In this round-up, we’ve targeted the key announcements for CoCo from Snowflake Summit. The aim is to discuss what each announcement means in practice, where we've already seen these features work, where the gotchas are, and what data leaders should prioritize in H2 2026.
This round-up has been created by Mike Droog, one of our three Data Superheroes.
Snowflake CoCo Desktop: Snowflake Just Left the Browser
Since the day Snowflake launched, the workflow has been:
- Open a browser
- Navigate to Snowsight
- Write your SQL
Every developer tool has a native desktop experience (VS Code, DBeaver, DataGrip, etc) but Snowflake's own tooling has always lived in a tab.
That changed today. CoCo Desktop is now Generally Available. Not preview. Not beta. GA, now.
What was announced
Snowflake CoCo (Cortex Code) for Desktop is now GA. This is a native desktop application that brings Snowflake's AI-powered development experience to your local machine. No browser required.
But this isn't just "Snowsight in an app". CoCo is an AI coding agent that understands your Snowflake environment, including your schemas, your tables, your queries and your pipelines. It helps you build, debug, and optimize directly from your desktop.
And it's not just hype. On ADE-Bench (a real-world analytics and data engineering benchmark created by dbt Labs), CoCo scored 72.1%, outperforming both Claude Code and OpenAI Codex (each at 65.1%). It also does it more efficiently, with 51% fewer tokens and 8% less time than Claude Code on Opus 4.7. That's not a marginal difference, it's a structural advantage from being purpose-built for data work rather than general-purpose coding.
Alongside CoCo Desktop GA, Snowflake also announced CoCo plugins for:
- VS Code — work within your existing editor
- Cursor — for AI-native development workflows
- Excel — bring CoCo intelligence to spreadsheet users
- Claude Code — integrate with Anthropic's coding agent
And the rebranding: Cortex Code is now officially Snowflake CoCo.
Why this matters for your team
- The "meet developers where they are" promise is finally real. Desktop. VS Code. Cursor. Excel. This isn't one interface, it's everywhere your team already works. The friction of "open Snowsight" disappears.
- GA means production-ready. Public preview features have caveats. GA means supported, SLA-backed, and enterprise-ready. You can roll this out to your team tomorrow without the "it's still in beta" caveat.
- Excel integration changes who can interact with Snowflake. Your finance team lives in Excel. Your ops team lives in Excel. CoCo in Excel means they can query Snowflake, build visualizations, and get AI-assisted analysis without ever learning SQL or opening Snowsight.
- Desktop performance. Browser-based tools have latency from page loads, session management, tab memory, etc. A native app is simply faster for the daily developer workflow. If you spend 4+ hours a day in Snowflake, that accumulated friction matters.
What actually sets this apart from Copilot/Cursor
CoCo is platform-aware in a way generic coding agents can't be. It knows your warehouse. It knows your permissions. It knows your data lineage and query patterns. When you ask it to help write a transformation, it doesn't guess at table names; it knows them. That context makes it meaningfully better at Snowflake-specific work than any general-purpose tool.
And with the plugin model, you don't switch tools. CoCo comes to you through VS Code, Excel, desktop, or whatever else you already have open.
For teams building on top of CoCo, the new CoCo Agent SDK packages the same tools and agent loop into an installable library. You get programmatic access to querying Snowflake, reading files, running shell commands, executing SQL, and editing code — so you can build custom applications, internal tools, and domain-specific workflows on the same engine that powers CoCo for thousands of customers.
What to do about it
If you're a data engineer:
→ Download CoCo Desktop today. It's GA. Try it for a week alongside Snowsight and see which becomes your default.
If you're a team lead:
→ Roll out the VS Code/Cursor plugins to your engineering team. Evaluate whether it reduces context-switching and improves development velocity.
If you have non-technical stakeholders:
→ The Excel plugin is your bridge. Finance teams, ops teams, and analysts who refuse to leave Excel can now interact with Snowflake data through a familiar interface with AI assistance.
If you're building Snowflake applications:
→ CoCo is now integrated into the app development workflow. You can build and deploy CoCo for Snowflake apps directly — collapsing the development cycle.
CoCo Plugins and Code Bundles: Snowflake Stops Asking You to Context-Switch
I want to talk about something that's easy to gloss over in the keynote chaos, but actually matters more for daily productivity than half the flashier announcements.
CoCo Desktop going GA was the headline. But alongside it, Snowflake quietly dropped something that changes how most people will actually interact with the platform day-to-day.
CoCo now lives in Excel, and VS Code, and Claude Code
Not as some watered-down integration. As actual plugins that bring CoCo's full Snowflake-aware context into tools people already have open. There are three plugins:
Excel — I think this one is underrated. Your finance team isn't going to learn SQL. They're not opening Snowsight. They live in spreadsheets. Now CoCo can query Snowflake, run analysis, and surface results right there. No export-to-CSV confusion. No "I'll send you a dashboard link." Just the data, in the tool they actually use.
VS Code — If you're already writing SQL or Python in VS Code (and most of us are), you now get CoCo's schema awareness, query suggestions, and Snowflake context without switching to another window. It knows your tables. It knows your permissions. It knows your recent queries.
Claude Code — This one's interesting. If your team has already adopted Anthropic's coding agent, CoCo layers in Snowflake-specific knowledge alongside it. You don't choose one or the other.
Code Bundles: finally, no more wrappers
This is a smaller announcement but one that made me genuinely happy. If you've ever written a perfectly good Python script and then spent 30 minutes figuring out how to package it as a stored procedure or UDF for Snowflake... you know the pain.
Code Bundles let you take your Python or Java file and run it directly in Snowflake. No wrapper. No DDL ceremony. No "CREATE OR REPLACE FUNCTION" boilerplate. Your code, as a file, executes.
Public Preview now.
It sounds simple. It is simple. That's the point. The gap between "it works on my machine" and "it works in Snowflake" just shrunk from 30 minutes of packaging to basically zero.
Mike's take
The CoCo plugin model is a smart move. The biggest adoption barrier for any new tool isn't functionality, it's "I don't want to switch away from what I'm already using." Plugins eliminate that. You try CoCo inside VS Code for a week. If it helps, you keep it. If not, you uninstall it. Zero commitment.
And honestly, the Excel plugin might end up being the most impactful one. Not for data engineers, but for the 80% of people in your org who need data but won't learn SQL. That's a much bigger audience than the VS Code crowd.
What we recommend
Install the VS Code plugin this week. Give it a real workload, not a toy query. See if the Snowflake-aware context actually helps your workflow or just adds noise.
For Code Bundles, next time you need to deploy Python to Snowflake, try it the new way first. Compare the time against your current approach. I suspect you won't go back.
Cortex AI Is Becoming a Platform: Functions Studio, SpaceX Models, and Agentic Search
There's been a quiet shift happening in how Snowflake thinks about Cortex AI, and this Summit made it explicit.
Cortex started as a collection of pre-built functions. SUMMARIZE. TRANSLATE. COMPLETE. Useful, but you used what Snowflake shipped; nothing custom. If you needed something specific to your business, you went elsewhere.
Three announcements at the keynote together tell a different story. Snowflake wants Cortex to be the platform you build your own AI on top of.
Cortex Functions Studio
You can now create your own AI functions. Not just call Snowflake's own, but build yours.
Your prompts. Your evaluation criteria. Your business logic. Deployed on Snowflake's infrastructure. Governed by the same RBAC you already manage. Same compute model. Same security.
I think of it like this: Snowflake used to give you a fixed menu. Now they're giving you the kitchen. The infrastructure scales the same way. The governance applies the same way. But the function does whatever your business needs it to do.
If you have a repetitive AI task — classifying support tickets, extracting entities from contracts, scoring leads — Functions Studio is how you turn that from a prompt you re-run manually into a governed, scalable, callable function.
SpaceX AI Models in Cortex
SpaceX's AI models are now available inside Cortex. Private Preview started at Summit.
The signal here isn't just "one more model." It's that Snowflake is building a model catalogue. Multiple specialized models, available natively, with your data never leaving Snowflake. The models come to your data; you don't move your data to the models.
Expect this list to grow. The architecture is clearly designed for a marketplace of models, not just the ones Snowflake builds.
Agentic Search
Search that actually thinks. Now in Preview.
Instead of writing SQL to find things, you describe what you're looking for in natural language. The system breaks your query apart, calls AI functions, searches across structured and unstructured data, and brings back results with context.
This matters most for knowledge-heavy organizations. If your data lives across 50 tables, a document store, and three semantic views (and nobody knows the right SQL to query across all of them), Agentic Search is the interface that makes that mess navigable.
What connects these three
Models (SpaceX) give you the foundation. Functions Studio gives you customization. Agentic Search gives you the interface. Together, they turn Cortex from "Snowflake's AI features" into "the platform where you build AI features."
Data leaders, you can now build AI capabilities custom to your business, on models you choose, accessible to users who don't know SQL, all inside the governance framework you're already running. That's a meaningfully different value proposition than "here are some pre-built functions."
What we recommend
Pick one AI task your team does repeatedly and build a custom function in Functions Studio. Just one. See how it compares to your current approach (probably a Python script or a prompt that someone re-runs manually). If the governed, scalable version is better, and I suspect it will be, you'll find ten more use cases quickly.
Mike Droog is a Data Superhero and Solution Architect at Aimpoint Digital, a Snowflake partner helping teams build custom AI solutions on Snowflake. If your team is exploring custom AI functions or agentic capabilities, we can help.
