Overview
- Snowflake announced four updates to its AI coding agent Cortex Code, adding general availability in the Snowsight UI, native Windows support for the CLI, a new multiagent feature called Agent Teams, and a set of packaged skills for common data tasks.
- In Snowsight, Cortex Code runs as a persistent, context‑aware helper that knows a user’s databases and schemas, can generate or fix SQL and Python with a diff view, and can execute verified ML pipelines directly in Snowflake Notebooks within Workspaces.
- Agent Teams turns Cortex Code into a coordinator that spins up parallel subagents with defined roles such as research, coding or testing, while keeping users in control to review progress, approve steps or redirect work without losing context.
- The new skills act like specialist playbooks that the agent can load for targeted jobs, including machine‑learning lifecycle tasks, cost‑intelligence for tracking credit use and budgets, Openflow data movement, Streamlit app development, and Cortex AI SQL functions for text tasks.
- Snowflake says its integrated Snowflake ML platform lets Cortex Code train models three to seven times faster than common open‑source libraries and serve XGBoost with up to 10x lower latency, and a TS Imagine executive credits the agent with large productivity gains; these are company‑reported claims without independent benchmarks.