Snowflake Introduces CoCo, an AI Coding Agent for Data Engineers

Snowflake has launched CoCo, an AI coding agent specifically designed for professional data engineers. CoCo aims to help users generate repeatable, high-quality code and text for data engineering tasks by leveraging advanced AI models.

CoCo offers several benefits tailored for Snowflake users, such as data inference running within Snowflake's security perimeter and governance controls, native integration for reading schemas and query history, and built-in skills for various Snowflake workflows like Dynamic Tables and dbt. It also provides model flexibility, allowing users to choose among top frontier models.

To achieve reproducible outcomes, CoCo supports an open standard for 'skills,' which are Markdown files that inject domain-specific instructions and knowledge into an AI agent session. Skills can bundle scripts and reference materials, encoding institutional knowledge, and are managed through plugins for professional SDLC practices.

The post outlines key best practices for using AI coding agents, including starting minimal and iterating, understanding model intelligence, treating conciseness as a constraint, and prioritizing reproducibility. It also stresses that agents should not replace enterprise tools or make direct changes in production environments due to non-deterministic outputs.

CoCo is accessible through three distinct environments: a command-line interface (CLI) for terminal-native engineers, a full VS Code-based Desktop IDE for local repo integration, and directly within Snowsight for cloud-based long-running tasks. Installation instructions are provided for the CLI and Desktop versions, with Snowsight requiring no installation.