Snowflake Bolsters AI Agents for Enterprise

Snowflake is pushing its Cortex Agents platform further into enterprise readiness, announcing a suite of new capabilities designed to tackle the operational complexities of deploying AI at scale. The company’s focus is shifting from agent creation to robust management and operation.

StartupHub data

Companies working on this

Profiles of the companies named in this story, with funding and a one-liner from our database.

Snowflake

$12.4B

A cloud-based data platform enabling data warehousing, data lakes, data engineering, and data sharing.

The latest enhancements address challenges surfaced as customers move from initial agent deployments to managing enterprise-wide rollouts. These include orchestration, execution, and governance hurdles.

Building Smarter Agents

A public preview is coming soon for a new Coding Agent, leveraging the same runtime that powers Snowflake CoCo. This allows developers to embed AI coding capabilities directly into their own applications.

New features like Skills Package and Agent Toolset aim to streamline capability management. Skills Package lets users define a set of skills once and reuse them across multiple agents via a single URI. Agent Toolset allows agents to reference tools from other agents, eliminating redundant definitions and ensuring consistency.

Tool Search will optimize context building by allowing agents to discover and load relevant tool definitions on demand, rather than pre-loading all definitions upfront. This is crucial for large deployments with numerous integrations.

Operationalizing AI at Scale

For longer-running tasks, the generally available Async Agent API will allow agents to execute jobs in the background without holding open connections. This is particularly useful for complex data processing or report generation, aligning with the capabilities of products like Snowflake CoWork.

The new Code Execution Tool provides a sandboxed Python environment within Snowflake for data processing, calculations, and generating polished outputs like PDFs or presentations, all within a single agent run.

Interrupt and Resume functionality, also generally available soon, introduces human steering to agent workflows. Users can pause an agent, correct its course, and resume execution, simplifying error handling and state management.

Partial Access is another upcoming feature designed to serve users with different permission levels from a single managed agent. The agent will dynamically evaluate user permissions to determine accessible tools, enforcing granular access controls natively within Snowflake.

Streamlined Management

A new Versioning UI, currently in public preview, offers a visual way to manage agent configurations. It allows for comparison, rollback, and promotion of agent versions, enhancing governance and auditability, especially for regulated industries. This effort is part of Snowflake's broader strategy to control costs and ensure responsible AI deployment, as seen in discussions around Snowflake's AI Cost Controls.

These updates underscore Snowflake's commitment to providing a robust platform for enterprise AI, moving beyond initial development to focus on reliable, scalable operation, a move that also benefits their expanding ecosystem like the Snowflake Marketplace.