Snowflake Introduces Background AI Bots and Cortex Training
November 6, 2026 · 2 min read
Also published in Português (Brasil)
Snowflake has launched new capabilities for Snowflake CoCo and Snowflake CoWork at the Snowflake Expedition 2026 event. The company introduced CoCo and CoWork Bots, which are AI agents that operate in the background and only notify users when decisions or approvals are required.
Snowflake CoCo serves as a coding agent for data teams, while Snowflake CoWork acts as a personal work agent for knowledge workers. Currently, there are over 9,100 accounts using CoCo and over 5,800 accounts using CoWork. Organizations including Cisco, Docusign, Pacific Life Insurance Company, and United Rentals have implemented these tools.
The new Bots inherit the roles and permissions of the user to execute goals, such as preparing customer meeting briefings or maintaining data pipelines. These Bots will be available in private preview soon. Meanwhile, CoCo Automations is generally available, allowing data teams to schedule recurring workflows with audit trails. CoWork also supports scheduled data analysis, such as weekly sales reports sent to an inbox. Additionally, the Async Agent API is generally available to support multi step tasks that can resume after interruptions.
Snowflake expanded agent access through Snowflake for Slack, which combines CoCo and CoWork into one bot. CoCo for Excel and Sheets allows natural language queries on Snowflake data from spreadsheets, and the Snowflake Mobile app enables business users to request data and developers to manage coding sessions. These features are in public preview.
Governance updates include Skills and Plugin Sharing, which is generally available and allows CoCo workflows to be shared as official skills. Profiles in CoCo and Profiles in CoWork are in public preview to provide role and team context. A Memory feature allows agents to remember previous decisions, which users can edit or delete.
For data and costs, Snowflake launched the Snowflake AI Data Package in public preview, featuring over 60 industry datasets. Agentic Discovery on Snowflake Marketplace is also in public preview for finding and installing AI and data via conversation. The Snowflake Cortex AI Gateway added auto model tiers and dynamic model routing for CoCo in private preview to match tasks with appropriate models.
Snowflake also introduced Cortex Training in public preview, allowing organizations to fine tune open weight models using serverless reinforcement learning. This process occurs entirely within the Snowflake environment. Based on benchmarks using LoRA SFT on H200 GPUs, Cortex Training costs up to 61 percent less than serverless training services from Fireworks and Tinker. Cortex Training is developed on an open source project.