Getty Images/iStockphoto
Snowflake's new Cortex Agents enables agentic AI development
The vendor's new managed service, now in preview, provides developers with an API for combining structured and unstructured data with LLMs to create agentic applications.
Snowflake on Wednesday unveiled Cortex Agents, a managed service that enables developers to build agentic AI applications.
Agents are generative AI applications built using a combination of an enterprise's proprietary data and large language models (LLMs) so the applications understand the enterprise's business. With that knowledge, agents -- unlike AI-powered assistants and chatbots that respond to questions and directives -- can be trained to act autonomously, surfacing insights on their own and taking on specific tasks.
Generative AI agents gained prominence during the second half of 2024 and are now the dominant trend in AI development, given their ability to make users smarter and more efficient. Vendors such as Google Cloud and Salesforce, the parent company of Tableau, provide capabilities for developing agents that can perform tasks that previously had to be done manually.
Cortex Agents, now in public preview, is Snowflake's version.
Because service lets developers build natively within the confines of the governed Snowflake platform, it is a significant addition for the vendor's customers, according to Stephen Catanzano, an analyst at Enterprise Strategy Group, now part of Omdia.
"Cortex Agents is a major step forward," he said. "It enables Snowflake customers to deploy AI agents at scale with unified governance, integrating data retrieval from both structured and unstructured sources."
Andy Thurai, an analyst at Constellation Research, similarly noted that Cortex Agents is an important new feature for Snowflake customers.
"[Cortex Agents] allows Snowflake customers to use a native platform to build AI to solve business needs," he said. "Most customers prefer a native platform to build things rather than creating tool sprawl. In this case, Snowflake being one of the dominant data lake players, it is easy for users to engage Cortex Agents."
New capabilities
Generative AI has evolved quickly in the two-plus years since OpenAI's November 2022 launch of ChatGPT marked a significant improvement in generative AI capabilities.
Initial applications included chatbots that enabled users to query and analyze data using natural language rather than code and text-to-code translation tools. Quickly, however, generative AI has matured to the point at which applications can be trained to act autonomously.
As a result, enterprises are rapidly increasing their investments in AI development. With data providing AI its intelligence, data management and analytics vendors have responded by building environments where customers can use their data in conjunction with generative AI models to do their development.
While not at the forefront, Snowflake, along with its competitors, has been in on the trend. Cortex, Snowflake's AI development environment, became generally available in May 2024, and in November, the vendor introduced tools to streamline the development of conversational AI applications.
Now, with Cortex Agents, Snowflake is providing customers with capabilities to build generative AI agents.
"Data is what unlocks and powers generative AI, and generative AI is what unlocks and powers business transformation," said Christian Kleinerman, Snowflake's executive vice president of product, during a virtual press conference. "Cortex Agents is the next frontier for Snowflake to help customers innovate and obtain better business outcomes."
Cortex Agents is available as a REST API. With the REST API, developers can access structured data through Cortex Analyst and unstructured data through Cortex Search, combine their governed Snowflake data with an LLM to create an agent and integrate that agent into any application.
Following the establishment of a partnership between Snowflake and Anthropic, Cortex Agents is optimized for Anthropic's Claude LLM with agents providing greater accuracy when used with Claude than other LLMs. However, Snowflake customers have choice when selecting LLMs.
Because optimal development is limited to one LLM, Cortex Agents, while an evolution for Snowflake, is not as robust as competing agentic AI development environments, according to Thurai.
"It's an OK solution," he said. "It limits the users … whereas many other agentic platforms provide options to use a choice of many underlying LLMs."
In addition, though a fully managed service that simplifies agentic AI development, it limits how developers can build agents, failing to provide options such as low-code/no-code visual development capabilities, Thurai added.
Despite limitations, Cortex Agents is nevertheless a solid framework for carrying out its intended purpose of enabling Snowflake customers to build AI agents, according to Catanzano. Claude is one of the highest-performing LLMs, and the REST API provides a straightforward method for retrieving relevant data.
"It's a strong, well-rounded framework," Catanzano said. "Cortex Analyst ensures accurate SQL generation for structured data, while Cortex Search provides state-of-the-art unstructured data retrieval. Anthropic Claude offers robust reasoning and complex workflow execution, making it highly effective for developers."
Competitive standing
With Cortex Agents, Snowflake continues to build up its AI development environment. However, unlike Databricks and tech giants such as Google Cloud and Microsoft, Snowflake sat back for much of the first year after ChatGPT was launched and has had to catch up.
The vendor has fully embraced AI since Sridhar Ramaswamy became CEO in February 2024. However, its capabilities still don't quite measure up, according to Thurai.
"They are still way behind when it comes to AI and generative AI embracement," he said. "Other vendors, including Databricks and hyper-scalers, are innovating and offering tools much faster."
To further catch up, the vendor should add support for more LLMs and provide a full platform for agentic AI development that includes more than an API, Thurai continued.
"Snowflake offerings are a decent stop in the right direction, but they have a lot to catch up on to be a good player in agentic AI platforms," he said.
However, given the point at which Snowflake started after spotting competitors the better part of year, Snowflake is making noteworthy progress, according to Catanzano.
"Snowflake is catching up quickly," he said, adding that Cortex Agents represents progress if not revolutionary innovation. "While [Snowflake] started slower than competitors, Cortex Agents seems well positioned with its focus on unified governance, scalability and integration with enterprise data."
Regarding what Snowflake can do to catch up, Catanzano suggested improving AI observability, refining multi-modal data integration and improving real-time monitoring and governance capabilities.
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.
