Copilot Cowork: Microsoft 365's New Agentic AI

Microsoft 365 Copilot has always been good at answering questions and drafting a paragraph. Where many teams have struggled is the next step: the multi-hour, multi-app work that sits between a prompt and a finished deliverable. That is the gap Copilot Cowork is designed to close.

This guide introduces Copilot Cowork from scratch, explains where it fits in a business productivity stack, and covers the enterprise security controls Microsoft has wrapped around it.

Microsoft Copilot Cowork analysing a bike share rental spreadsheet, producing key driver metrics, charts and an executive summary in a single agent run
Copilot Cowork working through a spreadsheet analysis task: reading the file, building visualisations and writing an executive summary across a single agent run.

What Is Copilot Cowork?

Cowork is a new way of using AI for work. Rather than asking a question and receiving a single response, you describe an outcome. The AI then plans the work, breaks it into steps, carries it out across the applications and files it needs, and checks in with you along the way.

The idea began with Anthropic, the company behind Claude. In January 2026 they launched Claude Cowork, a desktop agent that worked with files on your computer. The model was popular enough that Microsoft, already a significant Anthropic customer, partnered with Anthropic to bring the same agent technology into Microsoft 365.

The result is Copilot Cowork: the same agentic approach, now running inside the Microsoft 365 environment your organisation already uses, grounded in your work data, and subject to your existing identity, permission and compliance controls.

Copilot Cowork was announced on 9 March 2026 and became available through Microsoft's Frontier program on 30 March 2026. It is rolling out to enterprise customers progressively through 2026.

How Copilot Cowork Differs From Regular Copilot

Most organisations using Microsoft 365 Copilot today use it as an assistant. You ask it to summarise a document, draft an email, or analyse a spreadsheet, and it responds in a single turn. That is useful, but it sits inside one app and one prompt at a time.

Copilot Cowork changes three things:

Change 01
Multi-step execution

A single request can trigger a plan that runs across Outlook, Teams, Excel, Word and PowerPoint, over minutes or hours, producing multiple deliverables.

Change 02
Grounded in your context

Through a layer called Work IQ, Cowork draws on signals from your emails, calendar, Teams chats, SharePoint files and relationships, so its plan reflects how your organisation actually operates.

Change 03
Outputs stay in M365

Cowork generates files directly into Word, Excel and PowerPoint, stored in OneDrive or SharePoint, rather than downloading local copies that create version sprawl.

A helpful way to think about it: regular Copilot is reactive. You ask, it answers. Cowork is agentic. You describe the outcome, it figures out the steps and carries them through, checking in before anything consequential happens.

How Work IQ Makes Copilot Cowork Practical

Cowork would be much less useful if you had to spoon-feed it context for every task. Work IQ is the layer that solves this. It sits on top of Microsoft Graph and provides Cowork with contextual signals from across your Microsoft 365 tenant: emails, calendar entries, Teams transcripts, SharePoint documents, and the relationships between people and projects.

The practical effect is that when you ask Cowork to "prepare for the Q2 customer meeting," it can find the relevant customer emails, pull the last three meeting notes, check the calendar for who is attending, and draw on the right SharePoint folder, without you having to link to any of it manually.

Work IQ only surfaces data the user is already permitted to access. If someone cannot see a SharePoint site in their normal work, Cowork cannot use it on their behalf either.

Copilot Cowork Business Productivity Use Cases

Microsoft demonstrated four flagship scenarios at launch, and enterprise early adopters like Capital Group have added several more. The common thread is work that spans multiple apps, takes real time to complete, and produces a concrete deliverable at the end.

1. Proposal Generation

Writing a client proposal is one of the most common time sinks in professional services and sales, and one of the best fits for Copilot Cowork. The building blocks usually exist already: a discovery call transcript in Teams, a proposal template in PowerPoint or Word, product and pricing information in SharePoint, a brand voice guide that defines how your organisation sounds, and a customer website that tells you what they care about.

A typical request looks like this: point Cowork at the Teams recording of the discovery call, reference your branded proposal template, link to the product information and brand voice guide in SharePoint, and provide the customer's website URL. Cowork then reads the transcript to extract the customer's priorities and language, pulls the relevant product details, reviews the customer's website to understand their business context, applies your brand voice to every section it writes, and assembles a proposal in your branded slide deck that reflects the specific conversation you had.

The draft is a starting point, not finished work. The value is that a senior person can spend their time refining positioning and commercials rather than assembling the first draft from scratch. Organisations that run on repeatable proposal patterns tend to see the largest gains here.

Inputs
Source 01
Teams call recording
Source 02
Branded proposal deck
Source 03
SharePoint product info
Source 04
Brand voice guide
Source 05
Customer website URL
The engine
Microsoft 365
Copilot Cowork
  1. Reads the transcript
  2. Extracts priorities and language
  3. Pulls the right product details
  4. Reviews the customer website
  5. Applies your brand voice
  6. Assembles the branded deck
Output
Deliverable
Your branded proposal
  • Slide 1: Cover
  • Slide 2: Your challenge
  • Slide 3: Proposed solution
  • Slide 4: Approach
  • Slide 5: Timeline
  • Slide 6: Investment
  • Slide 7: Why us
  • Slide 8: Next steps

Saved to OneDrive. Ready for review.

Work IQ grounds every step in your M365 tenant; a human refines positioning and commercials at the end.

2. Meeting Preparation

You ask Copilot Cowork to prepare for an upcoming customer meeting. It reads the relevant email threads, pulls prior meeting notes from OneNote or Teams, checks calendar context for who will be there, and produces a briefing document in Word, a presentation in PowerPoint, and an Excel file with the customer metrics you typically review. Everything saves directly to OneDrive.

Early Frontier customers have highlighted this as the most immediate productivity win. What used to take two hours of copy-paste across four applications becomes a single request with a review pass at the end.

3. Calendar Triage and Focus Time

Most weeks start with a packed calendar and not enough time for deep work. Cowork reviews your Outlook schedule, asks what you want to prioritise, flags conflicts and low-value meetings, and proposes a reworked week. After your approval, it accepts, declines or reschedules meetings and blocks focus time automatically.

4. Monthly Budget Reviews

A repeatable finance workflow that used to require manual assembly each month. Cowork pulls figures from the relevant Excel files, compares them against previous months, drafts the commentary in Word, prepares the summary slides in PowerPoint, and sends the pack to stakeholders. With scheduled tasks, you can set this to run on the first of each month.

5. Project Status and Executive Briefings

Cowork can pull progress signals from Teams channels, project documents in SharePoint, and meeting notes, then assemble a status report or executive briefing in the format your organisation expects. Useful for leaders who spend hours each week turning scattered updates into a single coherent document.

6. Research and Knowledge Synthesis

Through the Researcher agent in Copilot, now powered by Claude models in addition to Microsoft's own, Cowork can combine web research with internal SharePoint content to produce grounded analysis. A new "Critique" feature pairs one model to generate and another to review for source reliability, which is particularly useful in regulated settings.

The Copilot Cowork Security and Governance Layer

This is where Copilot Cowork differs most clearly from Anthropic's original Claude Cowork, and where it matters most for organisations evaluating agentic AI at scale.

Where Claude Cowork is a desktop product that sits outside organisational boundaries, Copilot Cowork is built to run inside them. Microsoft has integrated it with the identity, compliance and governance controls that enterprise IT teams already rely on.

Identity and Permissions

Copilot Cowork operates under the signed-in user's identity and permissions at all times. It cannot access files, mailboxes, SharePoint sites or Teams conversations that the user cannot access directly. If someone's permissions change, Cowork's access changes with them.

Sandboxed Cloud Execution

Cowork runs in a protected, sandboxed cloud environment within your Microsoft 365 tenant. Data stays inside the Microsoft 365 service boundary and is covered by Microsoft's Enterprise Data Protection commitments. Work in progress continues safely as users move between devices.

Human-in-the-Loop Checkpoints

Cowork produces an explicit plan before it starts and shows checkpoints throughout. Actions that send emails, reschedule meetings, or change files are proposed for approval rather than taken silently. Users can pause, redirect, or cancel at any point.

Auditable Actions and Outputs

Every action Cowork takes and every output it produces is auditable. Administrators can trace which files were read, which tools were called, and what changes were made. This integrates with Microsoft Purview for compliance reporting and with Microsoft Defender for security monitoring.

Admin Controls and Data Loss Prevention

Administrators manage Cowork through the Microsoft 365 admin centre. Controls include:

  • Enabling or disabling Cowork at the tenant level
  • Scoping access by user, role or group
  • Applying Purview Data Loss Prevention (DLP) policies to Cowork prompts and outputs
  • Restricting which sensitive data types can be used for grounding
  • Blocking web search grounding while allowing internal Work IQ grounding
  • Using Insider Risk Management to detect patterns of inappropriate usage

Anthropic as a Subprocessor

Because Cowork uses Claude models from Anthropic, administrators must explicitly enable Anthropic as a subprocessor in the Microsoft 365 admin centre before their users can access Cowork or Claude-powered features like Researcher. This is a deliberate opt-in, giving IT and security teams a clean decision point.

What Cowork Does Not Do

Microsoft has been clear about current limitations. Cowork runs within the Microsoft 365 boundary, so data stays inside your tenant. It does not download files to local machines, does not create version sprawl, and does not move data outside the compliance perimeter. For regulated industries, this is a materially different risk profile than tools that copy data to external services for processing.

Copilot Cowork vs Claude Cowork vs ChatGPT Agent

A reasonable question from anyone evaluating agentic AI today: why Copilot Cowork rather than Claude Cowork, ChatGPT Agent, or the other emerging options?

The honest answer depends on context. For an individual knowledge worker or a small team operating outside an enterprise IT perimeter, Claude Cowork on the desktop is often simpler, cheaper to start with, and works against local files. For organisations already standardised on Microsoft 365 and governed by enterprise compliance requirements, Copilot Cowork fits the environment that is already in place: identity, DLP, auditing, sensitivity labels, retention and admin controls all carry over automatically.

Many AI pilots stall not because the technology failed, but because adoption required a parallel identity stack, a separate compliance story, and a new approval process for every document the tool touched. Copilot Cowork avoids that by operating inside the governance framework the organisation already runs on.

Agent Comparison at a Glance

Desktop agent
Claude Cowork

by Anthropic

Environment

Local desktop (macOS, Windows)

Data access

Files you authorise

Compliance

Not integrated with enterprise tooling

Auditability

Local history only

Best for

Individuals and small teams

Web agent
ChatGPT Agent

by OpenAI

Environment

Web and connected services

Data access

Web browsing and connectors

Compliance

Limited enterprise integration

Auditability

Limited

Best for

Web research and task automation

Enterprise agent
Copilot Cowork

by Microsoft

Environment

Microsoft 365 cloud

Data access

Work IQ across the tenant

Compliance

Purview, Defender, DLP, EDP

Auditability

Full tenant audit trail

Best for

Enterprises on M365 needing governed AI

Feature Comparison

Capability Claude Cowork ChatGPT Agent Copilot Cowork

Recommended for enterprise
Primary environment Local desktop (macOS, Windows) Web and connected services Microsoft 365 cloud
Works with Local files, folders, MCP connectors Web services, limited file handling Outlook, Teams, Word, Excel, PowerPoint, SharePoint, OneDrive
Data grounding Local files you authorise Web browsing and chat history Work IQ across the Microsoft 365 tenant
Identity and permissions User-level access on local machine OpenAI account Microsoft Entra ID identity and permissions
Compliance integration Not integrated Limited Purview, Defender, Enterprise Data Protection
Auditability Local conversation history Limited Full audit trail via M365 admin centre
DLP support No Limited Yes, via Purview DLP policies
Admin controls Individual installation Account-level Tenant-level enable/disable, role scoping
Typical licence Claude Pro, Max, Team or Enterprise ChatGPT Plus or Enterprise Microsoft Copilot licence, including related business and enterprise bundles

Practical Considerations for Australian Businesses

A few points worth thinking through before enabling Copilot Cowork in your organisation.

Data foundation matters more than the model

Cowork is only as useful as the content it can ground on. Well-governed SharePoint, sensible sensitivity labels and clean permissions produce stronger results than sprawling access and scattered documents.

Start with specific, repeatable workflows

The highest-value early use cases tend to be tasks people already do regularly: monthly reports, meeting prep, status updates, customer briefings. These are easier to evaluate against current effort.

Train people on the approval workflow

Cowork's value depends on users reviewing plans and checkpoints properly. A rushed approval is as risky as no approval. Treat checkpoints as real decision points, not formalities.

Frequently Asked Questions About Copilot Cowork

Is Copilot Cowork the same as Claude Cowork?

They share underlying technology from Anthropic, but they are different products. Claude Cowork is a desktop agent that works with local files. Copilot Cowork runs in the Microsoft 365 cloud, grounds itself in your work data via Work IQ, and operates under Microsoft's enterprise security framework.

Does Copilot Cowork use my data to train models?

Data stays within your Microsoft 365 tenant under Enterprise Data Protection. Microsoft does not use customer data from Copilot or Cowork to train foundation models.

Can admins turn Copilot Cowork off?

Yes. Administrators control Cowork access through the Microsoft 365 admin centre, including enabling or disabling it at the tenant level and scoping access by user, role or group.

How does Copilot Cowork handle sensitive information?

Cowork respects existing Microsoft 365 permissions, sensitivity labels and Purview DLP policies. Administrators can block specific sensitive data types from being used in prompts or grounding.

Is Copilot Cowork available in Australia?

Cowork is rolling out progressively through 2026 to Microsoft 365 Copilot customers. Australian organisations on the Frontier program have access, and broader availability follows the standard Microsoft 365 rollout pattern.

What is the difference between Copilot Cowork and Agent 365?

Copilot Cowork is the built-in agent that works across Microsoft 365 on your behalf. Agent 365 is Microsoft's platform for building custom AI agents, analogous to how Power Platform relates to Microsoft 365. Many organisations will use both: Cowork for general knowledge work, Agent 365 for bespoke workflows.

Where Hypergen Fits

Adopting agentic AI inside Microsoft 365 is less a software installation and more an organisational change exercise. The technology is ready. What usually determines whether value lands is the quality of the data foundation, the clarity of the governance model, and the depth of user capability building.

Hypergen helps Australian organisations think through those questions in a practical, business-first way. Our work spans Microsoft 365 Copilot training and adoption, Copilot Studio and Power Platform automation, and custom AI development where the problem calls for it. We are platform-agnostic, but we have deep experience with the Microsoft ecosystem and the governance questions that come with it.

For teams already running Copilot, our M365 Copilot support packages are a practical way to draw down expert hours as you explore Cowork use cases, build custom agents, and get more from prebuilt agents like Analyst and Researcher. If you missed our earlier piece on the product landscape, see why we're now recommending Microsoft Copilot.

Evaluating Copilot Cowork for your organisation?

We are happy to talk through the relevant considerations, from governance to data foundation to practical use cases for your context.

Talk to Hypergen
Explore M365 Copilot support