The AI Productivity Gap for Data Centric Workloads
More than half of professional developers use AI tools daily; Claude, Copilot, and ChatGPT have become as routine as Stack Overflow once was. But for data teams, these tools tell only half the story. The same platforms that accelerate individual productivity consistently fall short when it comes to moving complex enterprise data work from idea to production.
There are two fundamental gaps that generic AI tools have never been designed to close.
The first is context. Most AI coding assistants understand your code, but they stop at the repository boundary. They have no awareness of your data catalog, your schemas, your semantic models, or the governance rules that govern who can see what. This translates into developers spending their time manually pasting in table definitions, explaining data models to an AI that forgets it, and debugging errors that stem from the tool simply not knowing your environment.
The second is security. Generic tools typically send your queries, schema names, and business logic to third-party servers. For any organisation operating in a regulated environment, or one that takes data governance seriously, this is a non-starter.
Cortex Code is Snowflake's answer to closing the gap between your data and the tools that exist.
What is Cortex Code and why am I excited about it?
Launched in November 2025, Cortex Code is Snowflake's native AI coding agent that is purpose-built to not just write code but to understand the environment that code will run in.
Unlike a traditional chatbot, Cortex Code is an agent that doesn't stop at a response, it acts. It can read your data catalog, write and execute SQL, build dbt pipelines, deploy AI agents, and optimise existing workflows, all through plain language conversation. And critically, it does all of this within your governed Snowflake environment, meaning it only sees and acts on data your user is permitted to access.
The depth of understanding is what sets it apart. Cortex Code knows your specific databases, schemas, tables, and semantic models. It understands Snowflake's Role-Based Access Control (RBAC). It's aware of your query history, compute costs, and production pipelines. When you ask it to "update the top performers' query to show the top 100" or "which five service types are using the most credits," it knows what you're referring to.
Cortex Code isn't limited to one type of user. From admins managing permissions, data engineers building pipelines, analysts exploring data, and developers deploying AI applications Cortex Code is designed to meet people where they are, in the language they already use.
How Do You Access It?
Cortex Code is available through two interfaces, and the right one depends on how and where you work.
Cortex Code in Snowsight is the web-based experience, built directly into the Snowflake UI. It's context-aware in real time and knows which workspace you're in, the data you're looking at, and what you were working on in your last session. It's ideal for SQL development, admin tasks, data discovery, and data science notebooks. It operates like an expert colleague looking over your shoulder in the platform, ready to help you refine a query, explain a schema, or fix a failed execution with one click.
Cortex Code CLI takes the same agent and brings it to your local terminal and code editor. You can access it in VS Code, Cursor, or any environment of your choice, on macOS, Linux, or Windows (native Windows support as of March 26th, 2026). This is where it gets particularly powerful for engineering-heavy workflows. The CLI can read and write to your local repositories, supports dbt and Apache Airflow natively, and is built for end-to-end project work: building from scratch, deploying agents, orchestrating pipelines, and managing everything from a single interface.
Example task: "Build me a monthly revenue mart in my dbt project and make sure it's tested properly before we deploy".
It's worth reiterating: both interfaces operate within the context of your Snowflake data and are governed by your existing RBAC policies. The agent doesn't bypass controls - it works within them.
What Can You Actually Build?
The use case surface area is broad, but it maps cleanly to three areas:
Admin and Data Discovery - Search and explore the catalog, manage permissions, understand schema relationships, and perform governance tasks through conversation rather than manual navigation.
Building on Snowflake - Generate synthetic data, build AI agents, create Streamlit apps, and develop backend analytics logic. Use cortex code to automate complex data integrations and enrich their insights layer, essentially enabling engineers to spend less time on context-setting and more time on meaningful outputs.
Data Engineering - Build, manage, and optimise dbt pipelines; migrate legacy code to Snowflake-compatible syntax; orchestrate workflows with Airflow all without leaving your development environment.
We won't be elaborating on machine learning workflows, agentic development, or the growing library of specialised built-in capabilities that Snowflake continues to add. That's best saved for a future article but what is worth noting is that Cortex Code is a tool that has, so far, shown no signs of compromising its ability to keep pace with the rapid rate of AI innovation.
See Cortex Code in action across a real scenario - I've compiled a series of screenshots that walk through my deployment of a streamlit app that analyzes sales performance across call centers, a 20 minute build with Cortex Code.
What This Means for You: New vs Existing Snowflake Customers
New to Snowflake? Accelerate Your SSIS Migration.
One of the most common journeys we see at Keyrus is organisations moving from SQL Server Integration Services (SSIS) to Snowflake. This is the framework Keyrus recommends for SSIS migrations, combining Snowflake's SnowConvert AI and Cortex Code with a structured delivery approach.
The approach follows a structured five-phase framework: Assessment → Planning → Mapping → Implementation → Validation. Each phase builds on the last, and the combination of tools means that much of what used to require manual effort is now automated or AI-assisted.
It starts with SnowConvert AI, which parses your existing SSIS packages, generates visual dependency graphs (DAGs), and produces assessment reports that give you full visibility into migration scope, complexity, and risk before a single line of new code is written. It's worth noting that SnowConvert AI is free to access for any Snowflake customer or partner, lowering the barrier to getting started considerably. Migration effort is then prioritised using EWI (Early Warning Indicator) severity, so teams focus first on the highest-impact gaps rather than working through packages in arbitrary order.
Cortex Code then takes over the heavy lifting of planning, code generation, and deployment. It generates the SQL, dbt models, UDFs, and test suites needed to replicate SSIS logic in Snowflake-native constructs. Even unsupported components like C# scripts embedded in SSIS packages have defined workaround patterns using stored procedures and Snowflake UDFs. Validation is built into the process, with functional equivalence testing ensuring that what arrives in Snowflake behaves exactly as it did at the source.
The framework scales from a handful of packages to enterprise-wide workloads, and the same structured approach applies throughout. Whether you're in the early stages of evaluating a migration or already mid-journey, this gives your team a repeatable, auditable path to Snowflake without starting from scratch on every package. Note that the SSIS Replatform feature in SnowConvert AI, which converts packages into dbt projects, is currently in public preview. We recommend engaging with Keyrus or your Snowflake account team to confirm current coverage before scoping your migration.
Already on Snowflake?
Cortex Code compounds everything you've already built.
The organisations that have invested in governing their data are the ones who will get the most out of Cortex Code immediately. The agent already knows what's there. It knows what's sensitive, what's expensive to query, what your top-performing pipelines look like. Data engineers can optimise without context-switching. Admins can automate repetitive governance tasks. Analysts can get to answers faster without waiting for engineering capacity. The platform investment you've already made becomes the foundation the agent stands on.
Best Practices: Getting the Most Out of Cortex Code
Cortex Code is powerful, but like any capable tool, the quality of output depends on the quality of input. A few principles that make a real difference:
Communicate naturally. Describe what you want, not how to do it. "Show me which service types are consuming the most credits this month" will get you further than trying to structure a prompt like a query. Iterate conversationally and just say what you'd like changed.
Plan before you build. For complex, multi-step tasks, use /plan to see the full approach before anything is executed. Start small, test frequently, and build component by component.
Review before accepting. Always review proposed changes, especially SQL, DDL, and DML operations, before they run. Use the built-in diff view to see exactly what will be created, edited, or deleted. This isn't just good practice; it's how you stay in control of an agentic system.
Stay secure. Review any privilege grants or RBAC changes carefully. Use specialised skills like semantic-view-optimisation where your workflow demands expert precision.
Conclusion
Cortex Code represents a meaningful shift in what it means to use AI for data work. Not a chatbot you query in isolation. Not a tool that forgets your environment the moment you close the tab. An agent that knows your data, respects your governance, and operates where you already work whether that's in a browser, a terminal, or an IDE.
For new Snowflake customers, it shortens the path to value during migration. For existing customers, it deepens the return on a platform investment already made. For many customers, it closes the gap that has made AI tools feel impressive in demos but frustrating in practice.
Ready to explore what Cortex Code could look like in your environment? Keyrus works with organisations at every stage of their Snowflake journey - from migration strategy through to AI agent deployment. Get in touch with our team to find out how we can help you move from pilot to production, faster. Contact sales@keyrus.co.za