Meet Cortex Code CLI: Your AI coding agent for the full data stack

Build confidently with your data, right from your terminal

Vibe coding showed how fast AI can generate code. It also revealed the ceiling: speed without context, governance, and real enterprise data only gets you so far. Cortex Code CLI is built for what comes next.

If you can describe what you want to do with data, you can do it with Cortex Code CLI. It understands your data stack deeply, so it can act on your instructions accurately across data engineering, analytics, ML, and agent development, directly inside your terminal or favorite IDE. 

See it in action in the 2-minute overview, or go deeper with the 30-minute live demo and AMA covering dbt pipelines, parallel query execution, Jupyter notebook generation, and a candid Q&A on governance and MCP integration.

In addition to supporting Snowflake workflows end to end, Cortex Code is expanding support towards all data systems, starting with data engineering tools dbt and Apache Airflow®. It allows developers to build transformations faster, optimize orchestration workflows, and debug with full context. 

Getting started

Cortex Code CLI is now generally available. Install it with:

curl -LsS https://ai.snowflake.com/static/cc-scripts/install.sh | sh        

Run cortex in your terminal and the setup wizard connects you to your Snowflake account. If you already have a ~/.snowflake/connections.toml from the Snowflake CLI, you can reuse that connection directly.  If you’re not yet a Snowflake customer, a new monthly subscription is available!

Check out full details in the docs.

Start by asking what it can do

Once you're connected, a good first prompt is simply:

Cortex Code will walk you through its available skills. Out of the box, those include managing and orchestrating data pipelines, building semantic views, and creating and optimizing Cortex Agents. It also supports MCP, so you can connect to external tools like GitHub, Jira, and Confluence to pull in project context alongside your data.

You can also extend it with your own custom skills, written as standard markdown files and loaded from your project or user directory. Use /skill list to see what's available in any session. The community is already building and sharing skills openly, including one that takes a completed analysis and turns it into a PowerPoint deck.

More accurate and more efficient than generic coding tools

Rigorous testing demonstrated that Cortex Code is more accurate and efficient than generic coding tools in tasks requiring deep awareness of your data stack:

• Higher accuracy: We ran Cortex Code and Claude Code against the same ADE-Bench tasks using dbt + Snowflake with the same underlying model, Claude Opus 4.6. Cortex Code successfully completed 28 of 43 tasks (65%), compared to 25 of 43 (58%) for Claude Code.
• Greater efficiency: Cortex Code required nearly 50% fewer total calls to reach the correct solution, including 2x fewer file reads and 4x fewer bash commands. This efficiency is driven by a SQL-native approach using the snowflake_sql_execute tool, which enables native execution within Snowflake, resulting in more targeted exploration.

How teams are using it

"When you can have all of the power of Snowflake at your fingertips, right on your local computer in the command line, it's allowed us to unlock new capabilities. Cortex Code has been there to answer simple questions, guide you through the process, and really be an AI companion for developing however you want." -- Matt Louisey, Senior Director of Analytics at WHOOP

Watch the WHOOP story

At Fanatics, a product leader pasted a requirements doc into Cortex Code, pointed it at the right Snowflake environment, and had a working end-to-end prototype in under an hour:

"Cortex Code is changing how our developers deliver work. It's collapsing weeks into hours and months into days. That combination of speed and enterprise control is what makes it really transformative."

Watch the Fanatics story

Pro-tips

The best practices guide is worth reading once you're past the basics. A few highlights:

• Use /plan mode when working against production data. It shows you exactly what actions are about to run before they do
• Use read-only roles in production environments and avoid ACCOUNTADMIN for day-to-day work
• Use /resume, /new, and /fork to manage sessions across complex, multi-step work

From vibe coding to agentic engineering

Vibe coding showed how fast AI can generate code. It also revealed the ceiling: speed without context, governance, and real production data only gets you so far. Agentic engineering is what comes next. When AI operates directly inside your trusted data environment, the constraint shifts from who can build to what's worth building.

Cortex Code CLI is GA, works across dbt and Airflow, and is open to teams outside Snowflake. Anything you want to do with data, you can build with confidence using Cortex Code.

What's Next

If you're in the Bay Area, join us Tuesday, March 17 at 5:30 PM PT. We're throwing a Cortex Code Launch Party at SVAI Hub in Menlo Park. Meet the PMs and engineers who built it, see live demos, and connect with other developers who are already shipping with Cortex Code. Save your spot.

All the links from this issue:

• Build In Snowflake Faster With Cortex Code CLI (2-min demo)
• Cortex Code in Action: Live Demos + AMA (30-min deep dive)
• How WHOOP Builds at Hyper-Speed with Cortex (Short)
• Redefining Innovation at Fanatics with Cortex Code (Short)
• Cortex Code CLI Expands Towards Supporting Any Data, Anywhere (blog)
• Cortex Code CLI documentation
• Best practices for Cortex Code CLI
• Extending Cortex Code with custom skills
• Cortex Code Launch Party, Menlo Park, March 17 (Bay Area)

For the latest on life at Snowflake including company news, events, and culture, check out The Snowcast.