Copilot Cowork Is Generally Available

Microsoft's Copilot Cowork reached general availability last week, and the most striking thing about the launch isn't the marketing—it's the pace of real-world adoption. After just three months in the Frontier preview, more than half of the Fortune 500 is already using Cowork. That's not a slide for an earnings call; it's a signal that agentic work, the kind where you delegate a multi-step task to an AI and come back to a finished result, is moving from experiment to enterprise standard.

Cowork is built on a straightforward premise: instead of drafting an email, summarizing a meeting, or answering a question, you hand off a job. "Compare the Q2 forecast across the four regional spreadsheets and flag the accounts where revenue recognition changed." "Find every contract that references our old SLA and draft a side-by-side with the new one." "Look at this stalled pipeline and tell me which deals need attention this week, in order of impact." Cowork plans the steps, calls the tools, and returns the deliverable.

That shift—from suggestion to delegation—is the heart of what Microsoft calls agentic work. And it is arriving faster than most IT roadmaps assumed.

From Chat to Completed Work

The previous generation of Copilot was powerful, but its natural habitat was the chat pane. You asked, it answered. You could iterate, refine, and copy the output into Word or Excel. Cowork changes the shape of the interaction. You define the outcome, set any guardrails, and let it run. The task might take minutes or hours. It can keep running after you close your laptop because it is hosted in the cloud, inside the same Microsoft 365 trust boundary that already governs your email and files.

In the GA announcement, Microsoft shared a few customer examples that illustrate the difference. One engineering team taught Cowork to safely edit batch-job spreadsheets and regenerate dependency flow charts after every change. Another team compared nearly four thousand files across two product versions, work that would have taken weeks. A sales lead pointed Cowork at a stalled pipeline and got back a ranked list of at-risk opportunities, with the exact follow-up that had gone cold on each. What used to be a week of manual review became a single morning.

These are not toy demos. They are the kind of operational work that sits in every mid-sized and large organization: cross-referencing data, reconciling versions, triaging queues, and producing structured output that a human can review and act on.

Why Microsoft Chose a Different Architecture

Cowork is not the only agentic assistant on the market, but Microsoft's design choices are worth understanding because they explain why enterprise buyers are comfortable adopting it at scale.

First, cloud hosting matters more than it sounds. When an agent is going to act on your behalf—write files, send messages, update records—you want it running in a controlled environment, not on a laptop with a local browser extension. Cowork runs inside Microsoft's cloud, which means tasks continue even if the user's device is off, and files are not stored locally where they are harder to govern.

Second, Work IQ gives the agent context without extra plumbing. Because Cowork is part of Microsoft 365, it can ground tasks in the organization's actual graph: who reports to whom, which documents are authoritative, what meetings decided what, and which SharePoint sites hold the source of truth. That context layer is hard to replicate with a generic assistant plugged into email via an API.

Third, security and compliance are inherited. Cowork prompts, responses, and generated artifacts flow through the same controls as the rest of Microsoft 365. Sensitivity labels are preserved end-to-end. Audit logs, eDiscovery, Data Security Posture Management, Insider Risk Management, and Communication Compliance policies all apply. Data Lifecycle Management for Cowork artifacts reached GA on June 22. DLP is coming soon. This is the kind of incremental, enterprise-grade release pattern that makes compliance teams comfortable.

Fourth, model choice keeps options open. At GA, Cowork runs on Anthropic models, including Opus 4.8 and Sonnet 4.6. Frontier customers can also use GPT 5.5. And Microsoft's own Cowork 1 model, a fine-tuned option optimized for everyday tasks at lower cost, is releasing in the coming weeks. The message is clear: customers are not locked into a single model, and Microsoft will route the right model to the right task.

Fifth, cost controls are built in from day one. Cowork is off by default. Admins enable it, assign access, and set spending limits at tenant, group, and user levels. Usage reporting breaks down spend by user and feature. Users can request additional credits when a task needs them. There are two payment options: pay-as-you-go at $0.01 per Copilot Credit, or P3 committed usage for a discount.

The Pricing Reality

Usage-based pricing is new for most Microsoft 365 customers, and the announcement spends significant time on how to budget for it. Microsoft observed three common task patterns during Frontier: light, medium, and heavy. Light tasks use a small number of knowledge sources, limited reasoning, and produce one or fewer outputs. Medium tasks draw on multiple sources, apply structured reasoning, and generate two or more outputs. Heavy tasks aggregate broadly, apply deep reasoning, and produce many outputs.

Microsoft also identified four user personas with distinct usage patterns. The idea is that IT can model cost by multiplying users in each persona by their expected mix of light, medium, and heavy tasks, then applying a per-prompt rate. A downloadable estimator spreadsheet is available for organizations that want to build their own model.

Frontier tenants that had at least one active Cowork user between March 30 and June 16 get a grace period: they will not be billed until July 1, 2026. That gives IT teams a few more weeks to turn on controls before usage starts hitting invoices.

What Else Ships at GA

Beyond the core task engine, the launch adds several practical pieces.

The Microsoft 365 Copilot app now includes a toggle that moves users from chat into Cowork's full experience. Plugins extend Cowork into partner and line-of-business systems. Nine are available now, including Enosix, Harvey, LSEG, Miro, monday.com, Moody's, Morningstar, S&P Global Energy, and TeamsMaestro. Eight more are coming soon, including Adobe, Atlassian, Box, Canva, CB Insights, Databricks, MoneyForward, and Templafy. Fabric and Dynamics 365 Sales, Customer Service, and ERP apps are also GA.

In Frontier, Cowork can browse the web through a local Edge browser, following existing enterprise policies. That opens up research tasks that require public data without bypassing security controls.

Why This Matters for OpenClaw Users

If you are running OpenClaw on Windows or inside a Microsoft 365 environment, Cowork is a useful complement, not a replacement. OpenClaw gives you local, transparent control over your own agents, skills, and model routing. Cowork gives you a managed, enterprise-grade agent that can operate across the Microsoft graph and keep running in the cloud.

The two can coexist cleanly. You might use OpenClaw for local coding agents, Obsidian-based knowledge workflows, or Windows automation that you want to keep on-device. You might use Cowork for cross-tenant research, complex document comparison, or long-running tasks that need to interact with SharePoint, Teams, and Dynamics. The boundary is not either-or; it is local control versus delegated scale.

What is changing is the expectation. Once employees see a task like "compare these four thousand files and tell me what changed" completed overnight, the bar for every other process rises. The organizations that adapt fastest will be the ones that treat agents as a new layer of infrastructure, with clear ownership, guardrails, and measurement.

A Practical Starting Point

For IT leaders who want to get ahead of this without opening the spending floodgates, a reasonable first step is to identify three to five repeatable, multi-step tasks that currently consume senior staff time. Document the inputs, the expected output, and the approval step. Then enable Cowork for a small pilot group, set per-user credit caps, and run the tasks side by side with the manual process for two weeks.

The goal is not to eliminate humans from judgment. It is to remove the mechanical overhead that keeps people from exercising judgment. Cowork is good at the former. The latter remains the job of the team.

Microsoft's bet is that agentic work will become as normal as sending a Teams message or running a Power BI report. The GA of Copilot Cowork is the most concrete step yet toward that future. For organizations already invested in Microsoft 365, the infrastructure, trust model, and cost controls are in place. The question is no longer whether agentic work will arrive. It is which teams will learn to delegate first.