Base Copilot use sat under 45 percent at six months, and the agentic demo argued for more seats

Cowork is the most compelling Copilot demonstration Microsoft has built, and it has no price of its own. It rides on a Copilot licence and the Frontier program, which means the thing being sold in the room is the commitment underneath it.

Prepared by Redress Compliance · August 15, 2026 · Microsoft advisory. Based on 30 to 40 Microsoft 365 Copilot engagements, 2025 to 2026.

• Delegated tasks: it drafts documents, sends mail, schedules meetings and posts in Teams on your behalf, rather than answering a single prompt.
• Reusable skills capture how you want a workflow done, which raises the value ceiling and concentrates dependence inside daily workflows.
• Cross application reach extends into analytics and business application scenarios, widening the governance surface at the same time.
• Model choice is real, with Anthropic Opus selectable alongside the Microsoft models, so confirm which model processes your data.
• Governance runs through Agent 365, so controls and audit belong upstream of enablement rather than after it.
• Switching cost rises with adoption, because skills embedded in daily work are harder to unwind than a chat assistant, per the Copilot licensing brief.

The account team pitch is that agentic Cowork justifies expanding your Copilot estate now, because the productivity ceiling has moved up. In roughly 6 of 10 Copilot rollouts we benchmarked, base seat adoption was still under half at six months, so adding agentic capability widened the gap between licence spend and realized value rather than closing it.

The structure of the offer is what makes it persuasive. Cowork genuinely is impressive, it genuinely is included, and it genuinely requires nothing new to be purchased. Each of those statements is true and together they produce a conclusion that does not follow: that because the capability costs nothing extra, expanding to reach more of it is free. It is not. The capability has no price precisely because the licence does, and the only lever anyone can pull in response to a Cowork demonstration is seat count or term.

That inversion is worth naming because it changes what diligence looks like. There is no Cowork contract to review, no Cowork rate to benchmark, and no Cowork clause to negotiate. What there is instead is a Copilot commitment that may grow on the strength of a capability nobody has yet measured in your own tenant. The honest test is therefore the same one that applied before Cowork existed: what share of assigned seats show measured active use, and is that number moving. Agent audit logs, not demonstrations, are what answer it, and Cowork's own footprint inside the applications your staff use makes those logs available.

The governance point compounds it. Agent governance was an afterthought in about half the rollouts we saw, which is unsurprising when the capability arrives as an included feature rather than a procurement event. Nothing forces a review, because nothing is being bought. Agents that draft, send and post inside your tenant deserve controls and audit before enablement, not after the first incident. Tie any Cowork driven expansion to measured base adoption, keep new commitments short until that number clears 60 percent, and settle governance upstream. The seat economics sit in the Copilot licensing brief, the cost model in the true cost analysis, and the renewal mechanics in the EA renewal brief.

Watch the briefing · 7:34Microsoft Copilot and CoworkHow the AI seats are priced against the estate you already own, what the bundling does to your renewal baseline, and which commitments are worth making before adoption is proven.

Across roughly 30 to 40 Microsoft 365 Copilot engagements in 2025 and 2026, the agentic features became the upsell hook well before base value was proven:

The patterns: demonstrations treated as evidence, expansion decided before adoption was measured, and governance scheduled after enablement.

The buyer side move is to price the commitment, since that is the only thing with a price. The wider library sits in the Microsoft practice.

Is Copilot Cowork a separate SKU?

No. Cowork rides on an active Microsoft 365 Copilot licence and is delivered through the Frontier early access program, which requires a tenant opt in. There is no standalone Cowork price to negotiate, so the commercial decision sits inside your Copilot contract rather than beside it.

What does Cowork actually do?

It runs long, multi step tasks rather than answering a single prompt. It drafts documents, sends mail, schedules meetings and posts in Teams on your behalf, and reusable skills capture how you want a workflow done so the output stays consistent across runs.

What is base Copilot adoption actually running at?

Measured active use of base Copilot sat between 25 and 45 percent of assigned seats at the six month mark in the engagements we advised on. Agentic capability layered on top of that gap widens the distance between licence spend and realized value rather than closing it.

Where does the negotiation actually happen?

In the underlying Copilot commitment: seat count, term length, and price protection. Since Cowork has no price of its own, any expansion justified by the agentic demo is expansion of the thing that does have a price, which is where the leverage and the exposure both sit.

How should governance be handled?

Upstream of rollout, not after it. Agent governance runs through Agent 365, and it was an afterthought in about half the rollouts we saw, which created audit and data exposure risk. Agents act inside the same applications your staff use, so controls and audit logging belong in place before enablement.

Does model choice matter commercially?

It matters for diligence. Anthropic Opus is selectable inside Cowork alongside the Microsoft models, so confirm which model processes your data and under what terms. That is a contractual and compliance question rather than a pricing one, and it is easier to settle before enablement.

What adoption level justifies expanding Copilot?

Tie any expansion to measured base adoption, and keep new commitments short until that number clears roughly 60 percent. Capability is not adoption, and expanding a licence estate on the strength of a demo prices the capability before the value is proven.

What is the honest measure of Cowork value?

Agent audit logs rather than demonstrations. Because Cowork acts inside the applications people already use, its footprint is observable: what ran, how often, and for whom. Those logs answer the value question in a way a scripted demonstration cannot.