There comes a point in the development of any tool when it stops being an «interesting experiment» and becomes part of the workflow. Snowflake's Cortex Code seems to be heading in exactly that direction. The company recently announced a significant expansion of the capabilities and availability of its AI development tool, and it's worth taking a closer look.
What Is Cortex Code and Why Do You Need It?
In short: Cortex Code (or simply CoCo) is an AI assistant within the Snowflake platform that helps write code, automate data tasks, and build entire workflows. It doesn't just suggest a line of code – it can independently plan a sequence of steps, execute them, and return a result.
Simply put, it's not an autocomplete but more like a digital colleague you can assign a task to and get a finished solution. This is precisely what is called agentic AI – a system that doesn't just answer questions, but acts.
Now It's Available to Almost Everyone – and Directly in the Interface
Previously, Cortex Code was mainly accessible via the command line and required a certain level of technical expertise. Now, it has reached general availability (GA) in Snowsight – Snowflake's graphical web interface, which is used by most people working with the platform. This is important: the barrier to entry has been significantly lowered.
In parallel, Windows support has been added to the CLI version. Previously, the command-line tool only worked on macOS and Linux, which automatically excluded a significant portion of developers. That restriction has now been lifted.
Agent Teams: When One Agent Isn't Enough
One of the most interesting updates is the support for so-called agent teams. The idea is as follows: instead of one AI agent performing all tasks sequentially, several agents work in parallel, each on its own piece of the task.
Imagine you need to analyze a large dataset, write a report on it, and automatically run several checks. One agent handles the analysis, another generates the report, and a third runs the checks. All at the same time, with no waiting.
This isn't just an acceleration – it's a qualitatively different approach to automation. Such systems are already being used in large projects, and Snowflake is embedding this logic directly into its platform. By the way, a similar concept can be seen with other market players: OpenAI, for example, positions its compact GPT-5.4 mini and nano models specifically as sub-agents – fast and inexpensive performers that operate under the direction of a more powerful coordinator model.
New Skills: What CoCo Can Do Now
In addition to architectural changes, Cortex Code has gained a set of new skills – built-in capabilities that the agent can use to solve tasks:
- Working with Git repositories – the agent can now interact directly with the version control system, which is critical for team development.
- Code execution – the ability not just to write a script, but also to run it in a sandboxed environment and then return the result.
- Working with Snowflake documentation – the agent can reference official materials to provide more accurate and up-to-date answers about the platform.
- Data analysis – a new skill for examining table contents and drawing conclusions based on the data.
Each of these skills is a separate tool in the agent's arsenal. By combining them, it can solve tasks that previously required several manual steps from a developer or analyst.
Why This Matters – And for Whom
Snowflake is a platform used by companies for storing, processing, and analyzing large volumes of data. Cortex Code integrates directly into this process, which means that people who work with data every day get an AI assistant right where they already spend most of their workday.
For developers, this means less routine: less boilerplate code, manual runs, and switching between tools. For analysts, it's an opportunity to automate repetitive tasks without deep programming knowledge.
For companies building complex data pipelines (i.e., chains of automated operations like collection, processing, storage, and visualization), the introduction of agent teams is a potentially significant speed-up. This is especially true if these pipelines need to be maintained and regularly updated.
What Remains Behind the Scenes
Agentic AI is still a young field. The tools are evolving rapidly, but so are the questions: How predictably do agents behave in non-standard situations? How can their actions be controlled if they operate autonomously? How can security be ensured when an agent gains access to real data and systems?
Snowflake has not yet disclosed all the details about how the control and restriction mechanisms in Cortex Code are designed. This isn't a cause for alarm, but it is a reason to monitor its development – especially for those planning to implement such tools in sensitive workflows.
Overall, the direction is clear: AI in development is ceasing to be just a «smart autocomplete» and is becoming a full-fledged participant in the workflow. Cortex Code is one example of what this looks like in practice when a real infrastructure stands behind the idea.
