Could Excel agents unlock $1T in economic value? (martinalderson.com)

A new wave of "Excel agents"—from Shortcut and Microsoft’s Agent Mode to Claude for Excel—is turning spreadsheets into first-class interactive environments for LLM-driven automation. Unlike past workflows that required heavy copying/pasting or uploading static XLSX files to a chat, these agents work against a live Excel instance, read only the cells they need (avoiding context-window bloat), iteratively test and debug changes, and can call out to scripts or bash to verify results. They can also generate VBA or Python glue behind the scenes and coordinate subagents to fetch live API data. That matters because most companies run critical, messy “software” inside Excel with no source control, testing, or validation—making them ripe for large, low-friction uplift when non‑technical users get agent help. The economic case is striking: research estimates 38% of knowledge‑worker time is spent in spreadsheets. Applying that to ~70.9M U.S. management/professional workers yields roughly 27M full‑time equivalents; at a conservative $90k average salary this implies ~$2.4T in annual labor tied to spreadsheet work. If agents cut that effort by even 20–50%, the productivity gains easily scale into the hundreds of billions — plausibly $1T+ globally — while also reshaping roles across businesses as routine Excel tasks get automated.