Skip to main content

Microsoft Switches On Use Case Classification of Copilot Cowork Work by Default for Frontier Tenants

Microsoft is turning on, by default, a feature that sorts Copilot Cowork activity and agent sessions into use cases. Who is caught, where the three switches are, and what to decide before the rollout reaches you.

KBY News Desk9 min read
Business professional analyzing bar chart on tablet in office setting, highlighting data insights.

Microsoft is turning on, by default, a feature that sorts the work people do in Copilot Cowork, and the work that agents do in a tenant, into a fixed list of use cases such as “executive summary writing” or “project scheduling”. It is a public preview for tenants in Microsoft’s Frontier programme, and the Message Center post announcing it (MC1557759, dated 17 November 2026) says the rollout begins in mid-November and is expected to finish by late November. Microsoft’s own wording is that “no action is required”. We think that is the wrong thing to tell an admin about a setting that is on until somebody turns it off, so this is where the controls are and what we would settle first.

#What Microsoft is switching on

Microsoft’s Learn documentation describes use case classification as a way to identify “the type of work people perform in Microsoft Cowork and agents perform in supported agent sessions”, matched against a built-in taxonomy. It covers two kinds of activity. The first is agent sessions: sessions with agents built in Copilot Studio or Azure AI Foundry, declarative agents, and third-party agents registered through Agent 365 connected platforms. The second is a person’s own work activity in Cowork. Microsoft says it is not yet available for Code, Autopilot or Researcher.

The taxonomy is predefined, in four levels (domain, category, sub-category, use case), and Learn says the same methodology is used in every tenant. Learn gives an example: documentation for a confidential internal project might be labelled “Functional requirements definition” and would not carry the project’s name. The results appear in the Microsoft 365 admin centre. For agents, that is the Activity tab on an agent and Agent Map. For Cowork, it is the Use cases tab under Cost management, then Consumption, and the Usage tab on the Cowork page.

The classification also feeds a second feature. Microsoft says AI value calculations, which put a cost and an estimated value against each use case, are on by default too, and are limited for now to Copilot Cowork. That is the commercial point of the exercise: the Message Center post says the insights help organisations see which use cases consume the most AI credits and find opportunities to consolidate agents.

#Who is and is not caught

Three conditions apply, all from Microsoft’s documentation. The tenant has to be in the Frontier public preview. Learn says use case classification is available to Frontier tenants with at least 50 Copilot licences, and that tenants below that will get it in a later release. For work activity in Cowork, Learn adds that the user must hold a Microsoft 365 Copilot premium licence; Microsoft’s support article for end users puts it as on by default for Frontier users who have Copilot licences.

If your tenant is not in Frontier, nothing changes today. If it is, the default is on for the whole tenant, for both kinds of activity, unless a policy or the tenant-wide switch says otherwise.

#What the classifier reads

This is the part we would want in front of a data protection lead. The support article says an AI classification engine analyses signals from a person’s work across Microsoft 365, Copilot, agents and Cowork. It lists three kinds: profile information (role, job title and function), activity signals (“key phrases from emails, meetings, and documents in Microsoft 365”), and AI tool usage data. It adds that the signals “may change over time”.

Microsoft’s reassurances are specific. Individual classifications are not visible to other people. Organisation leaders and administrators see work activity only as aggregated insights. Microsoft says it does not see the underlying content used to classify work. Learn adds that the results are “directionally accurate” and meant to show trends, “not to audit individual interactions”. We have no reason to doubt any of that, and it is the vendor’s account; the pages we read do not describe independent testing of it.

#Where the switches are

There are three controls, and they are separate. Learn names Global Administrator or AI Administrator as the roles that can change the first two.

  1. Agent sessions. In the Microsoft 365 admin centre, go to Agents, then Settings, then Use case classification, and turn it off for the whole tenant. There is no per-group option for agents. One discrepancy to know about: the Message Center post calls this page “Activity classification”, while the Learn page calls it “Use case classification”. We have followed Learn, and you should expect to find it under either name while the preview settles.
  2. Cowork work activity. This is a Feature access management policy. Search for Feature access management in the admin centre (or go to Settings, Viva, Viva settings), open the Use case classification tab, select Add a policy, choose the Use case classification app and feature, pick the groups or users, set the access, and save. Learn offers two settings: off completely, or off by default while letting people opt in from their profile. To cover the whole organisation, Learn’s tip is to create a security group that contains everyone and use that as the scope.
  3. AI value. The Message Center post places the switch for AI value calculations, and the inputs used for them, under Agents, then Settings, then AI value.

People can also act for themselves. Microsoft’s support article says a user can open their profile card, choose Update your profile, open the Data and privacy tab, and switch off “Allow AI to identify your use cases”. Learn says there is no equivalent for agent sessions yet.

When classification is turned off, Microsoft says it stops within about a day. Existing results are hidden, not deleted, and if you turn it back on it resumes from that point with no backfill of the gap. That matters for one reason: turning it off now and on later does not give you a clean record of the period in between.

#Why we would decide this before the rollout lands

Our view is that a default-on feature that reads email, meeting and document phrases to label what staff are doing deserves a decision, even though Microsoft’s design is aggregated and no individual is shown to a manager. There are two reasons.

The first is the order of events. Microsoft’s own documentation tells admins to use these controls to fit “local data protection obligations or Works Council guidelines”. That is a prompt to consult people before the setting is on, and the default does the opposite: classification begins at rollout and is only hidden if you turn it off afterwards. In the UK we would expect a data protection lead to ask whether the privacy notice given to staff covers this processing and whether the impact assessment for Copilot needs updating. We are not giving legal advice. We are saying the question will be asked, and it is easier to answer before the first classification exists than after.

The second is what the feature is for. The AI value numbers are a cost and justification tool for whoever signs the Copilot bill. That is a legitimate use, and it is also the use that makes people nervous about being measured. The aggregated-only design helps, but “directionally accurate” classifications attached to departments and groups can still be read as a verdict on a team. Learn itself warns against drawing conclusions from isolated results.

Where we lean: if you are in Frontier, leave agent-session classification on, because it describes what your agents do rather than what a person does, and it is the more useful half for spotting duplicate agents. For Cowork work activity, we would create the Feature access management policy that turns it off by default while allowing opt-in, and revisit once you have told staff what it does. That keeps the cost reporting for agents and gives people a choice about their own work. A different organisation, particularly one that has already consulted staff about Copilot analytics, could reasonably leave it on. Either answer is defensible. Finding out from the admin centre after the rollout is the one we would avoid.

#What to do this week

  • Find out whether your tenant is in Frontier, and whether it has at least 50 Copilot licences. If not, note the date and move on.
  • Open Agents, then Settings, and check what the use case classification page shows today.
  • Decide the Cowork policy and the AI value setting with whoever owns data protection and the Copilot budget, then record the decision.
  • If you leave it on, say so to staff in plain words, and tell them about the profile setting.
  • Ask your Microsoft account team, word for word: “Where is use case classification processed, and does it stay inside our data boundary?” The pages we read do not say.

We have covered the other recent Copilot default Microsoft is changing, Claude being switched on by default from 9 December, and our Microsoft Copilot technology page collects the rest. If a term here is new, our Copilot lexicon entry explains it.

#If you run a small business without an IT team

This almost certainly does not touch you yet. Microsoft’s documentation says the feature is for Frontier tenants with at least 50 Copilot licences, and Frontier is Microsoft’s preview programme for its newest AI features. If you have fewer than 50 licences, or never joined Frontier, there is nothing to switch off. If a provider manages your Microsoft 365, it is worth one question to them: “Are we in the Frontier programme, and has anything been switched on for us?”

#What we do not know

The pages we read do not say whether classification is processed inside the UK or EU data boundary, which signals are used for agent sessions as opposed to Cowork, or when the feature will leave preview and whether it will stay on by default when it does. We expect the default to carry into general availability, because Microsoft’s cost reporting depends on it. We would be wrong if Microsoft makes it opt-in before then. We will update this story if it does, or if Microsoft aligns the two names it uses for the agent setting.

Evidence trail

Sources and verification

Primary documentation and external technical references used in this article.

  1. 01Microsoft Learn: Understand use case classification in Microsoft Cowork and Agent 365 (Frontier)learn.microsoft.com
  2. 02Microsoft Support: Use case classification in Microsoft Frontiersupport.microsoft.com
  3. 03Microsoft 365 Message Center post MC1557759 (public archive copy)mc.merill.net

About this page

Credited to

Dates

Published

Corrections

No corrections are listed for this page. Some review sweeps corrected pages without naming each one; those are in the corrections log.

Reader Interaction

Comments

Add a thoughtful note on Microsoft Switches On Use Case Classification of Copilot Cowork Work by Default for Frontier Tenants. Comments are checked for spam and held for moderation before appearing.

Loading comments...

Get new fixes by email.

New runbooks, configuration traps and fixes from KBY Technologies, about twice a week.