Primary domain: Data & Analytics (Data Governance)
Secondary domain: Artificial Intelligence & Machine Learning (Agentic AI)
Article type: Product & Vendor Innovation
Technology: Snowflake CoCo for Excel and Google Sheets (data agent add-ins)
Innovation maturity: Early Adoption
Mission: Enterprise/Operations
Business impact: Lets analysts work with live governed data in spreadsheets without CSV exports or separate connectors
Audience: IT, Business/Leadership
Publication date: November 2026
Author / Contributor: Innovation team
Related technologies: Snowflake; Microsoft Excel; Google Sheets; Semantic Views; Cortex AI functions; data governance
Spreadsheets have been the main way people work with data for more than 40 years, and they are also where governed data tends to escape. Someone exports a CSV, emails it, and soon several versions of the truth are circulating. Snowflake’s answer, announced in early November 2026, is to bring its AI agent to the spreadsheet itself: Snowflake CoCo for Microsoft Excel and Google Sheets, now in public preview.
Snowflake’s release notes date the preview to November 3, and its product blog followed on November 5.
A chat pane connected to live data
According to Snowflake’s documentation, each add-in places CoCo in a chat pane inside the spreadsheet, connected to the user’s live Snowflake data. From there, users can query with natural language or SQL, build pivot tables, charts and formulas, and use the semantic views their role can access. Snowflake describes CoCo as its coding agent for data teams.
The company contrasts this with the three usual options it lists: CSV exports, ODBC connectors set up by a data team, and third-party connectors maintained by IT. Each, it says, either creates disconnected copies of data or depends on someone else to configure access, and none makes it easy to add AI.
Governance travels with the user
The governance details are the heart of the announcement. After sign-in, CoCo uses the user’s selected or default Snowflake role. Access controls, masking and row access policies apply to the results written to the sheet. CoCo previews every change before writing it, and the documentation says it records a citation to the source of each written value.
In Excel, the add-in also supports intelligent refresh. When CoCo pulls data, it saves the query text in the workbook’s settings. Users can refresh by button or with a /refresh skill, and Excel workbooks can refresh automatically when opened, which Snowflake says suits executives who want current numbers without clicking. Automatic refresh is Excel-only for now. If a query’s result size or columns change, CoCo updates the workbook while keeping existing data, formulas and assets intact.
Building real spreadsheet objects
Rather than only answering in a side panel, CoCo writes native spreadsheet objects. Snowflake says it chooses chart types based on the shape of the data, builds pivot tables, and applies conditional formatting, number formats and validation rules. Heavy aggregation is offloaded to Snowflake’s virtual warehouses, so pivots can be built on source data larger than Excel’s 1 million rows or the 20 million cells in Google Sheets. In Excel, users can also call Snowflake Cortex functions in cells, such as CORTEX.CLASSIFY and CORTEX.SUMMARIZE.
Two customers quoted by Snowflake offered early impressions. Matt Wyman of The Chefs’ Warehouse said analysts can use it without training, and Kate Savina of GTCR said there is no exporting or tool switching.
What the benchmark shows, and what it does not
Snowflake ran internal tests of 25 spreadsheet tasks across three trials and compared CoCo for Excel with an AI harness that reaches Snowflake through its MCP server. It reports similar accuracy, lower latency and fewer output tokens for CoCo. These are the company’s own tests, run October 26 to 28, 2026 in Excel for Web, not independent results.
What it means for data teams
For governance teams, the appeal is that the spreadsheet stops being a side door: the same masking and row-level rules follow the data into the workbook, and saved queries show which sources fed an analysis. For analysts, the draw is fewer manual exports. The caveats are practical. Both add-ins need a Snowflake account, they are in preview, and Snowflake says features such as spreadsheet lineage in Horizon Catalog and exporting visualizations are still to come.
The bottom line
CoCo for Excel and Google Sheets is a clear example of AI agents moving into the tools people already use, with governance built in instead of bolted on. Teams that live in spreadsheets should watch how it performs outside the vendor’s own tests.