What Microsoft Copilot Cowork actually is, what it costs, and how it compares to other AI agent tools on the market.
Tecnet co-hosted another session with Microsoft, this time on Microsoft Copilot Cowork, the newest addition to the Copilot family. Ashley Evans, the same Microsoft sales leader who joined our finance, HR, and operations session, walked through why AI is moving past simple assistance, what Cowork actually does, and how the pricing and security model works underneath it.
A quick poll at the start of the session found that 26 percent of attendees were already using Cowork, and 68 percent had heard of it but hadn’t tried it yet. If that describes your organization too, this post covers what’s worth knowing before you do.
The problem: chat can answer a question, but real work still spans a dozen apps
For the last couple of years, most people’s daily AI use has looked the same: open a chat window, ask for a summary, a first draft, or an answer to a question, and move on. That’s been genuinely useful. It’s also, for most teams, still limited to one turn at a time.
The work that actually eats a day rarely lives in one place. Preparing for a customer meeting means pulling from email threads, a proposal deck, calendar notes, and whatever was discussed last time. Cleaning up an inbox means moving between the inbox itself, a calendar, and a handful of related documents. Each of those handoffs (opening one app, copying something into another, coming back to check a third) is what people in this space now call context switching. AI chat tools were never built to do that part. They wait for the next question.
The insight: the shift is from giving advice to getting things done
Microsoft frames the change happening right now as three shifts, and together they explain why tools like Cowork are getting so much attention.
The first is a cognitive shift: from treating AI like a single specialist doing one job, to treating it like a manager who takes on a whole project and hands off pieces of it. The second is an orchestration shift: you describe the outcome you want, and the tool coordinates the work across applications on its own, so the juggling between five different apps becomes its job. The third is an action shift: AI moves from offering advice you then act on, to completing the task directly, with you approving the result at the end.
Copilot Cowork is Microsoft’s answer to all three. Copilot Chat answers a question, and Copilot inside an app helps with a single document. Cowork can look at an outcome you describe, plan the steps needed to get there, work across multiple files and applications, and keep you updated as it goes.
The framework: match the tool to the size of the job, and know what’s protecting your data
Once Cowork is in view, it helps to place it next to the tools most teams already use, since the real difference is how much work you hand over and how long the tool runs before it checks back in with you.
Copilot Chat is built for fast, single-session help: a draft, a summary, an answer to a question, done in seconds. Copilot inside an application, like Word or Excel, works with the context of the specific file you have open. Copilot Cowork plans across multiple applications and files, works through the sequence on its own, and brings back a finished result for you to review. Microsoft also has a fourth tier now available through its Frontier program, called Microsoft Scout, an always-on agent that keeps moving work forward, including tasks on your own PC, without being asked first.
None of that matters much if the underlying platform can’t be trusted with real business data, so it’s worth being specific about what Cowork does with your information. Four commitments sit underneath it. Your data is encrypted both at rest and in transit, and it stays inside your organization’s M365 security boundary, so nothing you work on is used to train or improve the underlying models. Cowork inherits the permissions and sensitivity labels already set up in your tenant, so it can only see what the person using it could already see themselves. It’s manageable centrally through the M365 admin center, giving administrators visibility into who’s using it, how, and to what extent. And it runs inside a protected sandbox where every action needs your approval and can be undone.
For an organization already running Microsoft 365, that’s a meaningfully different starting point than adopting a brand new AI platform. The governance, the permissions, and the audit trail are the ones you’ve already built.
The solution: budget for usage-based cost, start with one task, and know what you’re comparing it to
Knowing what Cowork can do is only half of the planning work. The other half is understanding what it costs and how to keep that predictable, since this is where Cowork departs from the flat, per-seat pricing most IT budgets are built around.
Copilot licensing is still the entry point, and it’s genuinely useful on its own for the everyday work most teams already do: summarizing, drafting, analyzing, and pulling from your own tenant’s information. Cowork sits on top of that as a usage-based service, billed in what Microsoft calls Copilot credits. As a rough guide from Microsoft’s own estimates, a simple, single-step Cowork task runs somewhere around $1.25, while a complex task spanning multiple documents and steps can run up toward $12. Those figures are illustrative, and worth checking against Microsoft’s current cost estimator before you budget against them. One practical tip: typing “/cost” inside a Cowork session shows exactly how many credits that particular task consumed.
Because the spend is usage-based, the M365 admin center includes a cost management dashboard built specifically to keep it visible and controlled. It covers three things: monitoring, so you can track how quickly credits are being used and by whom; configuration controls, so you can set spending limits and alerts at a user or group level; and purchase controls, so an administrator manages any additional credit requests centrally. For an IT leader used to predictable, per-seat software costs, that dashboard is the tool that keeps a usage-based model from turning into a budget surprise.
Three examples from the session show what a Cowork task actually looks like once it’s running. The first was customer meeting preparation: given a simple prompt, Cowork pulled from existing tenant data, reasoned across several documents, and produced multiple finished artifacts, a briefing note and supporting materials, in a few minutes. The second was inbox and calendar triage: Cowork surfaced draft email replies directly in the chat pane, cutting out the back and forth between inbox, calendar, and document tabs. The third was financial research: Cowork worked across several file formats, kept a visible log of what it was doing at each stage, and built out a full analysis.
One detail worth building into your own rollout: if a task turns out to be one your team will repeat, Cowork lets you save the instructions behind it as a reusable skill, so the next person doesn’t need to describe the job from scratch.
It’s also worth knowing how Cowork compares to other agent tools on the market, since the comparison question comes up in almost every conversation about it. Cowork runs on Anthropic’s Claude Cowork models, brought into the Microsoft 365 ecosystem through a partnership between the two companies. What matters most for a business evaluating it is everything built around that model: where it runs, what context it has, and how it’s governed. A standalone desktop agent typically reads files directly off your local hard drive and runs outside any central governance. Cowork runs inside your governed M365 tenant, grounded in what Microsoft calls Work IQ, the organizational context it already has because it’s built natively into the platform you use every day. It carries the same enterprise security, permissions, and audit controls as the rest of your M365 environment, and it’s multi-model, meaning it can select the most efficient model for a given task, which matters once you’re paying by consumption.
For most organizations already invested in Microsoft 365, that combination (native context, existing governance, and cost controls built for a usage-based model) is the practical case for starting with Cowork over a separate standalone tool.
A reasonable way to begin is small and specific. Pick one recurring, multi-step task, something that currently means opening four or five things to get done, and try it in Cowork a handful of times before deciding whether to expand access further. Microsoft’s cost estimator and its guide to managing Copilot credits in the admin center are both worth reading first, so you know roughly what to expect before you start.
Curious whether Copilot Cowork is worth adding to your organization?
Tecnet helps organizations evaluate Microsoft Copilot, build AI agents, and plan a rollout with cost and governance considered from the start.
Visit tecnet.ca or reach out at [email protected].
For more on AI agents and how Tecnet evaluates them for clients, see our companion posts on AI agents for business teams and AI agents in finance, HR, and operations.