platform, the low-code offering, expected to be in private preview soon, is designed to help enterprises unlock insights from their data through the use of data agents, and to take action on these insights.
Snowflake Intelligence will allow enterprise users to create their own data agents using natural language and will address question-and-answer tasks that require structured data to begin with, according to Baris Gultekin, Head of AI at Snowflake.
“It will expand to use cases that also require the use of unstructured data currently siloed in places like Google Workspace, Confluence, Salesforce, and more. Snowflake Intelligence will also support business intelligence tasks, ranging from generating quick answers to creating interactive charts for deeper insights,” Gultekin said, adding that the goal is to start supporting use cases that leverage both structured and unstructured data.
One such example could be a user deriving insights that require using data from multiple sources such as an analytical table, a document from SharePoint, or notes from a Salesforce account.
The platform also supports API calls that can be used by enterprises to allow data agents to take actions such as making modifications to the data, Gultekin said.
Inside Snowflake Intelligence
The Snowflake Intelligence platform is a combination of multiple capabilities that Snowflake released earlier, including Cortex AI, Cortex Search, Cortex Analyst, and the Horizon Catalog.
“Snowflake Intelligence uses Snowflake Cortex AI as the generative AI engine to access industry-leading large language models and retrieval services for responses grounded in enterprise data, which include Cortex Analyst for structured data and Cortex Search for unstructured data,” Gultekin said.
Cortex is a fully managed (serverless) service inside the Data Cloud that is designed to provide enterprises with the building blocks to use LLMs and AI without requiring any expertise in managing complex GPU-based infrastructure.
Separately, Cortex Search and Cortex Analyst, introduced earlier this year, was designed to help enterprises build chatbots that could answer questions about an enterprise’s data using natural language.
Snowpark inside Snowflake handles the custom code execution, such as Python, and the external API calling.
The agents created via the Snowflake Intelligence platform are natively integrated with the Horizon Catalog, which makes the agents compatible with open table formats such as Apache Iceberg and Polaris.
Harnessing the Cortex Chat API capabilities
Snowflake’s Intelligence platform will also harness Cortex Chat API’s capabilities, according to Gultekin.
Cortex Chat API, which is expected to enter public preview soon, is aimed at helping developers connect the chat application front-end with Snowflake and ensure that answers to questions or queries from users are grounded in enterprise data.
The API, according to Gultekin, combines structured and unstructured data into a single REST API call that helps developers make use of retrieval-augmented generation (RAG), which is useful for creating AI-based applications.
Other Cortex updates, such as Cortex Knowledge Extensions, can also be made use of by the platform, the company said.
Cortex Knowledge Extensions, which will be available inside the Snowflake Marketplace and are currently in preview, are designed to provide data teams a way to enrich enterprise AI chatbots with content from third-party providers, such as research or newspaper publications.
The ability of data agents to take actions on behalf of a user makes the Snowflake Intelligence platform similar to several agentic AI offerings introduced by vendors such as Salesforce, Microsoft, Anthropic, IBM, and ServiceNow.
The platform also seems very similar to other offerings, especially in terms of how a low-code or a no-code platform is used to create the agent and later used to manage it.
Snowflake Connector for SharePoint, currently in public preview, can also be used via the Intelligence platform. The connector allows data teams to access data without the need to manually set up any pipelines, Gultekin explained.
A natural evolution
Bradley Shimmin, Omdia’s chief analyst for AI & data analytics, sees the combination of Snowlake Cortex AI and Snowflake ML as “the natural evolution” of the company’s objectives, enabling developers to deploy new agentic processes alongside more mature AI technologies to build more mature and complex applications.
“I think what we’re seeing here is Snowflake attempting to match pace with Salesforce, which rolled out a pretty mature and comprehensive agent platform, Agentforce, a month or two ago,” he said.
Shimmin expects Snowflake to focus on automating functions such as writing back generated information to the datebase, a boon for data teams struggling to keep up with business demands.
And while he he sees Snowflake and its partners creating pre-built agentic workflows for enterprises to adopt at some point in the future, “Early adopters of Snowflake Intelligence will be building out their agentic processes using Cortex AI to set these up on a case-by-case basis,” he said.
Snowflake is yet to announce the general availability of the Intelligence platform.
