How to use new AI agents in Word, Excel, and PowerPoint

Microsoft will introduce agentic AI in Word, Excel, and PowerPoint with technology that gives Microsoft 365 subscribers a digital assistant to write for them while they think about what comes next, drafting puff pieces — and formal contracts — from the briefest of prompts. These agents generalize Copilot beyond suggesting prompt fills to goal-driven assistance. The appeal is clear to active knowledge workers. According to the Microsoft Work Trend Index, the vast majority of early Copilot users say they’re more productive, and agents nudge that productivity deeper into everyday documents, spreadsheets, and slides.

What the new agents do across Word, Excel, and PowerPoint

Traditional Copilot answers a prompt. Set a goal with agent mode, cascade a sequence of actions throughout your file, and let the Microsoft Graph provide context-aware results on top of the files you have access to. When working in Excel, that’s selecting ranges, constructing formulas, and creating charts. In Word, that means changing structure, applying styles, and bringing in some referenced content. In PowerPoint, that’s creating from brief to complete decks, combined narratives, and visual slides alongside data-driven ones.

Who can use them and where these agents currently work

Agent capabilities are currently available to Microsoft 365 Copilot business subscribers and Personal or Family subscribers participating in public preview. Currently, Excel and Word agents are optimized for the web experience, with desktop support soon to follow. The PowerPoint experience is powered by the Office Agent in Copilot Chat on the web and initially available in English for certain regions. Administrators can oversee access and policies using familiar Microsoft 365 controls.

Excel agent setup and how to use it effectively

Begin in Excel for the web. When prompted, download the add-in and allow Agent Mode. Open a workbook and kick off Copilot. In business terms only, state your goal description, with no formulas included. For instance, “Analyze this quarter’s sales tab for anomalies, segment by region, and calculate a forecast with a confidence interval. Visualize it and explain the method you used.”

The agent will select functions, build helper sheets, and create charts. Most importantly, it reveals its reasoning steps and verifies outputs so you can examine — and massage — assumptions. If your data model is dirty, ask for it to normalize column headers, fix data types, or flag outliers before analysis. Gartner and others have been pointing out for years the time it takes to prepare data; allowing the agent to do the grunt work can give hours back per project.

Word agent setup and how to use it effectively

Open Word for the web and summon Copilot with a clear command, as well as any sources you need in your copy. An example could be: “Update this monthly report by pulling the latest numbers from the finance email thread and compare to last month’s Excel document. Use house style and bring the executive summary down to 150 words.”

The agent writes and formats while asking clarifying questions when it needs more context (like which template to apply or what data is authoritative). You can command it to “show changes in track changes,” “rewrite for a technical audience,” or “apply title case to headers and italicize partner names.” As the agent is Word feature-aware, it can manage styles, TOCs, captions, and cross-references while preserving formatting.

PowerPoint agent, what is This and how do I use it?

PowerPoint’s agent persona resides in the Office Agent in Copilot Chat. Begin a chat, and write what you’d like done, such as: “Make me a ten-slide employee briefing that improves participation in our retirement match program. You should have a compelling narrative with benchmarks, visuals, and basically an ask that makes sense.”

The agent researches and outlines designs that offer a bit of everything in consistent, balanced ways.

What the agent returns in PowerPoint

The agent offers a live preview of the deck type. You might request “three alternate title slides,” “simplification of jargon on slide five,” or “a chart comparing participation levels before and after incentives.” For heavy-research decks, write a brief that is as short as possible but disambiguates what the agent should be looking at, and attach references or point to approved docs so the agent can ground facts in your sources rather than from the open web.

Practical tips to get better results with AI agents

• Write prompts like briefs. Specify audience, results, limitations, and success criteria. Requesting “a board-ready two-pager summary, with three risks and mitigations,” is better than telling someone to “write this up.”
• Iterate conversationally. Request the agent to “show your steps,” “explain the formula choices,” or “try another way,” and then accept it or refine it. This keeps you in the loop, an important part of getting it right.
• Base the agent on your data. Cite particular files, folders, or emails the agent is able to open. McKinsey estimates knowledge workers spend a quarter of their day looking for information; grounding cuts that overhead and raises quality.
• Handle it like a junior analyst would. Farm out setup, first drafts, and eye candy to the agent and use your professional judgment. In internal pilots in various fields, teams report quicker first drafts and more time for strategy once routine work has been automated.

Security and governance considerations for Microsoft 365 AI agents

The agents respect Microsoft 365 permissions, sensitivity labels, and audit controls, and the responses are based in the Microsoft Graph rather than pulling in uncontrolled external content by default. Admins can implement data loss prevention, plug-in controls, and usage monitoring via common compliance tooling. As with any AI system, remember to scope sensitive data to the right workspaces, review outputs for accuracy, and ensure that human sign-off is necessary on decisions.

For businesses or individuals who are already in Microsoft 365, such agents might be less a novelty and more of a new mode of working. They transform the mundane office file into a collaborative surface on which you establish goals and the software does much of the heavy lifting, step by step, right there in context.