Automate the Path from Data to Predictive Insights with Agentic ML in Snowflake

Developing and moving models through the machine learning (ML) pipeline has traditionally been slow and manual, forcing data scientists into tedious troubleshooting cycles. Now, Snowflake is changing how teams work by introducing agentic ML through Cortex Code (generally available), Snowflake’s AI coding agent. Data science teams can use Cortex Code to automate development of production-ready ML solutions from natural language prompts on the same platform as their governed data.

Many customers are already leveraging the power of agentic ML to speed up production workflows in Snowflake ML including First National Bank of Omaha, a multistate holding company with $35 billion in assets and more than 4,500 employees.

Agentic ML is expanding access to machine learning across the business. At Kargo, a creative performance platform reimagining modern media buying, non-technical teams use Cortex Code to explore ML concepts—enabling deeper collaboration with data science teams.

Agentic ML workflows with Cortex Code

Agentic ML enables more trusted insights by automating tedious work, freeing teams to focus on higher-impact initiatives. Cortex Code comes with a rich set of ML-specific skills that streamline design, implementation and optimization of end-to-end workflows in Snowflake ML. Whether you’re training models, deploying models for inference, running distributed training, tuning hyperparameters or monitoring performance, Cortex Code intelligently accelerates one or more steps in the ML lifecycle by triggering the relevant specialized skills. 

Data science teams can rapidly build fully functional, high-quality ML pipelines by using Cortex Code to agentically plan, reason and select the optimal technique for each step across development and inference. Instead of spending cycles on navigating documentation, debugging errors or stitching together APIs, you can focus on applying your domain expertise and intuition to refine models and drive insights.

One example where agentic ML is driving productivity is feature engineering. Evaluating feature importance and identifying new feature recommendations were previously manual and time-consuming tasks. In this demo video, you can see how easy it is to use just a few natural language prompts in Cortex Code to quickly iterate on a churn model prototype, evaluate the importance of features across model types, clean up redundant or fragile features and surface the exact place where the model is falling short with new features ideas.

As Snowflake ML innovates, Cortex Code evolves with it to incorporate new capabilities and optimizations so you don’t have to worry about the complexity of distributed training, DAG-based orchestration or multimodal inference. This translates into significant productivity gains, enabling faster iteration and quicker paths from idea to impact.

Under the hood, Cortex Code leverages Snowflake ML’s fully integrated platform to scale workloads seamlessly across CPUs and GPUs with built-in optimizations — delivering performance advantages such as training up to three to seven times faster than open source libraries and inference latency for XGBoost models that’s up to 10x faster than legacy cloud providers.

Getting started 

Building agentic ML pipelines in Snowflake can be easily integrated directly into your existing workflows, regardless of where you prefer to work. Cortex Code is available directly in the Snowflake Snowsight UI (generally available) or within any terminal or code editor such as VS Code or Cursor with the CLI (generally available). 

In Snowsight, Cortex Code provides verified solutions in the form of fully functional ML pipelines that can be directly executed from a Snowflake Notebook in Workspaces. Snowflake Notebooks run in a Jupyter-based, container runtime environment purpose-built for large-scale AI/ML production workflows. 

Cortex Code CLI allows developers to transform faster, optimize orchestration workflows and debug with full context directly from the terminal. 

To get started building agentic ML pipelines with Cortex Code today, try this real-time fraud detection quickstart with Cortex Code in either the CLI or Snowsight from the free trial experience and the best practices guide for technical examples.