Snowflake launches new AI tools, unveils OpenAI partnership
New features such as an agent-powered code generator and automated semantic modeling simplify developing cutting-edge applications and improve the vendor's competitive standing.
As winter's chill blankets much of the U.S., Snowflake continues to drop new capabilities that simplify developing agents and other advanced applications.
On Tuesday during Snowflake Build London, a user event in the United Kingdom, the vendor launched Cortex Code, Semantic View Autopilot and the native integration of Snowflake Postgres in its AI Data Cloud, among a spate of other capabilities.
Cortex Code is an agent that enables users to generate code for building pipelines and applications while applying an enterprise's security and governance controls. Semantic View Autopilot is an AI-powered service that automates creating and governing the semantic views that give agents proper context. And Snowflake Postgres is a PostgreSQL database that Snowflake acquired in June 2025.
In addition to the new features, Snowflake on Feb. 2 unveiled a $200 million partnership with OpenAI that makes OpenAI models natively available within Snowflake's Cortex AI development environment. In addition, it includes plans for collaborating to build and deploy customized AI capabilities.
Collectively, the partnership and new capabilities are important advances for Snowflake, according to William McKnight, president of McKnight Consulting. In particular, he noted the value of eliminating costly and complex data pipelines by natively embedding Snowflake Postgres in the vendor's AI Data Cloud and Cortex Code's understanding of an enterprise's data environment.
"Snowflake, in this trove of announcements, wins the year in data so far and [furthers] its transition from a specialized data warehouse to a comprehensive AI and application platform," McKnight said.
Sanjeev Mohan, founder and principal of analyst firm SanjMo, similarly called Snowflake's latest slew of features significant for the vendor's customers.
"Snowflake already innovates fast, and the pace has picked up," he said. "Collectively, they give customers more optionality. And there was a big emphasis on skills, helping users codify their complex processes. That's a big benefit."
Based in Bozeman, Mont., but with no central headquarters, Snowflake is a data management vendor that has added AI development capabilities over the past few years in response to surging interest from customers in AI. Build London marks the third event in the last eight months at which Snowflake has unveiled a multitude of new AI capabilities, following Summit last June and Build last November.
Driving development
Snowflake's aim is to enable customers to create a connected data estate that can be trusted as a foundation for building and deploying AI and analytics applications, according to Christian Kleinerman, the vendor's executive vice president of product who spoke during a virtual press conference on Jan. 28.
Snowflake, in this trove of announcements, wins the year in data so far and [furthers] its transition from a specialized data warehouse to a comprehensive AI and application platform.
William McKnightPresident, McKnight Consulting
Poor data foundations and improper alignment with governance policies are among the main reasons that the failure rate remains so high.
Each of the three main capabilities Snowflake unveiled on Tuesday are designed to help customers create a connected, trusted data foundation for AI and analytics.
AI-powered code generation capabilities are not uncommon. However, tools that align natural language-generated code with governance and security policies from the outset of the development process are uncommon. And when enterprise-grade governance and security policies are applied to code late in development, the AI-generated code often doesn't align with an enterprise's governance and security standards, and the project never makes it past the pilot stage.
Domo recently launched App Catalyst, an AI-powered code generator that integrates governance and security from the outset of a project. Now, Snowflake is doing something similar with the release of Cortex Code.
Cortex Code, now part of Snowflake's Cortex AI development suite, understands user data, governance and operational semantics to give it context for creating code, and maintains an enterprise's governance and security standards to ensure that the code is enterprise-grade. Using the tool, data and AI teams can create production-ready applications far more efficiently than when they write code on their own.
"The most significant announcement we're making at Build is we're introducing Cortex Code," Kleinerman said.
While Cortex Code aids AI development by simplifying code generation, Semantic View Autopilot automates the creation of a semantic layer so that metadata and metrics are consistent across an organization and data can be discovered and trusted to inform analytics and AI applications. Similarly, running Snowflake Postgres natively within Snowflake's AI Data Cloud rather than externally through an integration advances development by simplifying access to unified transactional and analytical data that informs applications.
"I like Semantic View Autopilot," Mohan said. "For non-technical users to create agents with Snowflake Intelligence -- Cortex Code is for techies -- really well, there has to a robust semantic layer. That, to me, is the most important of the new items."
McKnight, meanwhile, called out integrating Snowflake Postgres into the AI Data Cloud as perhaps the most valuable of the new features.
"Snowflake Postgres [transforms] Snowflake from a purely analytical data warehouse into a transactional and analytical platform," he said. "Snowflake is not the first to bridge this gap, but it's significant because … it opens entirely new use cases, removes the cost and complexity of [extract, transform and load] pipelines, and enables zero-code migration."
Beyond the big three
In addition to the launches of Cortex Code and Semantic View Autopilot, and the native integration of Snowflake Postgres in the AI Data Cloud, Snowflake's new partnership with OpenAI is a significant move, according to McKnight.
Cortex AI enables users to access numerous AI models, including those from AI21 Labs, Anthropic, DeepSeek, Google Cloud, Meta and Mistral AI.
Even OpenAI models were available to users before the new partnership between the AI developer and Snowflake. However, they were only available through Snowflake's integration with Microsoft.
Native availability is a direct integration between a model and the architecture of an AI development environment such as Cortex AI, including its access to data and enforcement of governance and security policies. Unlike other methods of connecting models with development environments such API integrations or plug-ins, no complex configurations are required.
In addition to OpenAI's models, models from Anthropic, Google Cloud, Meta and Mistral AI are natively available in Cortex AI.
"Moving OpenAI models natively into Snowflake is a game-changer because it keeps sensitive data entirely within the Snowflake security perimeter, effectively removing the complex governance and data egress obstacles that kill enterprise AI projects," McKnight said. "Instead of complicated engineering, analysts can trigger GPT using simple SQL functions, democratizing high-level AI across the organization."
Additional new features Snowflake unveiled during Build London include the following:
Expansion of the Snowflake Horizon Catalog to include the open-source Polaris Catalog, which lets customers securely access data in Apache Iceberg tables as well as create, update and manage data stored in Iceberg tables.
Open Format Data Sharing to extend Snowflake's zero-ETL capabilities to open table formats Apache Iceberg and Delta Lake.
Snowflake Backups to protect business-critical data from ransomware or disruptions.
Updates to Snowflake Notebooks, including an integration with Cortex Code and Experiment Tracking, to make it easy for teams to compare testing results and reproduce top-performing models.
Cortex Agent Evaluations so users can trace, measure and audit agent behavior.
An integration with Vercel that enables vibe coding -- AI-assisted code generation using natural language prompts -- to build applications that can be deployed in Snowflake through Snowpark Container Services.
An integration with the Brave Search API so users can integrate real-time information from the internet into Snowflake Intelligence, Cortex Code and Cortex Agents to augment an enterprise's proprietary data.
New features in Workspaces, Snowflake Notebooks and OpenID Connect aimed at better enabling collaborative development.
"Some of [the new features] are improvement on existing technologies, but in all instances it's customer-driven innovation," Kleinerman said regarding Snowflake's impetus for developing the capabilities introduced at Build London.
Competitive standing
Two years after making AI development its main priority, Snowflake might have finally caught up to Databricks and other data and AI platform vendors, according to Mohan.
After OpenAI's November 2022 launch of ChatGPT significantly improved generative AI technology, Databricks and hyperscale cloud vendors AWS, Google Cloud and Microsoft all quickly reacted. They created environments for customers to build AI tools, including development frameworks and integrations with AI providers such as OpenAI.
Snowflake was slower to react, and only fully committed to enabling AI development in February 2024 when Sridhar Ramaswamy was named CEO.
"They've caught up," Mohan said, noting that Google Cloud similarly had to catch up after being viewed as an innovator of machine learning capabilities with its 2017 release of the Transformer neural network architecture.
"Google invented the Transformer and then watched the whole world take off with not only OpenAI but Meta with Llama and others," Mohan continued. "But look where Google is now with Gemini. So, it is too early to call winners in AI, and Snowflake has demonstrated that it has caught up after a late start."
McKnight likewise noted that with the release of its latest set of features -- particularly the integration of its Polaris and Horizon data catalogs -- Snowflake has fully caught up with its peers.
"Snowflake is now arguably ahead of Databricks in its ability to unify transactional applications and analytics, while having simultaneously neutralized the 'lock-in' argument," he said. "By embedding Apache Polaris directly into the Horizon Catalog, Snowflake now offers the same open governance as Databricks' Unity Catalog."
Looking ahead to what Snowflake could do next to continue serving its users and perhaps even attract new ones, McKnight named adding agent governance capabilities and more cost transparency.
"In its highly competitive market, it needs to address agent governance with a layer that governs intent and action, and application-centric costing where instead of seeing costs by warehouse, there is a 'Product View' that bundles the costs of the Postgres instance, the Snowpark Container Services and the Cortex API."
Mohan, meanwhile, suggested that Snowflake take steps to unify transactional processing and observational data such as customer behavior.
"I would like them to show how I as a developer can access all my data in a unified manner through a catalog," he said. "Horizon doesn't handle observe data, and I'd like to see all data in one place."
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with three decades of experience. He covers analytics and data management.
