Close your laptop. By morning, the competitor analysis is done, the invoice backlog flagged, and the weekly report structured. That's the actual promise behind Copilot Cowork, Microsoft's new agentic mode for Microsoft 365, which went globally available last week to the platform's 400 million commercial subscribers. You describe the outcome you want. Cowork builds a plan, executes it step by step, and returns a finished result. You don't have to be there while it runs.
For most office teams, this is a different kind of AI feature. Where standard Copilot responds to prompts, Cowork takes ownership of a complete task. The question that shifts for your organization: not whether to use AI, but how much of your workload you're ready to hand off entirely.
What does Copilot Cowork actually do?
Copilot Cowork turns your M365 assistant from a chatbot into a background contractor. You describe the end result in plain language, Cowork builds an execution plan, and carries out that plan step by step, without you at the keyboard.
Think of it as the difference between asking a colleague "can you look into this?" and asking them to "have the finished report on my desk by tomorrow morning." Standard Copilot does the first. Cowork does the second.
Microsoft shared three examples from the preview phase worth understanding. An engineering team automated the generation of dependency diagrams after every change to a batch spreadsheet. A product team compared nearly 4,000 files across two software versions, a task that would have taken weeks manually. A sales manager got a prioritized follow-up list from a stalled pipeline, with Cowork identifying exactly which deals had gone quiet and why.
Here's the thing: Cowork doesn't stop at Office apps. Third-party plugins already connect it to Miro, monday.com, Dynamics 365, and Microsoft's Fabric data platform. Adobe, Atlassian, Box, and Canva are on the roadmap. A task like "turn the key points from the Q2 board deck into a Miro board" becomes possible without switching apps manually.
Cowork can also browse the web via Edge, within your company's policy settings. If a task needs current competitor pricing or a recent news item for context, Cowork retrieves it without requiring you to supply the source.
What does it cost for a team?
Cowork layers on top of the Microsoft 365 Copilot User Subscription, priced at $30 per user per month. If your team already uses Copilot in Word, Excel, or Teams, that base license is already in place. Cowork adds usage-based charges on top.
Microsoft prices in Copilot Credits at $0.01 per credit. What a task costs depends on four variables: which AI model you select, how much context Cowork retrieves, the number of tool calls required, and how long the task runs.
| Task type | Credits | Cost (USD) |
|---|---|---|
| Light (limited sources, minimal reasoning) | 100-300 | $1-$3 |
| Medium (multiple sources, structured output) | 400-700 | $4-$7 |
| Heavy (broad aggregation, extensive output) | 700+ | $7+ |
Worth noting: that's cheaper than it sounds. A human analyst spending half a day combing through a stalled pipeline costs $150-250 in labor at average European knowledge worker rates. Cowork runs the same analysis for $7.
For a team of ten Copilot users running two light tasks per week each, expect roughly $80-240 per month in Cowork charges on top of the $300 base subscription. Total: $380-540 per month for a team of ten. Less than one day of external consulting.
Two payment options exist. Pay-as-you-go makes sense for testing. The P3 subscription ($200 per month for 25,000 credits) saves roughly 20% compared to pay-as-you-go rates. Microsoft's documentation covers all pricing tiers and includes a cost estimation tool.
One practical detail for your IT team: Cowork is off by default. Your tenant admin enables it per user, per group, or for the whole organization, and can set spending limits per team. No surprise invoices at the end of the month.
How does a Cowork task work in practice?
In the Microsoft 365 Copilot app, you switch from chat to Cowork mode with a toggle. You describe what you want to achieve in plain language, the same way you'd brief a colleague.
Cowork translates your request into a plan: which steps it will take and which sources it will access, your Outlook, calendar, Teams messages, SharePoint files, OneDrive documents. You review the plan, approve it, and Cowork starts. During execution, you get control points where you can redirect, add context, or pause. The output arrives as a finished document, spreadsheet, or analysis ready to share.
A concrete example: ask for a competitive analysis based on the last three months of customer feedback. Cowork searches your Outlook, pulls relevant attachments from SharePoint, reviews Teams conversations with the sales team, and delivers a presentation with key findings and a prioritized action list. Everything stays inside your existing M365 environment. No third-party platform required.
For context: Cowork can also browse externally via Edge, within your corporate policy settings. It retrieves competitor pricing or a news item if the task calls for it, pulling the context you'd otherwise have to find yourself.
Which AI model does Cowork use?
You choose. The default is a Microsoft model, but you can switch to Claude Opus 4.8 or Claude Sonnet 4.6 from Anthropic, or GPT-5.5 from OpenAI.
Model choice affects both quality and cost. Claude Opus 4.8 handles the deepest analytical tasks but consumes more credits. Sonnet 4.6 is lighter and less expensive. Microsoft's own Cowork 1 model is on the roadmap and promises significantly lower costs for standard tasks.
The fact that you can select an Anthropic model inside a Microsoft product confirms a broader platform shift. Enterprise tools are becoming model-agnostic. Organizations want the best model for each task, not just the one their primary vendor offers. If you've already worked through integrating Claude in Microsoft 365, Cowork extends that to full autonomous task execution.
Microsoft reports that across 125 internal test runs, Copilot Cowork is 30-40% cheaper than running equivalent tasks through Claude Cowork using Opus 4.8. The gap is structural: Microsoft optimizes for its own infrastructure. Independent verification is limited, but the pricing logic is consistent.
Privacy and your organization's data
Cowork runs entirely inside your Microsoft 365 security boundary. Your data doesn't leave it. Sensitivity labels on existing documents carry through into all Cowork output: a SharePoint file marked "confidential" stays "confidential" in whatever Cowork produces from it.
Audit logs, eDiscovery, and Insider Risk Management function identically to standard Copilot usage. Data Lifecycle Management became available on June 22. Data Loss Prevention support is coming soon.
For organizations operating under GDPR, this matters. Cowork processes your data in the same Microsoft datacenters already running your M365 environment. If you have a data processing agreement for M365 Copilot, it already covers Cowork. All prompts, responses, and generated files fall under your existing Microsoft 365 policy.
Our Generative AI Statistics data shows that 44% of European SMBs processed business data with a free AI tool in the past year without knowing where that data went. Cowork addresses that risk structurally: your data stays in your environment, under your control, visible in your compliance logs.
For organizations covered by the EU AI Act's general-purpose AI transparency requirements, which entered full effect in August 2025, Cowork's integration with existing M365 compliance infrastructure also simplifies the documentation obligations for AI system use in regulated industries.
Copilot Cowork vs. the alternatives
| Feature | Copilot Cowork | Claude Cowork | ChatGPT |
|---|---|---|---|
| Full task delegation | Yes | Yes | No (chat only) |
| Runs in background | Yes | Yes | No |
| Native M365 integration | Yes, native | Via M365 Copilot | No |
| Accesses company data | Outlook, SharePoint, Teams | Limited | No |
| Model choice | Claude, GPT-5.5, own model | Claude only | GPT only |
| Cost per task | $1-$7+ | Comparable | N/A |
The key advantage for Copilot Cowork is organizational context. It knows your email, your calendar, your shared files. Claude Cowork handles deeper reasoning tasks, but it doesn't have that business context unless you supply it manually each time. ChatGPT remains a chat interface: it doesn't run tasks in the background and doesn't connect to your organization's data.
For most teams already running M365, Copilot Cowork is the lowest-friction starting point. The data is already in the ecosystem. These tools aren't mutually exclusive, either: a team using Claude for complex analytical work can run Copilot Cowork for M365-native tasks that require company context. They're complementary.
Three things to try this week
Getting started doesn't require a project plan. Three steps work.
- Check your license. Does your organization have an M365 Copilot subscription ($30 per user per month)? No license, no Cowork access. That's the only hard prerequisite.
- Ask your IT admin to enable Cowork for a small test group. It's off by default. Start with two or three people, set a monthly spending limit of around $100, and track what tasks they run first.
- Begin with a light task. Have Cowork generate a weekly summary of all client communications, or compare two versions of a proposal. Learn how it structures its output before trusting it with heavier work.
Copilot Cowork isn't the first product to promise full AI task delegation. But it's the first built natively into the productivity environment where most office teams spend their working day. The integration advantage is real, and the pricing makes it accessible for organizations that would never budget for a separate enterprise AI platform.
The underlying shift: AI is moving from "help me write something" to "complete this work for me." Copilot Cowork is the clearest version of that shift available inside mainstream enterprise software today.
