Copilot Cowork and the shift to usage-based AI: what FY27 means for mission-driven organizations

On June 16, Microsoft made Copilot Cowork generally available. If you have watched Copilot mature over the past two years, the capability itself is the logical next step: instead of answering a question or drafting a paragraph, Cowork takes an outcome you describe and does the multi-step work to deliver it. You ask it to clean up your inbox and calendar, or to pull email, calendar, and CRM context into a meeting-prep briefing, and it plans the steps, works across your apps, and carries the task forward over minutes to hours while you stay in control with checkpoints. It runs on Work IQ, the layer that grounds it in your real business context, and it stays inside the Microsoft 365 security and compliance boundary you already have.

That is a genuine shift in what Copilot does. But it is not the part of the announcement I would put in front of a board first. The part that changes how you plan is quieter, and it is in the fine print: Cowork is billed by usage.

The real headline: seats, plus usage

For years, Microsoft AI has been a per-seat cost. You buy a Microsoft 365 Copilot license for a user, and that user gets Copilot. Predictable, budgetable, done. Cowork keeps that license as a prerequisite, but the work it performs is metered separately through a new currency called Copilot Credits. This is Microsoft following the whole industry from a fixed subscription toward a subscription plus consumption model. The license stays the floor. The heavy, long-running, agentic work is billed on top, in proportion to the work actually performed.

The mechanics are straightforward once you see them. Copilot Credits are pooled at the tenant level and cost one cent each on a pay-as-you-go basis, or less if you prepay for an annual pack at a volume discount that runs from about 5 percent up to 20 percent at the largest tiers. The same credits also cover other agentic work, like custom agents built in Copilot Studio or calls to the new Work IQ APIs. Microsoft publishes rough planning ranges: a light task like a recurring weekly status draft runs somewhere in the range of 70 to 200 credits, a medium task like a meeting-prep briefing runs a few hundred, and a heavy task like analyzing six months of usage data into a leadership-ready report can run well past 1,500 credits. In dollars, that is a few cents to a few dollars for routine work, and around fifteen dollars or more for the heavy lifts.

None of those numbers is alarming on its own. The thing to understand is that they add up in a way a fixed seat cost never did, and they add up faster the more useful the tool becomes.

Why this lands differently in the mission-driven sector

Microsoft writes for the median enterprise customer, and the FY27 materials frame usage-based billing as an opportunity to scale AI spend as value appears. That framing is correct. It is also written for organizations that can absorb a variable AI bill without a board conversation. A 200-person foundation or a critical-access hospital is not that organization.

Two things make the mission-driven sector different here. The first is budget: when every dollar is accounted for against a program or a grant, an AI line item that moves month to month is not a rounding error, it is a governance question. The second is scrutiny: the board, the finance committee, the auditor, and the funder all expect technology spend to be predictable and attributable. Usage-based AI can meet that bar. It can even come out cheaper than blanket seat licensing, because you pay for the work that gets done instead of for capability that sits idle. But it only meets that bar if someone sets the guardrails before the usage starts, not after the first surprising invoice.

This is the same pattern we have written about with shadow AI: the capability arrives faster than the controls, and finance finds out last. Cowork does not create that risk out of nothing. It just moves it from the browser tab to the invoice.

The switch is easy. The meter is the work.

Turning Cowork on is a toggle. Governing it is the job. Microsoft has done the responsible thing here and shipped the controls alongside the capability: a Cost Management dashboard in the Microsoft 365 admin center where an administrator sets a default spending policy, then scoped policies by group or by user, each with a monthly budget, a hard cap, alerts as spend approaches the limit, and reporting that attributes credits to specific users, groups, and agents. Prepaid credits are drawn down before pay-as-you-go, so the discount lands first. The tools are good. They still have to be configured correctly, run on a cadence, and translated into a decision the budget owner can actually make.

Here is what we tell clients to put in place before the first user runs their first Cowork task:

• Set the budgets and caps first. A default policy for the tenant, then tighter policies for the teams and individuals most likely to be heavy users. Hard monthly caps prevent a single unbounded task or an over-eager power user from spending a quarter's allocation in a week.
• Size a credit plan against real usage. Estimate how many people will use Cowork, how often, and for what kind of task, then model the prepaid-versus-pay-as-you-go mix. Prepay for the predictable baseline to capture the discount; leave pay-as-you-go headroom for the spikes.
• Pick the use cases worth delegating. The value of Cowork shows up only when staff hand it work that is genuinely worth a few dollars of compute. Name those workflows deliberately, the recurring report, the meeting prep, the inbox triage, and start there, rather than letting everyone experiment on the meter.
• Report monthly, in plain language. Where did the credits go, by team and by task, and did the spend earn its keep? That is the report that keeps AI a line item the board can interrogate instead of a surprise it has to explain.

What we are telling clients today

Cowork is real, it is available now, and it is worth adopting. It is also the first place most mission-driven organizations will meet usage-based AI billing, which makes it the right moment to build the cost discipline that every agent platform from here will require. The license is the floor. The credits are the meter. Govern the meter, and usage-based AI becomes what it is supposed to be: paying for outcomes, with the outcomes and the spend both visible enough to steer.

We have built a dedicated engagement around exactly this: standing Cowork up in a governed configuration and running the cost controls, either as a fixed-fee setup or as an ongoing managed service. If you have Copilot licenses and you are looking at Cowork, that is the place to start. And if you are not yet sure where Cowork fits in your sequence, a Discovery Sprint will place it against everything else on your roadmap.

The move from seats to seats-plus-usage is not a Microsoft quirk. It is where the whole market is going. The organizations that handle it well will be the ones that treated the meter as a governance problem from day one, not the ones that found out what it cost at the end of the month.