Snowflake Cortex Code CLI adds dbt and Apache Airflow support for AI-powered data pipelines

Snowflake Cortex Code CLI, Snowflake’s coding agent that was announced at the end of February, recently added support for dbt and Apache Airflow. There are several significant implications for extending Cortex Code’s utility to the enterprise.

The announcement not only marks the first time the coding agent has been applied to workloads outside Snowflake but also reaffirms the vendor’s commitment to the open-source community. Both dbt and Apache Airflow are popular open source frameworks for transforming data and implementing data pipelines.

Because Cortex Code CLI makes use of Agent Skills—folders of instructions and scripts for specific tasks—Snowflake’s support of dbt and Airflow all but democratizes what Umesh Unnikrishnan, Snowflake Head of Developer Experiences, termed “public standards” for one of the audiences that needs them most.

Additionally, Snowflake introduced a new self-service monthly subscription model that allows anyone to access the coding agent, including those who aren’t Snowflake customers. Although Snowflake initially targeted the developer and data engineering markets by expanding Cortex Code to dbt and Airflow, the agent is equally viable for front-office workloads such as sales and Business Intelligence applications.

According to Unnikrishnan, the agent was designed to fortify “agentic engineering, where you still have to follow sound engineering practices, not just push out a pipeline and hope it works. You actually have to ensure that you have harnesses in place so that when a pipeline breaks, it knows what’s wrong and how to fix it. So, when you’re building a dbt pipeline or an Airflow job, you’re also going to build all these harnesses around it, and Cortex will help you.”

Agent skills for data pipelines

Anthropic released Agent Skills in the final months of 2025; Cortex Code, powered by the latest models from Anthropic and OpenAI, includes Agent Skills relevant to dbt and Airflow use cases. Specifically, it contains packages for debugging, optimizing, and testing pipelines. These resources are influential for the speed at which Cortex Code can solve nontrivial data pipeline problems, including propagating changes downstream.

“When you’re building a dbt pipeline or an Airflow job, you’re also going to build all these harnesses around it, and Cortex will help you.” — Umesh Unnikrishnan, Snowflake Head of Developer Experiences

Agent Skills are designed to equip language models with the acumen to rapidly become well-versed in jobs they otherwise have nominal familiarity with. “They tell these LLMs how to do some very specific tasks in a very predictable, deterministic, and structured manner,” Unnikrishnan said. “With these skills that we built with Airflow, dbt, or our own product, Openflow, we tell the model, ‘don’t just tell the user random things: ask them first, then set up this thing. And then you help them set up this other thing, and create a test for it, and then you run all three together.”

dbt Models

Cortex Code can produce a profound effect on implementing dbt models in two ways. In situations in which tables for data are what Unnikrishnan called “well-named and well-structured”, they can build semantic models for the data pipeline tool. With the coding agent, “You can just point it at a table and say, ‘ Hey, go create a semantic model for me,” Unnikrishnan explained. “If you are building one of those things by hand, it will take you an hour or two. Now, it’s possible within minutes.”

“Now, Cortex Code can go inspect all of that. It knows the lineage of that data and updates everything that depends on it. It can do this in minutes.” — Umesh Unnikrishnan

The agent may be even more helpful for complex semantic models involving dbt—especially when something in them changes. When adding a column, for example, engineers are required to update the model and all places that depend on it, which can be time-consuming. “Unless you have all that written down or it’s in your brain on a giant white board, it’s hard to figure that out,” Unnikrishnan commented. “Now, Cortex Code can go inspect all of that. It knows the lineage of that data and updates everything that depends on it. It can do this in minutes.”

Airflow tags

Many data pipeline jobs in Airflow are predicated on tags that perform different actions for engineers. These actions are typical for data integration or transformation use cases and include ingesting, cleaning, extracting, aggregating, and loading the resulting data into a target.

Instead of manually writing the code for each of these operations, organizations can now employ Cortex Code to automate them. “Setting up that tag and then triggering a pipeline so that it runs every hour or day that your business requires it, that is something that Airflow does trivially now,” Unnikrishnan said.

Front office applications

Snowflake’s agent is equally viable for facets of natural-language interactions, including ad hoc question answering. For example, customer support applications might integrate and aggregate data from several sources to provide service representatives with relevant information. The language model powering Cortex Code can assess numerous aspects of a business end user’s question to deliver accurate responses.

With Snowflake’s agent, “The LLM does the translation of that Natural Language Query to something that comes out of a database like Snowflake,” Unnikrishnan explained. “Behind that, there’s the question of how that data gets in there, and how the LLM understands what a random column in a database means, and how it responds to what this salesperson is asking for.”

Just the beginning

Cortex Code CLI’s recently added support for dbt and Airflow is positioned by the vendor as just the beginning of what it hopes to accomplish with the agent. The grand vision is for it to eventually support any data, regardless of its location and tooling. By starting with two widely adopted open source resources, the vendor has set the bar high. However, its intent may be gleaned from the subscription model it offers to agents, so that Snowflake customers are not the only ones who can use it.

Time will tell which tooling the Cortex Code CLI will support next.

The post Snowflake Cortex Code CLI adds dbt and Apache Airflow support for AI-powered data pipelines appeared first on The New Stack.