Unlike earlier Copilot functions that mainly assisted with edits or summaries, Agent Mode takes on multi-step processes, which include generating, testing, and refining outputs until they reach professional quality. 

In Excel, this means the system no longer just suggests formulas, but runs complete analyses, builds visualisations, and even fixes errors without manual intervention. Microsoft claims this is comparable to handing over work to a trained analyst while you remain in control.

Sumit Chauhan, corporate vice president of Microsoft’s Office Product Group, said: “Today we’re bringing vibe working to Microsoft 365 Copilot with Agent Mode in Office apps and Office Agent in Copilot chat. In the same way, vibe coding has transformed software development, the latest reasoning models in Copilot unlock agentic productivity for Office artifacts.”

The new experience extends beyond Excel. In Word, Agent Mode is designed to turn writing into a conversation. A user can ask Copilot to draft a monthly report, highlight insights, or reformat documents according to corporate style guidelines. Instead of just delivering a block of text, the tool asks clarifying questions and proposes refinements, creating what Microsoft calls “vibe writing.”

Office Agent, meanwhile, takes the same approach to PowerPoint and Word but directly from Copilot chat. A simple instruction, such as preparing a deck on consumer trends, prompts the system to clarify the brief, gather external data, draft slides, and show previews before handing over the final presentation. 

Chauhan noted: “It’s not just simple assistive short answers, but board-ready presentations or documents. It’s work, quite frankly, that a first-year consultant would do, delivered in minutes.”

In internal testing using the SpreadsheetBench benchmark, Agent Mode in Excel scored an accuracy rate of 57.2%, higher than competitor tools like Shortcut.ai and Claude Files, though still below human-level accuracy of 71.3%. 

Microsoft stresses that every output is auditable and refreshable, an essential safeguard given Excel’s role in critical business operations.

While OpenAI’s models continue to power Agent Mode in Excel and Word, Anthropic’s models are now responsible for Office Agent in Copilot chat. This dual approach, Microsoft argues, allows it to combine strengths from different model families depending on the task. “We are committed to OpenAI, but we are starting to explore with the model family to understand the strength that different models bring,” Chauhan said.

The features are being released gradually. Agent Mode is now available in the web versions of Word and Excel for Microsoft 365 Copilot subscribers, with desktop support expected soon. Office Agent is live in the United States for personal and family subscribers through the Frontier programme, starting with English-language support.

Productivity tools are no longer limited to editing files, they’re becoming full partners in the work process. Chauhan highlighted, “Productivity is our DNA, we’re Office. While others will try to replicate us, there is no substitute for the real thing.”

Copilot, a generative AI chatbot based on GPT-4, was first introduced as Bing Chat on February 07, 2023. It was integrated into both Bing and Edge as Cortana’s successor, but by September 2023 it was released into the enterprise as Microsoft 365 Copilot as a tool to boost enterprise productivity.

As of October 2023, Microsoft CEO Satya Nadella reported that the company had more than one million paid Copilot users across more than 37,000 companies.

By 2025, that number is estimated at hundreds of thousands of customers, according to Nadella’s official LinkedIn account.

The smart Microsoft tool’s uptake has been impressive. Integrated across Word, Excel, Teams and Outlook, it blends generative AI with Microsoft Graph data to summarise meetings, write content, and automate repetitive tasks.

The solution has fundamentally changed the narrative for companies wanting to improve workload management and optimise human talent. Companies are leveraging it to work faster and think better.

Today, Copilot is everywhere – Word, Excel, Teams, Outlook and beyond. And it does more than assist people, it collaborates with them to draft reports, summarise meetings, generate presentations, and automate workflows.

Copilot combines large language models (LLMs) with contextual awareness across calendar invites, Teams messages, SharePoint files and other data to provide users with optimised time and resource management capabilities.

The AI knows your next meeting, the prep doc you worked on, the stakeholder comments that need to feed into the document and the meeting, and the last presentation you shared.

This deep integration allows it to generate emails, write PowerPoint decks and even build Excel dashboards that provide you with recommended insights. And all this is done in minutes, not the hours usually spent on this admin-intensive work.

However, perhaps one of the most profound value-adds is how Copilot has transformed access to knowledge. It is a zero-cost interaction which allows users to retrieve anything. When the right prompts are used, users have access to infinite insights on demand.

Copilot is also no longer just reactive. Users can use Copilot Studio to create AI agents capable of executing multi-step workflows such as emailing reports, updating CRM systems or analysing Excel data.

This agentic approach to AI changes the story from AI that responds to AI that acts autonomously across integrated platforms. Agentic AI is defined by its ability to understand context, chain tasks and make decisions within defined boundaries.

You can instruct an agent to create a proposal, attach key figures to it, and then email it to the relevant stakeholders without any intervention on your side.

The process just gets done with the AI deftly managing the entire action stack. Microsoft’s Copilot Studio allows users to create these workflows with minimal coding which democratises a task that was once only possible with engineers.

While productivity is an obvious gain, there are other, unexpected benefits. You can synthesise internal knowledge as easily as Googling a fact, which is both a cultural transformation and a process improvement.

Other use cases include intelligent meeting recaps, real-time proposal drafting, and internal document search that rivals enterprise-grade search engines.

Companies are now deploying thousands of copilots just to crawl through petabytes of data because the answers they provide are invaluable.

However, the implementation of AI remains challenging. Companies have AI on their strategy list but aren’t sure where to start. Data is messy, security frameworks aren’t compliant and employees often don’t know how to prompt the AI effectively.

There is a growing need for training that takes people beyond just clicking an AI button and into the realm of understanding the right questions to ask and how to interrogate the AI correctly.

The future of Copilot and AI in general comes down to orchestration. Companies want bespoke Copilots tailored to their workflows and an agent ecosystem that solves their problems.

They also want to create a culture which embraces AI and this means building skills in curiosity, experimentation and critical thinking.

There is a danger in over trusting AI, it has to be validated, challenged and directed to ensure it is delivering value.

Winning at Copilot, or any form of agentic AI, will come down to companies being willing to learn how to ask better questions, explore the potential of the technology, and that are open to learning, failing and experimenting.

These are the routes to finding new ways of benefitting from AI’s capabilities, and ensuring it works hard for the business.