Snowflake just made AI code suggestions in Cortex Code (CoCo) generally available. But the most interesting part isn’t the autocomplete—it’s what it knows about your environment.
Most SQL assistants guess schema structures. They suggest tables that don’t exist or columns you don’t have access to.
Same query. Different contexts. Zero execution.
CoCo inline suggestions fix this by operating directly within Snowsight Workspaces, using your actual query history, table schemas, and RBAC policies to predict the next portion of your SQL statement.
Why this approach works better for data teams:
- Suggestions are grounded in actual catalog context, not best-guess schemas
- It respects Snowflake RBAC, referencing only authorized objects
- Integrates directly into SQL files and Notebook SQL cells
- Already adopted by enterprise teams like Fanatics, Thomson Reuters, and WHOOP
The real differentiator: it’s not just generating syntax. By pulling context from Snowflake Horizon Catalog, it understands lineage and business metric definitions before suggesting code.
My honest take: AI autocomplete is becoming a commodity, and relying solely on inline suggestions won’t magically fix broken data models. But by treating the warehouse as context rather than just a target, Snowflake is moving from generic syntax generation to actual environment-aware development.
📚 Source: https://lnkd.in/dcNZ2Sh8
P.S. How much time does your team spend fixing AI-generated SQL that hallucinates non-existent columns? 👇
