Snowflake on Tuesday unveiled a storm of new capabilities that further aim of becoming a control plane for agentic AI.
These new features include Business Ontology, which provides agents with preapproved business definitions builds on Snowflake's semantic modeling capabilities, and Cortex AI Gateway to govern how agents access models and take action across enterprise systems address trust and help control the cost of building and running agents.
In addition, new CoCo and CoWork capabilities such as Bots inherit employees' access rights to act on behalf of users, while Automations to schedule repeated workflows help users work more intelligently. And autonomous data engineering, which now includes Pipeline Builder to simplify building AI and data pipelines, enables customers to scale agents more efficiently.
"Snowflake's new capabilities collectively mark an architectural transition from a passive data warehouse into an autonomous execution platform, addressing critical enterprise pain points like heavy [extract, transform and load] maintenance, data latency, unreliable AI context and ungoverned agent costs," William McKnight, president of McKnight Consulting, told TechTarget.
He added that competing data and AI vendors such as AWS, Databricks and Microsoft with its Fabric platform likewise offer agentic frameworks and semantic context layers. However, the openness Snowflake provides through its Horizon Catalog, which is built on open-source capabilities such as Apache Polaris and the Apache Iceberg REST Catalog API, distinguishes the vendor's capabilities from the more closed environments competitors provide.
"Today, real enterprise lock-in comes from who owns the customer's business context, metadata and security rules, not the compute alone," McKnight said. "Snowflake Horizon delivers an open, multi-engine catalog architecture, [while] Databricks Unity Catalog functions as a closed ecosystem that treats foreign catalogs as read-only endpoints."
Advancing trust and efficiency
Many enterprises are more successfully moving agents into production now than at the start of 2026 with Deloitte noting that 28% of organizations surveyed for its October State of AI in the Enterprise report revenue gains from AI, up from 20% in January. However, that means 72% of organizations are still getting zero return on their investments in building agents and other AI applications.
Snowflake's aim is to become a control plane for agents so that customers can build and scale networks of AI tools that can be trusted to perform as intended in production, according to Christian Kleinerman, the vendor's executive vice president of product.
"The most important technology that informs the quality of insights is the data of an organization," he said during a virtual press conference on Oct. 27. "We've been working for 12-plus years to help organizations un-silo their data and govern their data so they can get more value out of it, and now, the opportunity to get more value is larger than before. … AI for Snowflake is not theoretical."
Snowflake's new capabilities collectively mark an architectural transition from a passive data warehouse into an autonomous execution platform, addressing critical enterprise pain points like heavy [extract, transform and load] maintenance, data latency, unreliable AI context and ungoverned agent costs.
William McKnight
President, McKnight Consulting
Business Ontology is perhaps the most valuable feature Snowflake introduced during its virtual user event Expedition, according to Stephen Catanzano, an analyst at Omdia, a division of Informa TechTarget. The feature, which will soon be in public preview, builds on Snowflake's existing semantic modeling capabilities by enabling enterprises to designate official, enterprise-wide definitions for terms and metrics so agents can operate on a shared, consistent data foundation.
"Business Ontology is particularly valuable -- everyone is talking about this -- because it solves the 'which number is right' problem that undermines agent credibility," Catanzano said. "By giving every business term one approved, traceable definition that agents must use, it directly targets [a] top cause of poor agent outputs."
Cortex AI Gateway, meanwhile, is similarly valuable given that it treats governance as infrastructure rather than an afterthought, he added.
Meanwhile, from a competitive standpoint, Catanzano noted that Snowflake is attempting to differentiate itself from competitors by providing an integrated platform rather than a series of individual capabilities that customers must piece together.
"Snowflake's pitch is that these capabilities live natively on one governed data copy rather than being assembled from separate products," he said. "That architectural bet, rather than any individual capability, is where Snowflake is staking its differentiation, and it is a credible but not unique position since competitors are moving in the same direction."
Like Catanzano, McKnight named Business Ontology as perhaps the most significant of Snowflake's new features.
In addition, he highlighted CoCo Cloud Agents, which enable teams to run CoCo -- Snowflake's coding agent -- in isolated environments managed by Snowflake to improve security, and CoCo Automations that let teams run agents on a schedule to make agents more effective and cost-efficient.
Both CoCo Cloud Agents and CoCo Automations are generally available.
"CoCo Cloud Agents and Automations and Business Ontology stand out as most transformational," McKnight said. "CoCo Cloud Agents reduce the heavy pipeline maintenance tax by running in secure cloud sandboxes. ... Meanwhile, Business Ontology directly targets the AI trust gap by anchoring AI reasoning to human-approved business definitions and metrics rather than probabilistic model guesses."
More needed to enable full autonomy
Snowflake's latest barrage of capabilities directly addresses fractured context, unconstrained execution and uncontrolled costs, which are some of the main reasons many AI projects fail before making it into production, according to McKnight.
"Snowflake's new releases create a cohesive, production-oriented product that targets the primary reasons agentic AI projects stall in pilot testing," he said.
But Snowflake needs to add more features before customers can completely turn business processes over to AI, McKnight continued, noting that Snowflake still lacks automated feedback loops that continuously retrain agents and master data management to resolve issues with data in source systems.
"Despite a strong foundation, critical operational gaps remain before enterprises can fully automate mission-critical workflows without human handholding," he said.
Catanzano likewise noted that Snowflake's new features seem appropriately constructed to inform and control agents. However, there needs to be more proof that they work at scale in production environments before a conclusion on their effectiveness can be reached.
For example, Cortex AI Gateway capabilities such as Agent Abuse Detection that detect when agents operate beyond their permissions and Cortex Sandboxes that provide a secure space to run custom code next to proprietary data need real-world validation before Snowflake customers can trust them, according to Catanzano.
"The pieces fit together logically. … What still looks thin is maturity and proof at scale," he said.
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than three decades of experience. He covers analytics and data management.
