Governed Semantic Views for CoCo, CoWork, Power BI & Excel

How Snowflake Semantic Views reach CoCo, CoWork, Power BI, and Excel without leaving the platform

Executive Summary

At Snowflake Summit 2026, Snowflake and AtScale introduced a new way to extend governed Snowflake Semantic Views to CoCo, CoWork, Power BI, and Excel without copying data or rebuilding metric logic. AI is only as trustworthy as the business definitions behind it, making semantic governance a foundational requirement for enterprise AI.

Why Governed Semantics Matter for AI

This week at Snowflake Summit, Snowflake unveiled CoCo and CoWork, autonomous agents that write code and answer business questions on their own. Each one is only as trustworthy as the definitions underneath it. At the same time, AtScale announced a new product, built with Snowflake, to ground them.

We all understand the challenge of making AI accurate; we watch it hallucinate answers every day. The stakes climb when the question is a critical business one, where the right answer depends on the rules that tell an AI what “revenue” or “gross margin” actually mean.

Two customers have cracked this code; they were on my mind all week. Carrefour migrated 3,000 KPIs to a universal semantic layer. Blue Yonder gave up its identity as a BI team and rebuilt around a semantic-first strategy. Both stories start the same way, with metrics nobody could fully trust.

Snowflake, one of the largest data platforms on earth, just put its weight behind the same idea. More on that endorsement, and the partnership we announced this week, at the end.

What a Semantic-First Strategy Looks Like

A semantic-first strategy means governing your metrics in one model and serving them to every tool and agent, instead of locking them inside a single platform where they can’t travel. That lock-in is what creates drift. Nicolas Treanton, Head of Enterprise Analytics at Carrefour, hit that wall before migrating to AtScale, as he explained in his talk at the Semantic Layer Summit last month. His prior semantic layer served 1,000 users, 26 models, and 3,000 KPIs, and he still called it “a closed box.” Most BI tools, catalogs, and metadata layers work this way; they hold your definitions hostage.

Nicolas’s team freed those definitions from the BI tool and moved them to AtScale. One live, non-extracting architecture now serves consistent semantic context to every AI project and every legacy BI report. The closed box is open.

Blue Yonder made the same shift when it stopped being a BI team. Jeremy Arendt, Senior Director of Analytics Engineering, described it in his SLS ’26 session. Before the change, the company’s metric logic was “scattered and duplicated,” which made governance structurally impossible.

Blue Yonder rebuilt 800+ tables into one governed dimensional model and connected it to AI tools through AtScale’s MCP server. A multi-day analyst workflow collapsed into a single query. As Jeremy put it:

“It’s not about laying AI on top of BI dashboards. It’s about what becomes possible when our infrastructure is semantic first.”

A semantic-first strategy also slashes the cost of AI. When a model has to guess how to compute a metric, it scans far more data than it should. In one Tier 1 bank benchmark, routing the same queries through AtScale’s semantic layer instead of an unguided LLM cut cost by 99.995%, from $17.93 to $0.0008, and lifted accuracy from about 20% to 92.5%, and for context engineered queries, 100%. Governed makes AI both accurate and affordable.

What AtScale and Snowflake Announced at Snowflake Summit 2026

At Snowflake Summit, AtScale and Snowflake announced the Snowflake Semantic Views XMLA Endpoint, powered by AtScale (learn more at snowflake.atscale.com), available in private preview soon. It turns Snowflake Semantic Views into live XMLA endpoints for Power BI and Excel, so business users query the same governed metrics that CoWork and CoCo use, without copying data, rebuilding logic, mirroring, or breaking Snowflake’s role-based access controls. It deploys as a Snowflake Native App with a single DDL command.

Snowflake is putting money and engineers behind the same bet. Snowflake Ventures led AtScale’s largest equity financing to date, a sign that governed semantic logic belongs where the data lives, with no extracts or duplication.

“Snowflake customers should not have to move or duplicate data to give business users access to governed metrics in the tools they already use,” said Josh Klahr, Product Manager at Snowflake. “With the Snowflake Semantic Views XMLA Endpoint, powered by AtScale, customers can extend trusted business definitions to Power BI and Excel while keeping Snowflake as the foundation for consistent AI and analytics.”

How One Definition Reaches CoCo, CoWork, Power BI, and Excel

One governed Snowflake Semantic View grounds all four surfaces, so they return the same number. Picture a retailer’s finance team. An analyst defines gross margin once as a Semantic View. The data engineer asks CoCo, Snowflake’s coding agent, to build a margin-by-region pipeline, and CoCo reads that view and inherits the logic. The VP of Finance asks CoWork, Snowflake’s personal agent for knowledge workers, why margin slipped in the Northeast, and gets an answer built on the same definition. The board dashboard in Power BI and the team’s Excel pivot, both live through the AtScale XMLA Endpoint, show that exact number. One definition. Four surfaces. Same answer.

Snowflake renamed Cortex Code to CoCo and turned Snowflake Intelligence into CoWork at this year’s Summit, and both agents are only as trustworthy as the definitions beneath them. CoCo queries the relevant Semantic View before it ever touches a raw table. CoWork answers in plain language grounded in those same definitions. The XMLA Endpoint carries the identical logic to Power BI and Excel through live connections, with no mirrored data and no metric rebuilt in a second tool where it drifts.

When a team needs more, rolling 13-month trends, a product hierarchy, legacy Tableau dashboards, or compute in Databricks, Enterprise AtScale for Snowflake adds it without moving the foundation off Snowflake. The full universal semantic layer includes:

Complex metric modeling, hierarchies, and time intelligence
Git-backed development
Advanced DAX and MDX calculations
Multi-cloud flexibility
MCP connectivity that brings governed business definitions to Claude, ChatGPT, or any custom AI agent

Both editions keep data gravity, compute, and business logic in Snowflake, and an independent, standards-based, high-performance path to sharing those semantics with an LLM. Any AI cowork tool. Any BI tool. Your choice. All with accurate, optimized compute.

Why Portable Semantics Is the Foundation for Enterprise AI

Portable semantics is the foundation for enterprise AI, and four organizations proved it from four directions. Carrefour freed its definitions from a closed BI tool. Blue Yonder stopped being a BI team and rebuilt on a semantic foundation. A Tier 1 bank showed that going without one lets AI run up an $8.7 million guessing tax in wasted compute and wrong answers. And Snowflake, one of the largest data platforms on earth, put its capital and its roadmap behind the same idea.

“AI did not create the metrics consistency problem. It exposed it,” says AtScale CEO Chris Lynch. “Agents won’t get the benefit of the doubt.” Ask CoWork for gross margin, then ask Power BI the same question. If the answers disagree, you don’t have an AI problem. You have a semantics problem.

AI is only as good as the business context it runs on. Define that context once, govern it centrally, and make it portable enough to serve every tool and every agent, from CoCo and CoWork to Claude, ChatGPT, Power BI, Excel, and Tableau. Lock it inside one platform and you pay twice, in duplicated logic and in the guessing tax.

So the question for every data leader is which one you’re building. Your semantic layer is either an open asset that travels with your business, or a closed box that taxes every AI decision you make.

Key Takeaways

AI agents are only as trustworthy as the business definitions they use.
A semantic-first strategy enables consistent metrics across analytics and AI environments.
Carrefour unified 3,000 KPIs under a portable semantic architecture to eliminate platform lock-in.
Blue Yonder rebuilt its analytics strategy around governed semantic definitions and AI-ready infrastructure.
A Tier 1 bank demonstrated significant improvements in AI accuracy and dramatic reductions in compute costs through semantic governance.
The Snowflake Semantic Views XMLA Endpoint extends governed Snowflake Semantic Views to CoCo, CoWork, Power BI, and Excel without duplicating data or logic.
Portable semantics provide a foundation for trusted enterprise AI across tools, agents, and business users.

To see what a universal, standards-based semantic layer looks like on Snowflake, visit snowflake.atscale.com.