Snowflake Inc. today detailed a sweeping set of product updates aimed at accelerating enterprise adoption of artificial intelligence, positioning its platform as a foundation for building, governing and operating AI systems directly on enterprise data rather than as an experimentation layer bolted onto existing infrastructure.

The announcements, timed around the company’s Build 2026 event in London, span AI-assisted development, machine learning operations, data governance, open data interoperability and transactional workloads. They reflect what Snowflake executives said is a shift from AI pilots to production systems that must meet enterprise requirements for reliability, security and compliance.

“Even after many years of advancements in data technology, we still hear customers struggling with data that is siloed, the complexity of managing infrastructure, and how to deliver data insights with maximum security, governance and compliance and regulatory controls,” said Christian Kleinerman, Snowflake’s executive vice president of product.

Cortex Code available

Today’s most significant announcement is the general availability of Cortex Code, an AI coding agent designed for data-centric workflows on Snowflake. Unlike general-purpose coding assistants, Cortex Code is trained to understand Snowflake’s data models, governance constructs and operational semantics, the company said.

“It’s different because it’s a coding assistant focused on data operations, data pipelines and data transformations, and it has a lot of the enterprise data context that organizations have,” Kleinerman said.

Cortex Code will be available both as an embedded interface within Snowflake’s Snowsight console and as a command-line interface that works with local development environments. The CLI version was not previously announced.

Kleinerman said early usage inside Snowflake has shown dramatic productivity gains. “We think it’s a massive nonlinear productivity improvement in how data operations are done,” he said.

Development across the lifecycle

Snowflake also outlined updates aimed at streamlining the full lifecycle of AI and machine learning development. Snowflake Notebooks, now built directly on a Jupyter kernel, are designed to support end-to-end data science workflows while maintaining Snowflake’s security and governance controls.

Kleinerman said customer feedback drove the adoption of the Jupyter Kernel. “The most common feedback we got from customers was the desire to be more compatible with Jupyter,” he said.

The company is also bringing online feature serving and model inference into general availability, enabling low-latency, real-time AI applications.

To address governance, particularly around semantic consistency for AI agents, Snowflake is also introducing Semantic View Autopilot, an AI-powered service that automates the creation and maintenance of semantic views. The system draws on query history and existing business intelligence models to keep semantic definitions aligned over time, reducing manual effort and the risk of inconsistent AI outputs.

Semantic consistency ensures that the same business terms are used across reports, dashboards, AI agents, applications and users. Kleinerman said it has emerged as a major limiting factor for enterprise AI.

“The quality of results Snowflake Intelligence provides is a function of how good the semantic view is,” he said. “The semantic view autopilot accelerates the creation and iteration of semantic views based on the needs of customers.”

Snowflake Intelligence and agents

Snowflake is introducing new features in its Snowflake Intelligence conversational, agentic application while emphasizing its role as the primary interface for business users, distinguishing it from Cortex Code’s developer-focused approach.

New capabilities include saving and sharing insights and expanded access to open data formats such as Apache Iceberg and Delta Lake. Kleinerman said Snowflake has yet to lose a competitive evaluation involving Snowflake Intelligence.

“It’s not just the vision, but also the realization that in customers’ hands, with customers’ data, Snowflake Intelligence is performing extremely well,” he said.

Snowflake Postgres, the open-source database picked up with its acquisition of Crunchy Data Solutions Inc. last summer, is now generally available. It allows enterprises to run PostgreSQL-compatible workloads natively within Snowflake’s platform.

The bulk of the road from preview to general availability involved making Postgres a “good citizen” within Snowflake’s environment, Kleinerman said, citing integration with authentication, encryption and business continuity features. PostgreSQL has consistently been one of the most highly rated database management systems by database administrators and developers.

Kleinerman positioned Snowflake Postgres as complementary to UniStore, Snowflake’s transactional storage engine.

“If a customer wants compatibility with Postgres, then the answer is Snowflake Postgres,” he said. “If a customer wants low latency writes, [online transaction processing]-type activity with compatibility with Snowflake, the answer is UniStore.”

Open theme

A persistent theme across today’s announcements is openness, particularly around data formats and interoperability. Snowflake is integrating Apache Iceberg representational state transfer interfaces into its Horizon Catalog while continuing to support open-source Polaris.

“We have zero interest in creating artificial lock-in for our customers,” Kleinerman said. “I want customers using Snowflake because they want to, not because the migration off of Snowflake is too expensive.”

He said that philosophy extends to data sharing, with Snowflake expanding support for open formats and cross-cloud access.

Another current in the product updates is a shift in how Snowflake sees enterprise AI adoption. Kleinerman said many customers are moving away from building bespoke AI systems to adopting platform-native, managed AI capabilities integrated into their existing data environments.

“Most companies that we engage with are shifting to the original Snowflake promise, which is we don’t want customers spending time on the infrastructure,” he said.

But he cautioned against expecting to see fully autonomous AI systems in the near term. “Trying to get solutions where AI takes on an entire problem and solves it is much harder than providing smaller use cases with a human in the loop,” he said.

Photo: Snowflake