The Agentic Control Plane for Data Engineering

‍

TL;DR: Snowflake CoCo and Genesis are not competing tools. CoCo is the expert on everything inside Snowflake. Genesis sits above that layer, connecting GitHub, dbt, S3, Confluence, Jira and all other data tools, databases and platforms outside of Snowflake, and lets customers choose which LLM powers the work, including Cortex itself. Genesis’s Chief Data Agent, Eve, scans metadata across every connected system and data tool to build a context graph, using this context graph she maps out the entirety of your data estate, inside and outside of Snowflake. Since CoCo can’t reach outside of Snowflake, Eve can work with all external agents like Claude Code, Codex, and others to land data into Snowflake to handoff to CoCo.

The honest answer to "Genesis Eve or Snowflake CoCo" is not either-or. It is both, working together.

‍

Where Snowflake CoCo Ends and Genesis Eve Begins

Snowflake CoCo (formerly Cortex Code) is a Snowflake native coding agent built directly into Snowflake. It reads a customer's actual schemas, RBAC policies, and lineage before it writes a line of SQL or a dbt model, which is why its output tends to need less cleanup than a general-purpose coding assistant. CoCo lives inside the governed perimeter of Snowflake and is built to be excellent at that.

Eve and her team of data engineering agents work across the entire data lifecycle around the warehouse, ingesting and migrating source systems into Snowflake, building and monitoring the pipelines that keep it fresh, and connecting the wider data universe back so what lands there is complete and trustworthy. Where CoCo reasons over data already inside Snowflake, Genesis owns everything that gets it there and keeps it flowing, from source to warehouse and every stage in between. In Agent Server: Agent Access Control Explained, we wrote about how Genesis agents hand off work and share context with each other, and how that handoff actually stays secure.

‍

Eve, the Chief Data Agent

Genesis sits on top as the agentic control plane. The first thing a customer does after logging into Genesis is connect to their systems and data tools. We meet customers wherever they already are, so they can connect whatever tools they run. Our post on connecting data sources in Genesis walks through how that connection process actually works in the product.

The next step is choosing which LLM powers Eve. In a Snowflake environment we will natively use Cortex,  For customers with a Bedrock, Anthropic, or OpenAI subscription, they can connect those instead, and switch between them as needed. 

‍

How the Context Graph Gets Built

Once the source systems and an LLM are connected, Eve starts scanning metadata across every connected database and data tool to build a context graph. This is a one-time setup step. Multiple agents can run at once, users can collaborate in the same workspace, and Eve can hand off specific Snowflake work directly to CoCo when that is the more direct path. We cover how that kind of agent-to-agent handoff is governed in Agent Server: Agent Access Control Explained, RBAC, Caller Limits, and Safer A2A.

The practical effect is that a data engineer is not choosing between "the Snowflake AI" and "the everything-else AI." They are working with one system where the right specialist agent takes the right piece of work across the entire data lifecycle. That same philosophy shows up in how we built CoCo integrations, and in what happens when three CoCo sessions run in parallel on the same underlying context.

‍

Why the Control Plane Framing Matters

An agentic  control plane coordinates specialized agents and routes work to the right one using the tools you already have to do the work where your data already lives. That is closer to what a data engineering team actually needs than a single coding agent trying to be an expert in Snowflake internals, dbt modeling, GitHub history, and Jira ticket triage all at once.

Snowflake's own documentation on CoCo describes an autonomous agent that interacts directly with a customer's Snowflake environment, planning and executing multi-step tasks with a deep understanding of that environment's schemas and RBAC. Genesis extends that same agentic approach outward, to the rest of the stack a data team actually lives in day to day.

‍

Frequently Asked Questions

Does Genesis Eve and Snowflake CoCo compete with each other? No. CoCo specializes in work inside Snowflake. Genesis connects and works the rest of the stack, including GitHub, dbt, and Jira, and can route Snowflake-specific tasks to CoCo.

Can customers choose which LLM powers Genesis? Yes. Customers can use Cortex or connect their own key to Bedrock, Anthropic, or OpenAI, and switch between them.

What is Eve in Genesis? Eve is the Chief Data Agent that scans metadata across connected systems and connected data tools to build a context graph and can hand off specific tasks to other agents like CoCo, Cortex Code, Claude Code etc.

Can multiple agents and users work in Genesis at the same time? Yes. Multiple agents can run concurrently and multiple users can collaborate in the same workspace.