Copilot Cowork, your questions answered and how to build the business case

Microsoft made Copilot Cowork generally available on 16 June 2026, and the reaction has been exactly what you’d expect. On one side, genuine enthusiasm – more than half of the Fortune 500 were already using it, and Microsoft says it’s the fastest-growing, highest-satisfaction feature in the history of its Frontier program. On the other side, a reality check. The three-month window when Cowork was free in Frontier is over, consumption-based billing has begun, and the question that was always coming has arrived: who is going to pay for it?

That question isn’t a problem. It just needs to be understood. Cowork is not another feature bundled into your license – it’s agentic work that consumes compute based on what you ask it to do. Used well, it compresses a week of manual review into a morning. Used carelessly, it turns a task you could have done for free into a surprise invoice. This is the moment the whole market is moving through from experimenting with Ai to running it at scale – and Cowork is where that shift starts to show up on the invoice.

So before you switch it on across the business, it’s worth getting clear on what Cowork actually is, where it earns its keep, what you already own inside Microsoft 365, and how to put a credible business case in front of your CFO. Here are the questions we’re hearing most.

Q: What exactly is Copilot Cowork – isn’t it just Copilot with a new name?

No. The difference is the whole point. Copilot Chat advises and responds. Copilot Cowork plans and acts. Chat helps you explain, summarize or draft something while you stay in the driving seat. Cowork is designed to do the work – it plans the steps, executes them across your applications, and hands back a finished result rather than a draft.

The distinction is clean: Copilot Chat helps you generate content and find answers inside a single back-and-forth, while Copilot Cowork runs long, multi-step tasks in the background across Microsoft 365 to reach a finished outcome. It’s the shift from an interactive chatbot to a semi-autonomous agent – and that shift, from “you working with AI” to “AI working on your behalf,” is exactly the moment the meter starts running.

Q: Is Cowork only for big enterprises?

It’s priced and positioned for the enterprise, but we’d argue the value isn’t enterprise-only.

Look at how the competition is built. Anthropic’s Claude Cowork, for example, is an on-device assistant that typically requires a Claude Business or Enterprise license plus consumption, with its activity sitting outside enterprise audit logs and compliance APIs and its history stored locally on the device. That’s a model that assumes a large enterprise with its own security team to harden it.

Copilot Cowork takes the opposite approach. It’s built into Microsoft 365 and grounded in your existing Work IQ context, so any organization already on a Microsoft 365 Copilot license can adopt it – no separate enterprise platform, no new trust boundary to stand up, no security team required to make it safe to use. The discipline that makes Cowork pay off, theright tool for the right task, sensible credit limits, clear governance – scales down just as well as it scales up. A 200-person company that picks three high-value scenarios and caps its credits will see ROI as readily as a multinational. So while Microsoft aimed this at the enterprise, the case for “Cowork for organizations of all sizes” is real.

Q: Do we need to integrate it with all our other apps first?

This is one of Cowork’s quiet advantages: no, because it’s already built in.

Cowork is grounded natively in Work IQ – your email, calendar, documents, Teams, meetings and search, and operates inside the Microsoft 365 trust boundary. You’re not commissioning an integration project to connect it to where your work already lives. By design it’s confined to data held in OneDrive and SharePoint Online, and that cloud-native model is a genuine security advantage. Compare it with on-device rivals such as Anthropic’s Claude Cowork or the open-source OpenClaw, which install on the endpoint and leave IT to bring its own controls to harden them.

In short: the data, the permissions and the controls Cowork relies on are the ones you’ve already configured in Microsoft 365. That’s a faster path to value and a smaller attack surface.

Q: If it’s that capable, why not use it for everything?

Because capability isn’t the same as cost-efficiency. When you need to fix a wall plug, you reach for a drill, not a jackhammer. A jackhammer is more powerful – and for that job it’s unnecessary, expensive and likely to do more harm than good.

Cowork is the jackhammer.

A single task carries out many actions and calls multiple tools, which burns credits quickly. Microsoft’s own job patterns make the economics concrete (at the $0.01 pay-as-you-go rate):

Job type	What it looks like	Est. credits	Cost per job
Light	A weekly email summary; few sources, light reasoning, one output	~125	~$1.25
Medium	A customer-meeting briefing pulling from CRM, files and Teams	~500	~$5.00
Heavy	Six months of customer-data trend analysis with dashboards	~2,500	~$25.00

Now multiply by reality.

Microsoft’s own Frontier usage data puts a typical knowledge worker at 22 light, 11 medium and 5 heavy jobs a month – about $207.50, or roughly seven times the $30 Copilot licence that’s a prerequisite. Multiply that across a few thousand workers and you have an unpredictable, potentially budget-breaking situation.

The trap is using Cowork for work your licence already covers for free. Spending credits on a low-value task that Copilot Chat or Power Automate would have handled at no extra cost is exactly how bills balloon. The rule is simple: use Cowork sparingly, for the right tasks.

So here’s the map every user should have:

Task or need	Reach for	Credit cost
Quick answer, summary or text draft	Copilot Chat (incl. Opus 4.8 / GPT-5.5 model picker)	Included in license
Deeper research across sources, with citations	Researcher (up to 25 queries/mo.)	Included in license
Data analysis, trends, visualizations	Analyst (up to 25 queries/mo.)	Included in license
Create or edit a doc, deck or spreadsheet	Agent Mode in Word, Excel, PowerPoint	Included in license
Tasks, plans and work status	Planner Agent	Included in license
A recurring, rules-based process	Custom agent in Copilot Studio	Depends on agent
A multi-step task across M365 with complex orchestration	Copilot Cowork	Metered in credits

The headline most organizations miss: the $30 licence already does far more than they realise. Cowork is for the bottom row – the genuinely ambiguous, judgement-heavy, multi-step work a simpler tool can’t finish.

Q: What about security, is delegating work to an agent safe?

Security is the strongest argument for the Microsoft approach, and getting it right is where a partner earns their keep. At SofwareOne, security sits at the core of everything we do in the workplace, so we treat agentic AI the same way: not as a risk to be feared, but as a capability to be enabled safely.

The good news is architectural. Cowork runs entirely inside the Microsoft 365 trust boundary, inheriting your audit logs, DSPM, eDiscovery, Insider Risk Management and Data Lifecycle Management controls. It’s cloud-native (no local file access), it can’t delete your files by design, and it enforces human-in-the-loop checkpoints – a “zero actions” posture that requires explicit human approval before any sensitive external action. A sensible safeguard is to separate the “research and draft” phase of a task from the “execute and send” phase.

The part to plan for: this is a genuinely new class of risk. As the Gartner^® security note The Future of AI Security Is in Securing Agent Actions, Not Prompts puts it, with agentic AI: “the primary risk is not what the AI says, but what the AI does.”¹ The perimeter shifts from the words going into the actions the agent takes.

The specific concern is action injection, where malicious instructions hidden in an untrusted email or document hijack an agent that holds real credentials – a risk profile closer to malware behavior (lateral movement, privilege escalation) than to a chatbot mistake. Two gaps to note at GA: Copilot DLP support for Cowork is still in preparation, and for UK/EU tenants the underlying Anthropic models sit outside the EU Data Boundary, so Cowork is disabled by default and needs deliberate admin opt-in. None of this is a reason to avoid Cowork – it’s a reason to enable it consciously, with governance in place.

How to build the business case

Pull it together, and the case writes itself around four moves:

1. Anchor on workforce economics, not licenses. The right number isn’t “users × average persona.” Microsoft’s own modeling puts a typical knowledge worker near $200/month. Building the case on realistic role distribution – and naming the value each scenario delivers – is what separates a fundable plan from a frozen rollout.
2. Lead with what you already own. Before any Cowork spend, show that Chat, Researcher, Analyst, Agent Mode and Planner already cover the everyday work – included in the license. This both protects the budget and proves AI maturity. It's also where an experienced partner shortens the path – SoftwareOne has supported 1.8M+ Copilot users and delivered over 2,000 Copilot engagements, so we know which scenarios pay back and which quietly burn credits. (It also answers a common refrain among IT leaders: that the free Copilot Chat alone already meets many employee needs – a perception that quietly undermines premium licenses when advanced features go unused.)
3. Quantify the few scenarios where Cowork wins big. A stagnant-pipeline review compressed from a week to a morning; a 4,000-file version comparison done in a day. Put hours-saved against credit cost. That ratio is the business case.
4. Show the guardrails up front. Cowork is off by default. Set tenant, group and user credit limits, configurable alerts, and a pilot-first rollout. A CFO funds a metered tool far more readily when the spend is visibly capped.

Cowork is a real step change in how we’ll work inside Microsoft 365. It’s also not a silver bullet. The organisations that win are the ones that understand the licence they already have, choose the right scenarios, set limits, teach users what to use when, and measure value. Do that, and you get the upside without the surprise invoice.

Where to start

If E7 is on your renewal horizon, our Microsoft 365 E7 Envisioning engagement gives your CIO, CFO and Risk leaders a clear, defensible decision – a licensing roadmap, prioritized use cases and a governance-readiness view – before you commit. And if switching Cowork on now, our digital workplace and FinOps teams help you set the credit limits, scenarios and guardrails that keep the spend measured. Talk to SoftwareOne before you scale.

Sources & notes
¹ Gartner^® Emerging Tech: The Future of AI Security Is in Securing Agent Actions, Not Prompts, 20 February 2026 by Mark Wah, David Senf.

Disclaimer: GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved.

Cost, credit and usage figures are drawn from Microsoft’s Customer Cowork Estimator and published Frontier usage data.