Snowflake’s new Cortex Agents release notes describe thinking_effort settings, extra system instructions and stage mounts for coding-agent runs using code_toolset_all. A stage mount allows the managed sandbox to access files stored in a Snowflake stage or versioned stage; thinking_effort changes the orchestration model’s reasoning budget.
This is relevant if a client already uses Snowflake for data or application workloads. Instead of exporting data and project files into an unrelated tool, an agent can work closer to the governed Snowflake environment. The system instructions are appended rather than replacing the sandbox’s own controls, an important limitation.
It is not a reason to adopt Snowflake for a small personal app. But it is a useful integration capability to know when assessing an enterprise contract: ask whether their data, code and governance requirements justify a managed coding agent rather than an external API script.
Bookmark the new API controls for a future Snowflake integration proposal; test only in an account you already have.