Over the past few days I've noticed both excitement and concern in a lot of Microsoft 365 tenants. Microsoft announced Microsoft 365 Copilot Wave 2 on September 16th, and with it came a rebrand ("Copilot for Microsoft 365" is now officially "Microsoft 365 Copilot") plus a batch of new AI features, most of them in preview. The admins I work with are torn. The new possibilities are genuinely interesting, but governance, user communication and ongoing licence management don't go away. Since so many features are preview-only, you have to plan pragmatically from the start.

Most Wave 2 features arrive as previews over the coming months

Wave 2 is mainly a set of feature announcements, and a lot of it is preview. What interests me most is Copilot Pages, a persistent, shareable canvas built on Loop and embedded in Business Chat. Microsoft obviously wants it to become the new hub for teamwork and knowledge. Agents are arriving too. You can create specific Copilot agents with the Agent Builder in Business Chat, and SharePoint agents are on the way (preview). Excel now offers Copilot with Python functions (preview), PowerPoint gets a Narrative builder, and Outlook adds a "Prioritize my inbox" feature. The full list is in Microsoft's announcement and the official documentation.

As usual, real availability is patchy. Copilot Pages and agents are preview-only for now, typically for E3/E5 customers with Copilot licences, and mostly in English. If you expect instant productivity gains, you'll need patience. Rollouts will take months, and you can't skip the pilot phase. Even so, I'd start on the governance and communication groundwork now so you're ready when the features become broadly available.

Pages and agents bring old governance risks to a new canvas

Copilot Pages and agents give users a lot of autonomy all at once. That's good for innovation and bad for information protection. Agents can automate workflows, pull data together and interact with users, so admins have to watch closely what gets built and shared. The governance questions are the same ones we already know from Power Platform and Loop. You need to decide who can build agents and which data sources are allowed, and you need a way to monitor usage and access over time.

Microsoft has published some initial governance guidance for Copilot extensibility, but a lot of it is still vague, particularly during preview. Right now you manage the controls in the usual admin centers. You can enable or restrict Copilot Pages and agents for the whole tenant or for specific groups, but granular auditing and lifecycle automation are missing. Be especially careful with SharePoint agents. If permissions aren't set up properly, they're a fast track to oversharing. I'd apply the same principles I describe in /blog/restricted-sharepoint-search-copilot-oversharing to these new features.

This simple diagram shows how Copilot Pages, agents and data fit together:

flowchart TD
    User["User"] --> Pages["Copilot Page"]
    Pages --> Agent["Agent"]
    Agent --> Data["Data source"]
    Agent --> Action["Tenant action"]

Figure: How Copilot Pages, agents and data access interact.

Agents and Pages are persistent, so plan for visibility and clean-up controls early. Sensitivity labels, retention policies and regular access reviews are the obvious candidates.

Users need clear expectations and clear boundaries

These changes will raise questions, and fair enough. Many users don't understand how Copilot Pages and agents differ from the tools they already have, or why they can't use everything straight away. Communicating early pays off here, and not everyone needs every preview feature on day one. Be open about which features are live, what data gets processed and where the organisation draws the line. In my experience, short intranet updates that explain what's new and show the immediate benefit work well.

Practical tip: Start a FAQ for Copilot Pages, agents and governance now. It spares IT support a lot of repetitive questions and keeps users from getting confused.

You should also set up feedback channels where users can report problems and suggest improvements. That's particularly useful while the preview keeps changing.

Licence management and oversharing remain permanent admin tasks

Renaming the product to "Microsoft 365 Copilot" doesn't change the licence model. Copilot still needs an add-on licence, and for most organisations tracking assignment and usage is an ongoing job. You need to know who has Copilot and whether they actually use it. Proper reporting and assignment processes help here, as I explain in /blog/copilot-usage-reports-licence-assignment.

Oversharing is still one of the main risks in Wave 2, and Copilot Pages and agents make it more pressing. Copilot can access everything a user can see, whether that content was shared on purpose or not. Sensitivity labels, SharePoint permissions and user training are still the basics. Because agents and Pages collect and keep information, the risk of data leaking is higher than with ordinary chats. My advice is to review SharePoint and Loop governance together with your sensitivity labels and to set up regular review cycles. With so much still in preview, the complexity and uncertainty won't go away soon.

I see real potential and plenty of homework in Wave 2

Microsoft 365 Copilot Wave 2 delivers a lot of innovation, along with a fair amount of ambiguity. Admins have to switch on new features and protect the organisation at the same time. Good governance, careful licence management and clear communication matter more now than before. Personally, I think there's a lot of potential here, but also plenty of homework for the next few months. If you lay the groundwork now, you'll have fewer surprises later.