CoCo: Snowflake's AI Coding Agent for Data Engineers
Snowflake introduces CoCo, an AI coding agent tailored for data engineers, emphasizing repeatable workflows over one-off prompts and outlining best practices for production use.
Snowflake’s new AI coding agent, CoCo, targets professional data engineers rather than casual users, shifting focus from generating code to creating repeatable, high-quality outcomes. The tool is designed to integrate with existing Snowflake workflows, offering three environments: a CLI for terminal users, a VS Code-based Desktop IDE, and a browser-based Snowsight option requiring no installation. CoCo leverages existing Snowflake connections to execute tasks efficiently, prioritizing reliability over ad-hoc solutions.
CoCo’s design reflects emerging best practices for AI-assisted data engineering, including avoiding direct production changes due to nondeterministic outputs. The agent is positioned as a complement to enterprise tools like dbt for transformations and deployment frameworks such as schemachange, Flyway, or Terraform. Snowflake emphasizes that AI tools should enhance, not replace, established data engineering processes to maintain stability and reproducibility in production environments.
Skills are central to CoCo’s functionality, enabling data engineers to encode repeatable workflows in a standardized format. These skills use a SKILL.md file to define tasks, scripts, and conventions specific to a team, such as dbt project steps or naming standards. Skills are portable across AI coding agents, with plugins managing their lifecycle, versioning, and deployment through Snowflake’s Plugins Catalog for organizational sharing and governance.
Snowflake provides documentation and quickstarts for CoCo’s CLI, Desktop, and skills/plugins to help data engineers transition from one-off prompts to structured workflows. The company advocates for encoding institutional knowledge into skills and plugins to scale team capabilities, positioning reproducibility as the hallmark of professional data engineering in an AI-driven landscape.