Your AI Coworker now charges by the task

Copilot Cowork billing goes live—here’s how to get ahead of it.

If you run a Microsoft estate, per-seat licensing has a comfortable rhythm. You count the users, you true up once a year, and you can forecast next year with a straight face. AI doesn’t play by those rules. Copilot Cowork spends money on every task it touches, billed on usage rather than per seat. That turns a tidy entitlement question into a messy consumption one, and the old playbook doesn’t have a page for it. Now is the time to establish billing procedures and budgeting strategies for your agentic AI future.

Cowork billing is now mandatory

Cowork was released to general availability on June 16, 2026. It bills on usage in Copilot Credits in addition to the Microsoft 365 Copilot seats you already have. Organizations that implemented Cowork during the Frontier preview were given a short grace period to enable usage-based billing in the Microsoft 365 admin center, but must do so by June 30, 2026.¹ From July 1, any tenant without billing enabled loses access to Cowork entirely. Whatever workflows your teams have already built on it simply stop.

If you enable Cowork billing through an Azure subscription at scale, prepaid Cowork credits are MACC- and CSP-eligible, so the spend counts toward your existing Azure commitment instead of landing as a separate, untracked line. In other words, the deadline you have to meet anyway is also the moment to establish the right commercial footing, through CSP, rather than bolting on pay-as-you-go and sorting it out later.

This change is only the beginning

The same shift is rolling across the rest of the Microsoft AI stack. AI Builder credits are being retired too, with the credits seeded from existing licenses expiring on November 1, 2026, with no automatic conversion into the replacement model. Work IQ and Copilot Studio already meter in the same Copilot Credits currency. Taken together, the changes signify Microsoft moving its full AI portfolio from bundled entitlement to metered consumption. Cowork is simply the first deadline.

Two different commercial models, one estate

This split now runs throughout your Microsoft estate. One world is licensed: the M365 Copilot seat, Copilot for Security, the per-user entitlements you can count. The other is a consumption world layered on top; Cowork lives in both simultaneously. The seat gets your team in the door; actual Cowork tasks bill on usage on top of it. Around that sit the rest of the metered services: Azure AI, the agentic workloads in Agent 365 and Copilot Studio, and anything built on Microsoft Foundry. This two-layered billing structure creates serious difficulties in predicting spend and managing cost. You can know your license position down to the seat and still have no idea what you’re going to be billed for utilization.

On the consumption side, you buy the credit, and underneath it, the token. The bill moves with how often people use AI, how complicated AI tasks are, which model is used, and how much is automated. When everything was bundled, none of this showed up in your consumption bill. Now, a single Cowork task can fire off many calls before it returns an answer, adding up cost faster than anyone expects and usually without anyone watching. Managing a license position and managing credit consumption are genuinely different jobs; FinOps is the discipline that connects them.

Why the bill moves: how Cowork actually consumes

Using regular Copilot costs a small amount per request. Cowork doesn’t wait; you hand it a job and it runs end to end, working through steps, calling tools, and pulling in context, every piece of which burns credits. Microsoft prices each task based on four aspects: the model used, how much context it retrieves, how many tool calls it makes, and how long it runs. So a quick, single-step task on a light model is cheap, while a long, multi-output task on a top model is not: it’s entirely possible for one user to burn through a month’s budget in a day of heavy tasks. That variability is exactly why this is one of the fastest-growing line items in the enterprise, and why it can’t be managed like a flat per-seat fee.

Variability now impacts multiple Microsoft tools. Copilot for Security bills in Secure Compute Units that climb the deeper an investigation goes, and Microsoft Foundry hands your developers a much bigger box of consumption-based models and tools. To buy back some predictability, Provisioned Throughput Units (PTUs) let you reserve capacity up front for the workloads that run steadily, but don’t address the full problem for most organizations. The best approach combines committed capacity for the baseline and pay-as-you-go for usage that spikes.

The critical lever: benefit stack optimization

By now you’re managing several different layers of AI budget, and the order you draw them down in makes a difference. If you spend them in the right sequence, most of your usage is covered before you ever pay rack rate. A rough hierarchy:

• Spend included and prepaid credits first (Copilot Credits).
• Apply committed capacity (commit units and PTUs) to predictable demand.
• Use agent commit units for automation-heavy workloads.
• Let pay-as-you-go absorb genuine variability.

Correct sequencing avoids a cascade of problems down the line: prepaid credits expiring unused while the same month’s work gets billed on demand, on repeat. Sequencing can make the difference between optimization and overspend.

FinOps for AI

The same discipline that evolved to manage cloud and IT spend, when applied strategically, will help organizations stay on top of the fast-moving AI economy: FinOps. And if you understand FinOps, you know where to start. Before you sequence a single credit or commit to any capacity, you have to know what’s actually being used and where the money is going. That sounds obvious, but it’s the step most people skip, and you can’t manage spend you can’t see. The good news with Cowork is that the Microsoft 365 admin center shows consumption by user, group, and feature, and users can check the cost of a task as they run it. Understanding which AI services are switched on, who’s using them, which workloads run on their own, and what each one costs will allow you to get the full picture in front of finance before usage scales, not after.

From there, the discipline involves five factors:

• Token-level visibility: cost per interaction, per workflow, per outcome.
• Capacity and consumption planning: PTUs vs. on-demand vs. agent workloads, matched to demand.
• Benefit stack optimization: credits, commits, and consumption applied in the right order.
• Governance and controls: clear ownership across IT, finance, and the business, with policies, guardrails, and budgets.
• Continuous optimization: finding inefficiency and realigning spend to value, every billing cycle.

The window to act is now

June 30 is a hard deadline, and organizations need time to set up billing. Enabling Cowork billing appropriately through CSP against your Azure commitment allows you to set spending caps and policies before usage scales, plan for renewals with consumption already governed, and avoid surprise line items and uncapped pay-as-you-go invoices that derail your AI strategy and inhibit your innovation plans.

How SoftwareOne helps: FinOps meets Microsoft licensing

Thriving in the age of tokenomics demands a unique pairing of expertise: licensing and FinOps at the same table. SoftwareOne brings those practices together to support our customers as they scale innovation and control costs. In practice, that looks like:

• Getting Cowork billing right, fast. We can help you stand up usage-based billing before June 30 and route it through CSP against your Azure commitment so you clear the deadline and get MACC-eligible spend instead of an untracked PAYG line.
• Clarity on subscription licensing. We translate what the credit changes mean for your specific subscriptions: which seats unlock what, where bundled capacity is disappearing, and how Copilot, Cowork, and Power Platform now connect to paid consumption.
• Consumption impact analysis. We model how the shift to metered usage hits your spend, putting real numbers on your credit, PTU, and SCU exposure across Cowork, Copilot, Azure AI, and Foundry workloads before any of it lands.
• A clear roadmap. We turn that picture into a sequenced plan, including which workloads to commit, which to leave on demand, how to apply the benefit stack in order, and what FinOps governance to stand up, so the next deadline on the calendar, November 1, finds you ready instead of reacting.
• Hands-on optimization. If you’re already in the weeds on consumption, we’ll roll up our sleeves with you, finding the tasks and agent loops that cost the most, right-sizing models and capacity, and tying the spend back to the value it produces.

What comes after the first deadline

The immediate and November Cowork deadlines feel like a chore to clear, but they actually reflect an industry shift that will impact every organization eventually. The organizations that treat AI as a consumption-based cost center will be prepared for the changes that inevitably follow.

The clock is running—let’s talk this week

Cowork billing becomes mandatory on June 30. There are only days left to enable it the right way, through CSP, against your Azure commitment, rather than scrambling onto pay-as-you-go. SoftwareOne can map your entitlements, get billing set up before the cutoff, and build the FinOps governance to keep it under control.

¹ Source: Microsoft Learn, Partner Center announcements (June 2026): Frontier customers must set up usage-based billing for Cowork by June 30, 2026. Verify current dates and terms in the Microsoft 365 admin center before publishing.