Snowflake CoCo: The Coding Agent Built for Data Now Available Where You Work
At Snowflake Summit 2026, Snowflake CoCo becomes available where builders work. With a native desktop application, cloud agents, agent SDK, and (coming soon) mobile app and slackbot, CoCo will be fully integrated into the surfaces and tools where modern data teams live — all fully grounded in the enterprise data.
Adoption of AI in the enterprise has shifted from the theoretical potential of AI to the practical needs of production environments: putting AI to work safely, securely and governed, at scale.
The first wave of assistants offered autocompletion and query generation. The next wave is agentic: systems that inspect codebases, reason through complex tasks, modify files, run tests and manage pull requests with humans in the loop. In standard software engineering, this is already the new baseline.
For data and AI teams, the bar is higher. Enterprise AI runs inside governed data systems with live schemas, access controls, lineage and operational pipelines. Generic AI coding agents hit a wall here as they often generate code that looks right but lacks the grounding in data context and permissions to actually work in production.
Snowflake CoCo closes that gap, providing an agentic control plane for builders operating on modern data stacks.
"Thomson Reuters has built its data foundation on Snowflake to create a single source of truth across 37,500+ governed tables and 350 data sources, and now Snowflake CoCo is accelerating how we build on top of that. Our teams are modernizing legacy systems, scaling AI pipelines, and delivering insights in days instead of weeks, all within a governed environment. When you can go from idea to production that fast on top of trusted data, it fundamentally changes what's possible.
Caitlin Halferty
CoCo was built with a specialized harness for the data lifecycle, not as a generic wrapper around a model, but as an integrated system designed for how data engineers, analytics engineers, data scientists and AI builders actually work. It grounds the agent in data context, and connects it to the right tools and runtimes.
When it comes to real data engineering work, CoCo isn't just keeping pace with leading coding agents. It's setting the bar.
On ADE-Bench, an industry framework created by dbt Labs to evaluate AI agents on real-world analytics and data engineering tasks, CoCo achieved a 72.1% pass rate, outperforming both Anthropic's Claude Code and OpenAI's Codex (each at 65.1%)1. And the lead widens further on Snowflake native dbt project tasks, where CoCo's deep integration with the platform pays off.
Crucially, CoCo doesn't win by brute force. Compared to Claude Code running on Opus 4.7, CoCo uses 51% fewer tokens and takes 8% less time to get the job done. Two design choices drive that efficiency:
- Targeted vs. exhaustive exploration - Rather than scanning everything in sight, CoCo navigates directly to the data that matters, resulting in more efficient data exploration.
- A native tools approach - CoCo leans on native tools for working with data systems such as Snowflake, dbt and Airflow instead of falling back on bash based workflows, keeping work close to the data.
1 Efficiency score based on internal testing using ADE-bench, a framework created by dbt for evaluating AI agents on real-world analytics and data engineering tasks
What's new at Summit 2026
At Summit 2026, CoCo evolves from an AI coding agent into a full AI development platform. The new launches are organized around a simple principle: Your agent should work wherever you do and keep working when you can't.
“At Fanatics, our data demands shift constantly, and Snowflake CoCo gives our team the speed to keep pace. Engineers who used to spend days untangling pipeline issues and modeling data can now resolve those problems in hours, freeing them to build and ship new capabilities exponentially faster.”
Maddy Want
Use CoCo, where you work today
Cloud Agents are the foundation for CoCo's broader platform expansion. Automations, Slack integration and a native mobile app (the latter two coming soon) are all powered by Cloud Agents, bringing long-running, autonomous workflows to every surface where teams interact with CoCo.
For builders using CoCo directly within Snowflake environments, Cloud Agents bring the full agentic runtime directly into Snowsight. With Cloud Agents, each Snowsight session can now spin up an isolated, Snowflake-managed cloud environment behind the scenes, giving users the power of the CLI from any browser with no local setup, dependency management or infrastructure required.
When enabled, Cloud Agents provision a dedicated container for each CoCo session. The agent can execute shell commands, run Python scripts, install packages, read and write files, perform dbt builds and tests using a dynamically generated Snowflake profile, and search the web for additional context.
CoCo Desktop
CoCo is coming to Windows and macOS bringing the full power of Snowflake-native development into a native desktop app built for how data teams actually work.
“Snowflake CoCo is becoming a core part of how we operate as a company. We’re rolling it out across the entire organization, not just our data team. Because it understands our data, our environment, and our governance, teams can build and automate workflows on top of trusted data without needing specialized expertise.
Matt Luizzi
With CoCo Desktop, (generally available soon) builders get one governed surface to build across the data stack. You can create data pipelines, build applications, design agents, debug notebooks and visualize data flows without constantly jumping between screens. The editor becomes the place where code, data, context and execution come together so you can stay in flow from prototype to production.
Beyond the editor experience, CoCo Desktop brings an always-on AI agent to your local machine: a persistent assistant that understands project context, works across sessions and keeps momentum even when you step away.
With Automations, teams can move from reactive prompts to autonomous workflows, scheduling agents to handle recurring jobs like pipeline refreshes, data quality checks, model retraining and operational investigations — all governed by Snowflake role-based access control (RBAC) and backed by audit trails. The result is an AI development experience that behaves less like a one-off prompt box and more like a durable workforce, ready to investigate issues, make changes, run workflows and pick up exactly where it left off.
Also, because every team's stack is different, CoCo Desktop is extensible by design. With MCP integrations and a growing catalog of skills and plugins, teams can connect CoCo to the systems they already use and turn institutional best practices into reusable expertise. That means your team's standards, patterns and workflows can become repeatable building blocks helping every builder move faster while staying aligned with governed, enterprise-ready practices.
CoCo transforms the development experience for data and AI teams: local, agentic, extensible, and governed from the first prompt.
Coming Soon: Interact with CoCo on Slack or on the go with the latest mobile app
In addition to the editor, teams will soon be able to work with CoCo in Slack and on mobile. With the CoCo Slackbot, teams can ask questions, trigger workflows and share results directly in Slack, turning Slack into a governed interface for data work. Instead of relying on a generic chatbot, the Slackbot is connected to the enterprise data — tables, dbt models, Postgres schemas, Airflow pipelines and more — so answers are grounded in production context and constrained by the requesting user's access policies.
The CoCo mobile app will bring the same capabilities to iOS and Android for moments when work needs review, approval or action away from the desk. This is not a code editor on a phone; it is an agent interface for data leaders, analytics managers and builders who need to monitor pipelines, review scheduled task outputs, approve AI-generated workflows, inspect logs and ask natural language questions from anywhere.
Build on CoCo, programmatically — your way
CoCo is not just a tool builders use directly but also a platform teams can embed, extend and build on. CoCo supports Model Context Protocol (MCP) server and Agent Client Protocol (ACP), allowing other agents and enterprise systems to tap into its power.
Now, with the CoCo Agent SDK, builders can develop custom applications, internal tools, automations and domain-specific workflows on top of the the agentic engine that powers CoCo.
CoCo is now an agent platform with an SDK
The CoCo Agent SDK packages the same tools and agent loop that power CoCo for thousands of customers as an installable library. Developers get direct programmatic access to the capabilities CoCo uses in production: working with data, querying Snowflake, reading files, running shell commands, searching codebases, executing SQL, and editing code. The SDK includes built-in tool execution so your agent can start working immediately without you implementing any of that infrastructure yourself.
import { query } from "cortex-code-agent-sdk";
for await (const message of query({
prompt: `Profile the ORDERS table in MY_DATABASE.ANALYTICS:
- Total row count and date range covered
- Null rate for each column
- Top 5 customers by order volume
Summarize findings in plain English.`,
options: {
cwd: process.cwd(),
connection: "my-connection",
},
})) {
if (message.type === "assistant") {
for (const block of message.content) {
if (block.type === "text") process.stdout.write(block.text);
}
}
if (message.type === "result") {
console.log("\nDone:", message.subtype);
}
}
import asyncio
from cortex_code_agent_sdk import query
async def main():
async for message in query(
prompt="""Profile the ORDERS table in MY_DATABASE.ANALYTICS:
- Total row count and date range covered
- Null rate for each column
- Top 5 customers by order volume
Summarize findings in plain English.""",
options={"cwd": ".", "connection": "my-connection"},
):
if message["type"] == "assistant":
for block in message["content"]:
if block["type"] == "text":
print(block["text"], end="", flush=True)
asyncio.run(main())
Beyond the built-in tools, the SDK supports multi-turn sessions for stateful conversations across multiple interactions, structured output for returning typed, schema-validated JSON from your agents, MCP server integration for extending your agent with custom tools, hooks for intercepting and controlling agent behavior at each step, streaming output, and system prompts for customizing how the agent reasons about your domain.
Call it from pipeline scripts, backend services, or internal apps and tooling to embed data-native agentic capabilities directly into your stack. Platform engineers can integrate CoCo into CI/CD workflows, ISV partners can build data products on top of it, and teams can automate domain specific workflows.
Governance built in the DNA
CoCo brings AI-native development to every surface teams use, with Snowflake's enterprise-grade security and governance capabilities built in. Every operation runs under the user's existing Snowflake RBAC, LLM inference stays within Snowflake's security perimeter, and layered guardrails help protect against prompt injection and other LLM risks. Prompt and response logging, query tagging, usage monitoring, and admin-level cost and configuration controls give teams the auditability and governance they need to deploy agents safely at scale.
More than 7,100 Snowflake customers are already building with CoCo. At Summit 2026, CoCo becomes the AI development platform that powers work wherever it happens, powered by your data, aware of your stack and governed by design.
Learn more and try it yourself
Forward-looking statements
This article contains forward-looking statements, including about our future product offerings, and are not commitments to deliver any product offerings. Actual results and offerings may differ and are subject to known and unknown risk and uncertainties. See our latest 10-Q for more information.
