$30 per month. That's what Microsoft is charging for Copilot Cowork — on top of your existing Microsoft 365 subscription. Annualized per employee, that's roughly $360. The promise: an AI agent that handles multi-step work autonomously. I've spent two weeks researching whether that's worth the price since the March 9th announcement.
TL;DR
- Copilot Cowork is an AI agent that autonomously executes multi-step tasks inside Microsoft 365
- It integrates Anthropic's Claude model for handling long-horizon, complex work
- Works across email, calendar, Teams, Excel, and the full M365 suite
- Work IQ builds a model of your working context (emails, meetings, files) and informs every action
- Agent 365 ($15/user/month) is a management and governance add-on; E7 ($99/user/month) is the all-in bundle
- Currently in Research Preview, expanding to the Frontier program in late March
What Is Copilot Cowork?
The original Copilot handled one-shot requests: "Summarize this email." "Draft a report." One input, one output.
Copilot Cowork is different. Tell it "Prepare for next week's customer meeting" and it:
- Scans Outlook for relevant email history
- Reviews Teams conversations for customer context
- Finds relevant proposals and files in OneDrive
- Drafts a briefing document
- Generates a competitor comparison table in Excel
- Creates a PowerPoint deck for the customer meeting
- Blocks preparation time on your calendar
All of this runs in the background over minutes to hours. At defined checkpoints, Cowork surfaces for human review and can accept direction changes before proceeding.
Work IQ: The "Context-Aware AI" Core
The key enabler here is a system called Work IQ.
Work IQ continuously collects work signals across the entire M365 suite — Outlook, Teams, Excel, SharePoint, and more. It's not just reading file contents; it builds a model of who you communicate with, in what context, and which projects currently have your attention.
As I covered in my post on the reality of enterprise AI adoption, the biggest barrier to AI in enterprise has been "context gap" — AI that doesn't know your organization's situation produces generic results. Work IQ attacks this problem head-on using M365 data.
Work IQ vs General LLMs
| Dimension | General LLM (ChatGPT, etc.) | Copilot Cowork + Work IQ |
|---|---|---|
| Context source | User must paste it in | Auto-collected from M365 |
| Action scope | Text generation only | Sends email, edits calendar, creates files |
| Task duration | Single turn (immediate) | Minutes to hours (background) |
| Security boundary | Data sent to external servers | Runs within M365 security perimeter |
| Audit trail | None | Full audit log of every action |
Three Real-World Scenarios
Scenario 1: Weekly Calendar Cleanup
Prompt: "Reorganize this week's calendar. Protecting focus time is the top priority."
What Cowork does:
- Scans the full week in Outlook Calendar
- Identifies lower-priority meetings (based on attendee count, recurrence, missing agendas)
- Detects scheduling conflicts
- Generates a proposed change list → awaits user approval → applies changes (accept/decline/reschedule)
- Automatically adds Focus Time blocks
This alone could easily save 30 minutes every Monday. Calendar hygiene is a universal pain point in knowledge work.
Scenario 2: Competitive Analysis + Sales Deck
Prompt: "Before our meeting with Company A, run a competitive analysis and create presentation materials."
What Cowork does:
- Pulls relevant email and Teams conversation history about Company A
- Auto-generates a feature comparison table in Excel
- Identifies differentiation points and drafts a value proposition document
- Generates a PowerPoint presentation deck
- Organizes all related files into a single folder
Scenario 3: Project Kickoff Preparation
Prompt: "Set up the kickoff meeting for Project X."
What Cowork does:
- Reviews project-related documents and emails to extract goals and scope
- Drafts a project brief with milestones, owners, and next steps
- Generates a kickoff meeting agenda
- Prepares pre-read materials for distribution to the team
Agent 365 and E7: What's the Difference?
Microsoft announced several products simultaneously, which can be confusing. Here's the breakdown:
| Product | Price | What's Included | GA Timeline |
|---|---|---|---|
| Copilot Cowork | Included in M365 Copilot | Multi-step agent execution | Late March (Frontier) |
| M365 Copilot | $30/user/month | Cowork + existing Copilot features | Already GA |
| Agent 365 | $15/user/month (add-on) | Agent management, security, governance layer | May 2026 |
| M365 E7 | $99/user/month | Copilot + Agent 365 + Entra + E5 security bundle | May 2026 |
Agent 365 is the layer that lets IT administrators control which agents can access which data, and view complete audit logs. In enterprise environments, this isn't optional — without it, deploying Cowork in production is a governance risk.
What the Anthropic Claude Integration Means
What's notable is that Microsoft chose Anthropic's Claude — not its own GPT models — for Cowork. A company that has invested billions in OpenAI.
This tells us a few things.
First, Microsoft's internal evaluation concluded that Claude's strengths — long context handling, instruction following, safety — made it a better fit for multi-step agentic execution than GPT. Claude has demonstrated these strengths in MCP (Model Context Protocol) deployments as well.
Second, the AI model market is shifting from "bet on one company" to "multi-model." Even Microsoft, OpenAI's most committed backer, is choosing the right model for the job rather than defaulting to one supplier.
Developer Opportunities
1. Custom Agent Development
Cowork supports custom agents that run on top of its framework. With Microsoft Graph API integration, developers can build agents that connect third-party services to the M365 ecosystem — a real commercial opportunity as enterprises adopt this.
2. MCP Synergy
MCP has crossed 97 million installs, and agent interoperability is becoming a reality. Copilot Cowork is part of this trend. Developers who can build MCP servers will find increasing demand.
3. Security and Governance Tooling
Agent 365's existence signals that AI governance — controlling what agents can do and auditing what they did — is becoming a product category in its own right. Building in this space is a legitimate opportunity.
An Honest Assessment
This post would be incomplete without the counterpoints.
The price is real. $30/month is meaningful for individual developers, and for organizations, deploying to a full team is expensive. Add Agent 365 and you're at $45/user/month — $54,000 per year for a 100-person team.
It's still Research Preview. The pattern of "impressive demos, disappointing production" is well-established in enterprise software. How stable this runs in real organizational environments remains to be seen.
Data privacy concerns are legitimate. Work IQ reading your email, meetings, and messages is the point — but it also means an AI model has access to very sensitive data. Organizations with strong data sovereignty requirements may have valid reasons to hesitate.
M365 lock-in. This tool is only meaningful if your organization runs on Microsoft 365. For Google Workspace or Slack-centric teams, it's irrelevant. (For what it's worth, Google is pursuing a similar angle with Workspace Studio.)
How to Get Started
To access Copilot Cowork today:
- Requires a Microsoft 365 Copilot license ($30/user/month)
- Join the Frontier program (expanding in late March)
- IT admin activates it in the Copilot admin center
- Access the Cowork panel from Microsoft 365 apps (Outlook, Teams, etc.)
General availability isn't here yet. When Agent 365 and E7 GA in May, access will broaden significantly.
For developers, the most productive thing to do right now is get familiar with the Microsoft Graph API documentation — that's the foundation any Cowork-compatible custom agents will build on.
Is your organization actively evaluating AI agents, or is the timing still not right? I'd be curious to hear where teams are landing on this.
References
- Copilot Cowork: A new way of getting work done — Microsoft 365 Blog, March 9, 2026
- Powering Frontier Transformation with Copilot and agents — Microsoft 365 Blog, March 9, 2026
- Microsoft Launches Copilot Cowork, Powered by Anthropic's Claude — WinBuzzer, March 10, 2026
- Microsoft Copilot Cowork Marshals Corporate Intelligence, AI to Execute Complex Tasks — Cloud Wars, March 2026
Related posts:
- The Reality of Enterprise AI Agent Adoption - What actually blocks AI agents from reaching production
- MCP (Model Context Protocol): Connecting AI Agents in 2026 - The connectivity standard powering agent interoperability