Copilot gets hands-on in Word and Excel
Microsoft just made agentic capabilities in Word, Excel, and PowerPoint generally available, turning Copilot from a passive suggester into an active co-worker. Instead of just listing steps, it can now directly build visuals, restructure documents, and execute multi-step native edits right on your canvas. The update is now rolling out to enterprise and personal Microsoft 365 subscribers.

For the past year, using Copilot inside Office felt a bit like having an intern who only speaks in advice. You would ask it to fix a pivot table or adjust a slide deck, and instead of just doing the work, it would spit out five bullet points telling you where to click.
That dynamic is finally shifting. Microsoft announced that agentic capabilities in Word, Excel, and PowerPoint are now generally available. The feature is rolling out as the default experience for Microsoft 365 Copilot and Premium subscribers, and it is also available to Personal and Family plan users.
Basically, Copilot can now take multi-step, app-native actions directly on the canvas instead of just talking about them.
From giving instructions to doing the work
Microsoft admits that early foundation models simply were not smart enough to reliably command complex software interfaces. Copilot was stuck being a passive chatbot on the sidebar.
Thanks to recent leaps in multi-step reasoning and instruction following, Copilot can now handle real editing workflows. It understands pivot tables in Excel, citation styles in Word, and corporate slide templates and animations in PowerPoint.
Early internal numbers show users were clearly waiting for this. In Excel, weekly tries per user jumped 67%, new user retention climbed 50%, and positive feedback ratings jumped 65%.
Word saw engagement rise 52%, with retention up 11% and satisfaction up 21%. PowerPoint engagement grew 11%, while retention jumped 36% and satisfaction increased 25%.
Under the hood, Microsoft is using its Work IQ system to ground the assistant in corporate context, while adopting a multi-model approach that lets users pick different frontier models depending on the task.
Why it matters
We think this is the exact pivot Copilot needed to justify its existence in everyday work.
A chat assistant in an office suite was always a halfway solution. Telling someone how to write an Excel formula saves very little time compared to an AI that just builds the table and plots the chart for you. For everyday desk workers, this turns Copilot from an awkward search bar into an actual execution tool, and it is worth testing immediately if you have access.
Strategically, this plays right into Microsoft's biggest moat. Anyone can spin up an LLM, but deeply connecting an agent into proprietary, complex desktop application controls is an entirely different engineering challenge.
That said, the biggest hurdle remains accuracy in high-stakes workflows. Microsoft explicitly pointed out that deeper edits for legal documents and financial spreadsheets are next. When an agent touches balance sheets or contracts, even a subtle hallucination can be catastrophic. Until change-tracking and review mechanics feel completely foolproof, users will likely keep a very tight leash on it.
Sources (1)
- microsoft.com(opens in a new tab)Originalmicrosoft.com