Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableArs Technica→ original

Microsoft builds Word, Excel and PowerPoint into Copilot and launches the Autopilot agent

At Ignite, Microsoft moved Word, Excel and PowerPoint inside Copilot, which Nadella called a “new OS for work.” Instead of opening Word, you ask Copilot to put together a client report. The agent builds the document itself, and you can then edit the file in regular Office. A separate agent, Autopilot, takes on not a one-off task but a whole role, such as managing email and calendar. For companies, the more important decision now is how much authority to hand the agent.

AI-processed from Ars Technica; edited by Hamidun News
Microsoft builds Word, Excel and PowerPoint into Copilot and launches the Autopilot agent
Source: Ars Technica. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceArs Technica ↗
Editor-in-Chief
Key metric #1Autopilot is billed through Copilot Credits; pricing has not been announced yet
Key metric #2Each agent has its own ID, and an admin can restrict its permissions or block it
Key metric #340+ companies, including Adobe and Mastercard, are working on AI security standards

At its Ignite conference, Microsoft announced general availability of Word, Excel and PowerPoint inside Copilot, its AI assistant, which CEO Satya Nadella called a “new OS for work.” An example makes it easiest to understand. Before, you opened Word, wrote a client report and occasionally called on an assistant for help.

Now you give the task to Copilot itself. It pulls context from your email, meetings and files, creates the document, spreadsheet or presentation on its own, and you can then refine the finished file by hand in the familiar Word, Excel or PowerPoint. The office apps become plugins, and an agent, an AI that acts on its own, takes the wheel.

At the same time, Microsoft is launching the Autopilot agent, to which you can delegate entire responsibilities rather than individual requests. If you are choosing enterprise software or building products on top of AI, the main question is no longer “which model is smarter” but “how much authority are we ready to give the agent.

Office as a plugin for the agent

Nadella described what is happening as an inversion of the familiar setup: the app used to come first and the assistant was secondary, and now it is the other way around. Apps become part of the agent’s work, which draws on an enterprise knowledge layer covering emails, meetings, calendars, chats and files.

That layer is provided by Work IQ, part of the Microsoft IQ platform, which has also become generally available. Nadella described Work IQ as possibly a company’s most valuable database: it holds people, their relationships, emails, meetings and projects. For an agent, it is a kind of map of how work is organized inside the organization. Each company decides for itself, after review by its own IT department, whether autonomous software will read internal documents and correspondence.

Autopilot is a cloud agent, described as a “persistent and proactive” assistant for personal productivity. You can authorize it to work with your email and calendar, and it will carry out tasks within the role it has been given. Payment is usage-based through Copilot Credits, but final prices have not been announced ahead of general availability. The shared Copilot Workspaces environment is so far available only in the Frontier early access program. Later, you will be able to bring personal Autopilot agents into it. According to Nadella, communication with AI has so far been “single-user,” and Workspaces makes it multi-user: people and agents work toward a common goal with shared context.

“You can’t trust it” — and what to do about it

The most unexpected part of the presentation was what Nadella said about trust. He acknowledged that aligning models’ goals with human interests and understanding how they work internally remain unsolved problems. Asked how one can trust such a technology at all, he answered: “by definition, you can’t.” Microsoft sees the solution in architecture: separate the “supply of intelligence” from the authority it is granted.

In practice, each Autopilot agent will have a unique ID. An administrator will be able to restrict an agent’s permissions or block it entirely. All actions are recorded in a log for human review, and the agent may ask for confirmation before certain operations. In addition, Microsoft launched the AI Transparency Initiative, which more than 40 large companies, including Adobe and Mastercard, have joined. The goal is to develop open industry standards for AI governance and security.

Ars Technica rightly notes that Microsoft is asking businesses to believe in its own guardrails, and how well they will protect against errors, unsanctioned agent actions and data leaks has yet to be shown. The company’s logic is this: unreliable intelligence does not need to be made reliable, it is enough to control what it is allowed to do.

What the agent costs, and what about the outage

Agent 365, the control panel for AI agents in the enterprise environment, is also expanding. It extends the familiar identity, security and administration systems to agents, and Microsoft promises “observability at every level.” Nadella compared the new Agent 365 Business Analytics module to an ERP system for AI use: you can see exactly what drives token costs and match them against business results.

There is an important caveat: the estimate of working hours saved is calculated by a model, not measured directly. For companies deciding whether AI pays off, this is an honest warning: an attractive savings figure remains an estimate. Meanwhile, Judson Althoff, CEO of Microsoft’s commercial business, said that everything you can do with Claude or ChatGPT you can do better and cheaper in Copilot.

Nadella only briefly mentioned last week’s outage, which disrupted access to Exchange Online and Teams worldwide: it is a “reminder of how critical digital infrastructure is,” and “we have work to do.” Asking organizations to give agents more autonomy at a moment when there are more questions about Microsoft’s infrastructure is a bold move. For developers and product teams, the takeaway is practical: build in separate identifiers for agents, action logs and human confirmation for risky operations, and trust estimated savings figures only after real measurements.

Story

Key Questions Answered

What are the key details on Office as a plugin for the agent?

Nadella described what is happening as an inversion of the familiar setup: the app used to come first and the assistant was secondary, and now it is the other way around. Apps become part of the agent’s work, which draws on an enterprise knowledge layer covering emails, meetings, calendars, chats and files. That layer is provided by Work IQ, part of the Microsoft IQ platform, which has also become generally available.

What are the key details on “You can’t trust it” — and what to do about it?

The most unexpected part of the presentation was what Nadella said about trust. He acknowledged that aligning models’ goals with human interests and understanding how they work internally remain unsolved problems. Asked how one can trust such a technology at all, he answered: “by definition, you can’t.” Microsoft sees the solution in architecture: separate the “supply of intelligence” from the authority it is granted. In practice, each Autopilot agent will have a unique ID.

What the agent costs, and what about the outage?

Agent 365, the control panel for AI agents in the enterprise environment, is also expanding. It extends the familiar identity, security and administration systems to agents, and Microsoft promises “observability at every level.” Nadella compared the new Agent 365 Business Analytics module to an ERP system for AI use: you can see exactly what drives token costs and match them against business results. There is an important caveat: the estimate of working hours saved is calculated by a model, not measured directly.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…