Snowflake has spent years building a rock-solid reputation as the definitive destination for heavy-hitting enterprise data storage. But if you are still looking at them as a passive data warehouse where data just sits waiting to be queried, you are missing the bigger picture. At Snowflake Summit 26 in San Francisco, the data giant made an aggressive play for the future of software development by rebranding and refocusing its Cortex Code suite into Snowflake CoCo.
The goal? Moving organisations past basic AI experimentation and straight into the era of the autonomous, agentic enterprise.
From Coding Assistant to Autonomous Agent
This isn’t just a superficial name change designed to chase marketing hype. The transition to CoCo represents a fundamental shift in how companies interact with their corporate data layers. Instead of being a standard text-based assistant that simply drafts code snippets or answers SQL queries for engineers, CoCo is positioned as a fully autonomous coding agent.
Acting as a core component of Snowflake’s agentic control plane, CoCo provides developers and analysts with a unified, governed environment to build applications, automate workflows, and operationalise AI through simple, outcome-based prompts.
What sets this apart for daily use is how Snowflake is delivering this environment across the tools teams already use. CoCo is launching as a native desktop application (CoCo Desktop), alongside mobile apps and a dedicated Slackbot, allowing teams to check on active workflows or pull insights on the go. For hardcore developers, it extends natively into existing setups via a VS Code extension and a plugin for Claude Code.
Perhaps more importantly, it also introduces a Microsoft Excel extension. This opens the door for non-traditional builders, like data analysts and business users, to orchestrate complex data pipelines entirely on their own without needing specialised software engineering expertise.
Technical Autonomy Under Deep Governance
On a technical level, Snowflake is focusing heavily on background execution and isolated security. New Cloud Agents allow developers to spin up a task within Snowsight and let it run securely in the cloud. Because the agent executes in the background, users don’t need to keep a local laptop session open or active.
To ensure this autonomy doesn’t create compliance risks, Snowflake has introduced:
- Secured Local Sandboxes: An isolated environment that protects local system files and sensitive corporate resources while the agent builds.
- Role-Based Access Controls (RBAC): Every automation is locked behind established enterprise permissions and backed by comprehensive audit trails.
- CoCo Agent SDK & Skill Catalog: Teams can embed these capabilities directly into their own internal setups or share proven data engineering workflows through a centralised corporate catalogue.
Feeding the Brain with Real-Time Data
An AI agent is only as accurate as the data feeding it. To bridge the gap between disconnected application silos, Snowflake also introduced Snowflake Datastream. This is a fully managed, fully Apache Kafka-compatible streaming service designed to feed fresh, continuously flowing data directly into Snowflake tables.
Typically, streaming live enterprise data requires running a separate infrastructure layer alongside your data cloud, complete with customised brokers and fragile connectors that drain engineering resources. Datastream eliminates that operational complexity entirely. Real-time data streams straight into the database, automatically inheriting the exact same security masking, lineage, and access controls as the rest of the Snowflake platform.
When paired with CoCo, developers can use simple natural language prompts to create and operationalise real-time AI applications on top of live, streaming data.
Real-World Traction at Scale
This agentic framework is already being proven at scale across Snowflake’s base of more than 13,900 global customers.
At Thomson Reuters, teams are utilising CoCo to modernise legacy data pipelines across a massive single source of truth containing over 37,500 governed tables and 350 data sources, shifting development times from weeks to days. At Fanatics, engineers who used to spend days untangling pipeline issues are resolving modelling complications in hours, accelerating audience segmentation and real-time fan engagement. Meanwhile, WHOOP is rolling the environment out company-wide to let non-technical personnel automate operations in a single, governed environment.
Snowflake’s pivot to CoCo gears the platform to be one that empowers its clients to build on top of a sound data foundation. This further emphasises the company’s ethos that enterprise data is only as valuable as what you can safely build on top of it. By combining live streaming infrastructure with an autonomous coding agent, Snowflake is turning passive corporate memory into an active development engine. It’s a direct, highly utilitarian leap forward for enterprise tech.
