Making AI agents work with real-world spreadsheets is harder than it looks. A workbook in use for a few years leans on decades of Excel behaviour, from volatile functions to a date system that still treats 1900 as a leap year and number formats that change with the locale. Close isn't good enough, because a forecast that's off by a rounding error is a wrong forecast. An agent can read a workbook, but to change an input and trust what comes out, it needs an engine that calculates the way Excel does. We wrote about what building that engine taught us in 2022.
GRID started in Reykjavík in 2018. Our first product, launched in 2021, turned spreadsheets into interactive calculators, and most of the edge cases the engine handles came from years of running customers' real models. In 2025 we were named AI Solution of the Year at the Icelandic Web Awards. Today the engine supports 492 Excel functions, and our spreadsheet technology ships inside a leading LLM platform.
We've spent our careers building data and analytics products, and relying on spreadsheet models ourselves in airline planning, finance and operational research. We've written open source libraries for parsing Excel formulas and reproducing Excel's number formatting, built an Excel add-in, and checked our engine against Excel and Google Sheets one edge case at a time.