Microsoft Copilot | Vlad Talks Tech

Power Platform Just Hit 80 Million Users: PPCC 2026 Keynote Highlights

Vlad Catrinescu — Thu, 29 Oct 2026 12:37:28 GMT

Last year at the Power Platform Community Conference, Microsoft said that low code as we know it is dead. So before the PPCC 2026 keynote, a lot of people in Las Vegas were asking whether the Power Platform is dead as well. Well, I have good news for you, because Charles Lamanna said it pretty clearly: the Power Platform isn’t going anywhere, and it’s more important and healthier than ever.

Key takeaways

The Power Platform now has 80 million monthly active users, up from 25 million in 2022 and 56 million last year.
Microsoft announced an expanded and unified Microsoft Maker Community for everyone building across Power Platform, Copilot Studio and SharePoint.
You can now build Power Apps directly from the Microsoft Copilot app, and the result is a real app with real code behind it.
The new Copilot Managed Runtime gives vibe coded apps a Microsoft-hosted home that follows the policies your admins set.
Agents in Copilot Studio can now get their own identity, with a mailbox and OneDrive, and agent identities are rolling out to everyone right now.
The conference leaves Las Vegas and heads to Chicago on November 2 to 4, 2027.

Inside the Power Platform Community Conference 2026

I was at the Power Platform Community Conference 2026 in Las Vegas for the keynote, and the announcements matter whether you’re a maker, an IT pro, or the person who has to explain all of this to your team back at the office.

If you’ve never been to this conference, you’re really missing out. There were nearly 10,000 makers there this year, and between the sessions there were puppies. Yes, actual puppies in the expo hall. There was a makers market, an NFL experience where you could see how they use the Power Platform and tech to help their teams perform better, and even a human claw machine where you’re the one picking up the prizes.

The Power Platform now has 80 million monthly active users

Charles kicked things off by reminding us how much the world has changed since the first Power Platform conference in 2022. Back then, if you heard the word agent, you thought of Agent Smith in The Matrix. A prompt was a blinking cursor in a DOS terminal, a token was for playing Pac-Man at the arcade, and a vibe was just a feeling.

Since then, the Power Platform has grown massively, and I have the numbers to prove it. In 2022, the Power Platform had 25 million monthly active users, and today it’s at 80 million. That’s more than three times as many people in just four years. If you’ve been reading my recaps, it was 48 million in 2024 and 56 million last year, so this is the biggest year-over-year jump I’ve seen.

And it’s not just more people using the platform, it’s people doing more with it:

20% more people open a Power App every single day compared to last year.
There are 20% more flow runs, which shows that my favorite product, Power Automate, is doing better than ever.
AI usage in Power Apps has doubled.

So if you were worried the Power Platform was dying, the numbers tell you otherwise.

A new, unified Microsoft Maker Community

A big part of that growth is the community, and the Power Platform community is thriving, with members in more than 140 countries and regions, and over 90% of the answers coming from people like you and me. It really is the community helping the community.

The first announcement of the keynote was about that community: Charles announced an expanded and unified Microsoft Maker Community, one place for everybody building across Power Platform, Copilot Studio and SharePoint. I’ll be honest, that last part caught my attention, and a few other MVPs around me looked at each other and said “SharePoint?” You can find the new community at aka.ms/MakerCommunity.

Remember that at Microsoft, the engineering teams for Power Platform and Microsoft 365 are now under the same team, and everybody building is now a maker. So I’m really curious to see how those two communities come together. It might even have an impact on next year’s conference, but that’s my personal opinion, and I come back to it at the end.

Better tools: the first demo had no AI at all

Every demo in this keynote was a real use case from a real company. It wasn’t Contoso, where every demo just works all the time. Ryan Cunningham even joked that he’s starting to think those aren’t actually real companies.

The keynote was organized around four pillars: better tools, more makers, no cliffs, and value at scale. And funny enough, while Microsoft keeps saying there’s never been a better time to build with AI, the first demo had no AI at all. It showed the fundamental improvements Microsoft has made for people who build manually.

Power Apps: a pop-out formula bar and faster controls

You can now pop out the formula bar in Power Apps. So when your formula is dozens or hundreds of lines long, you’re not stuck with a small bar to read it in: you can put it beside your screen. Every new control is also more than 30% faster.

Model-driven apps are getting calendar and card views that you set up with just a few clicks. You could build those with code in the past, but now it takes a few clicks, and the calendar looks just like Outlook.

Power Automate: inline action cards and virtual network connectivity

Power Automate got some love too, which made me happy, because it’s my favorite product and I was a bit worried last year that there wasn’t much for it. The designer now has inline action cards, so you can view and edit an action’s parameters right on the canvas instead of in a panel on the left. You see more at the same time, so it’s kind of a modern take on the original designer.

And for my fellow IT pros, virtual network connectivity lets a cloud flow call a service that’s on-premises without you needing to build a custom connector. It’s safe, it’s secure, and it’s easy.

More makers: build Power Apps from Microsoft Copilot

The next pillar was more makers, and this is where AI comes in. In the new Microsoft Copilot (formerly Microsoft 365 Copilot) app, you have the chat you already use, plus Copilot Cowork, Code, and the brand new Autopilot. What’s new is that you can now build Power Apps from Copilot.

Charles even said he’s built more Power Apps in Copilot in the last two months than in the platform directly in the past two years. An app is now his favorite way to share information, rather than a traditional document or a PowerPoint deck.

We saw a demo where a marine underwriter at Markel Insurance, so a business user, built a full app for her team in Cowork just by describing what she wanted. It’s connected to their data, it sends emails through Outlook, it updates records in Dataverse, and sharing it is as easy as sharing a Word document. The person closest to the problem builds the solution, which is what the Power Platform has always been about, just in a more amplified way now.

No cliffs: low-code apps are code underneath

If you’re an IT pro like me, you’re probably thinking: okay, what happens when that app becomes business critical? This is where the no cliffs pillar comes in.

It took me a while to understand what that meant. In the past, there was a big limit to what you could do with no code and low code. Once you reached it, you had to go to pro code, and that was a big cliff to climb, because you couldn’t just keep working on the same app in Visual Studio Code. You had to rebuild it completely.

But behind every low-code app, there’s actually code. So when you build an app with AI, either as a code app or in Copilot Cowork, it’s still code deployed in the back. As soon as you reach your limit as a vibe coder, a developer can connect to it and take your solution to the next level without rebuilding everything.

The keynote even showed a solution built in Copilot Studio that a developer then opened in Visual Studio Code. He asked GitHub Copilot to explain it, and it drew a diagram of the whole solution. It also found that 3 of the last 12 follow-up emails asked suppliers for information they had already given, and the developer fixed that without ever opening the Copilot Studio designer. So even if you never look at it that way in the UI, it’s code in the back, and with AI and the new ways of hosting, that transition is easier than ever.

The Copilot Managed Runtime fixes the localhost:3000 problem

Let’s talk vibe coding, because Ryan had my favorite joke of the keynote: “Check out my cool new app, it’s at localhost 3000. What? That doesn’t work on your machine?” I’m sure that as an IT pro, someone has already shared an app with you that way, whether a colleague or a family member. We’ve all been there.

That’s exactly what the new Copilot Managed Runtime fixes. Whether you build in Cowork, Copilot Studio, Visual Studio Code, or even a third-party tool, you now have a dedicated runtime where you can safely host your apps. It’s hosted by Microsoft, and it follows the policies your admins set.

Makers finally have a way to host vibe coded apps in a secure, enterprise-compliant way. Personally, though, I’m wondering where the line is. When do you keep an app on the managed runtime, where you pay per use with Copilot credits, and when does a developer take it over and put it on Azure Static Web Apps for a fixed price of around $10 a month, so you know exactly what it costs? It’s brand new, so I’m curious to hear what you would do.

Copilot Studio workflows and the GitHub Copilot harness

The keynote also covered the workflows experience in Copilot Studio, which became generally available in August. So it’s not brand new, but it was still interesting to see it live, combining deterministic steps with everything AI, like classifying emails, calling an agent, or even asking Copilot Cowork to add a task.

In the demo, an insurance company’s shared mailbox got classified, the workflow called the right agents and automations, and the whole run, from the email coming in to a quote ready for the underwriter, cost less than 90 cents. Not 90 dollars, 90 cents. You can also test a single step without running the whole workflow, and you can choose to lay it out left to right or top to bottom. Ryan had some strong feelings about that during the keynote, and I suspect that’s why the top-to-bottom option exists now.

Staying with Copilot Studio, the GitHub Copilot harness isn’t new either, but we got numbers on how much better it is. In Microsoft’s testing, with the same model and the same tasks, code analysis went from about 41% to 88% on the new harness, with about 45% fewer input tokens. So your agents get smarter and cheaper at the same time, and that doesn’t happen very often.

Agents get their own identity, mailbox and OneDrive

I think the biggest showcase for a lot of people was agent identity. Until now, Microsoft mostly pitched agent identity to IT pros, as something to manage, rather than as something that makes you more productive.

What we saw at this keynote is that because an agent has its own identity, it has its own mailbox and its own OneDrive. You can chat with it in Teams, add it to a group conversation, email it, or tag it in a Word document exactly as if it were a user. That applies to Autopilot as well: Charles shared that his own Autopilot is named Apollo, and if you look at the org chart at Microsoft, you’ll actually see Apollo reporting to Charles.

Agent identities are rolling out to everyone in Copilot Studio right now.

The first governance demo on a PPCC keynote stage

The keynote closed with something we’ve never had before at a PPCC keynote: a security and governance demo. With everybody building apps, flows and agents right now, you want to make sure you manage them properly. We saw Agent 365 in the Microsoft 365 admin center, which isn’t new, but it was good to see it demoed.

My highlight was the Power Platform admin plugin with the Power Platform Advisor. It’s not new this week, but it was the first time I saw it live. It’s like an admin agent for Power Platform admins: you go into Cowork and ask it to build reports on what has been created in your environment. It reminds me of the SharePoint admin agent, but for the Power Platform, and I love that Microsoft is making governance easier for every company. It’s the first thing I want to try back at the office.

What this means for you

If you’re a maker, check out all the new experiences and see the new ways you can create apps and flows with AI, whether you’re building them with AI or putting AI inside them. If you haven’t tried it yet, plans in Power Apps are a good place to start building a whole solution from a prompt. If you have Copilot Cowork at work, try building an app with it. Charles’s idea of building an app or a dashboard to share information, instead of a PowerPoint deck or a Word document, is worth trying. It won’t fit every time, but it’s so much more flexible.

If you’re an IT pro, check out the latest in how to manage your environments. As more makers build apps, flows and agents, you need to keep an eye on them and keep them in check. Try the new Power Platform admin plugin, and see which of the tasks you used to do by hand or with PowerShell you can automate with AI. We’re about to see thousands of apps created every year, so keeping an eye on everything matters more than ever.

PPCC 2027 moves to Chicago

The Power Platform Community Conference is leaving Las Vegas and heading to the Windy City: Chicago, November 2 to 4, 2027. There’s a great promo if you register now, with 50% off and a free workshop if you do it fast, and you can register for the 2027 conference here.

The biggest change for me is the link itself: it’s now a Maker Conference. Charles included SharePoint in the new maker community, so I’m curious whether the 2027 conference will include more SharePoint and Microsoft 365 content, not just Power Platform. Will Microsoft have one big conference? Nothing has been announced yet, but it’s worth thinking about as the definition of a maker, and the tools makers use, keep changing.

Ready to plan your trip to Chicago? I’ll share every new PPCC promo on my Power Platform Community Conference page, so you can register for Chicago at the best price.

How Agent Approval Works in Microsoft 365

Vlad Catrinescu — Thu, 22 Oct 2026 09:41:16 GMT

Your users are already building agents in Microsoft 365, but what happens when somebody’s experiment becomes available to everyone? You need to know how and why, because agent approval in Microsoft 365 depends on the type of agent, and each type gets there differently.

Key takeaways

SharePoint agents no longer have an approval process, and the site owner decides which agent is the main agent for the site.
By default, anyone can share an Agent Builder agent with the whole organization without admin approval, and an admin can change that setting.
A Copilot Studio agent needs an admin to approve it before everyone in the organization can use it.
An agent only responds with knowledge each user already has access to, except for files uploaded directly to an Agent Builder agent.
Admins approve agent requests in the Microsoft 365 admin center, under Agents, All agents, Requests.

Agent approval in Microsoft 365 depends on the type of agent

How an agent goes from being built by a maker to being available for everyone depends on the type of agent:

SharePoint agents are managed by the site owner.
Agent Builder agents in Microsoft Copilot can be shared with specific users or with the whole organization, without any admin approval. That’s the default behavior if you just deployed Copilot and didn’t change anything, and an admin can change it.
Copilot Studio agents can be shared with specific people without an approval. To share one with the whole organization, the maker needs an administrator to approve it.

If you’re preparing for the AB-900 exam, remember the default approval process for every type of agent. The AB-900 study guide lists understanding the approval process for agents as one of the skills measured.

SharePoint agents don’t have an approval process anymore

There is no approval process for SharePoint agents anymore. There used to be one, but it was removed in June 2026. You’ll still find the Approved folder in the site’s Site Assets library, but it doesn’t have the same purpose it did before. It’s just another folder now, and if you want, you can still put your agents in there so you know where your approved agents are.

The closest thing to an approval process is the Site AI pane. A site owner can open it and choose the main agent for the site, and there can only be one per site. To highlight more agents, site members, or anyone with edit permissions or higher, can add the agent link web part to any page, like the home page.

If you’re responsible for SharePoint agents more broadly, I cover the reports and policy levers in SharePoint agent governance.

A maker can share an Agent Builder agent either with specific users they choose, or with anyone in the organization. By default, that second option is enabled, so anybody can share an agent with everyone in the organization without an approval. That’s the default that surprises most admins.

I know that can sound a bit scary, but remember that the agent will only respond with knowledge the end user has access to. If the agent’s knowledge includes a SharePoint site that another user can’t access, Copilot won’t use that site to answer that user, because it always respects the user’s permissions. If you want the full picture of how that works, I explain it in how Microsoft Copilot accesses your work data.

There is one thing to remember for Agent Builder. If you upload a file directly to the agent, that file travels with the agent itself, so everyone who has access to the agent can get answers from the uploaded file.

Copilot Studio agents need an admin to approve them

Copilot Studio has a more defined workflow. A maker submits the agent to the organization catalog, and an administrator needs to approve it before other users can see it.

Copilot Studio also has many ways to configure knowledge sources, whether through service accounts or by connecting as the user talking to the agent. So what a user can reach really depends on how the agent was configured. But as an admin, you still have that approval step before it’s available organization wide.

An admin can approve agents in two admin centers: the Microsoft 365 admin center, where Microsoft Agent 365 lives, and the Teams admin center. For the AB-900 exam, focus on the Microsoft 365 admin center. That’s where Microsoft has been unifying its agent management capabilities, and the unified agent inventory is part of that same story.

How each approval flow looks in practice

Set the main agent on a SharePoint site

As a site owner, go to the site settings and open Site AI. Under Main agent, pick the agent visitors open from the agent icon in the header, and decide whether to show the Copilot button to site visitors. Save, and that agent becomes the default agent for the site. You can only set one default for the whole site.

To showcase more agents, edit a page and add the agent link web part. You can select any other agent in your organization, and your users can click the web part to chat with that agent.

Open the agent in Agent Builder, select Share, and choose who it’s available to: only you, specific users in your organization, or anyone in your organization. Then copy the link, select Apply, and anyone in the organization can start using the agent from that link.

Remember that the agent will only respond with knowledge each user has access to, and that applies to SharePoint agents as well.

Submit a Copilot Studio agent to the org catalog

Copilot Studio is the more advanced one, and the exact steps to send an agent for approval aren’t on the exam. Once the agent is published to the Microsoft Copilot and Microsoft Teams channel, the maker can submit it to the organization catalog to give everyone access. The agent then waits in an approval queue, pending admin action.

Approve or reject the request in the Microsoft 365 admin center

As an admin, here’s how to review the request:

In the Microsoft 365 admin center, go to Agents, then All agents.
Open the Requests tab. Select Refresh if the agent doesn’t show up yet.
Select the agent to open its details.
Choose Publish to store or Reject submission.

Before you decide, you have access to the information about the agent: when it was created and last updated, its channels, its data and tools, insights from Microsoft Purview and Microsoft Defender, and the Microsoft Entra Agent ID permissions when the agent uses one.

When you publish it to the store, you choose who gets the agent, such as all users, and whether to pre-install it for them. You can also apply a policy template to it, which is a Microsoft Agent 365 feature. Once you make it available, the agent is available for everybody in the organization.

What this means for you

So who approves an agent? It depends on the agent type, and the Microsoft 365 admin center is where you stay in control. SharePoint agents follow the site owner, Agent Builder agents follow a sharing setting that’s open by default, and Copilot Studio agents wait for you in the Requests tab.

Go take a look at what agents are already live in your tenant. Check whether the Agent Builder sharing setting is what your organization wants, and know who’s watching the Requests tab so Copilot Studio submissions don’t sit in the queue. And remember that a file uploaded directly to an Agent Builder agent is available to everyone who can use that agent.

If you’re studying for the AB-900, make sure you can name the default approval process for each type of agent, and spend your time in the Microsoft 365 admin center rather than the Teams admin center.

Ready to get AB-900 certified? This walkthrough comes from my three-course AB-900 path on Pluralsight, which covers the core Microsoft 365 services, data protection and governance, and the basics of Copilot and agent administration. Start with my AB-900 path on Pluralsight, and keep the AB-900 study guide open while you study.

Copilot in SharePoint FAQ: Your Agents, Skills and Autofill Questions Answered

Vlad Catrinescu — Wed, 21 Oct 2026 15:48:18 GMT

Fellow Microsoft MVP Gokan Ozcifci and I ran over time in our Copilot in SharePoint webinar, probably because I went about 15 minutes over the 20-minute slot he gave me. That left a lot of your questions unanswered, so as I promised at the end, I’ve answered every one of them here. Everyone who registered gets the recording by email, and if you didn’t, you can still watch the recording and get the slides.

Key takeaways

Copilot in SharePoint needs a Microsoft Copilot license (formerly Microsoft 365 Copilot) for each user. General availability started rolling out on September 30, 2026, so it can take a little longer to reach your tenant.
Everyday work, like creating skills and using autofill, is included with that license. Advanced work, like image generation, analytics, real-time autofill and anything at scale, uses Copilot Credits, through a spending policy your admin sets up.
Classic SharePoint agents saved as .agent files keep working until the end of March 2027, and starting in mid-November 2026, you build new agents for SharePoint with Agent Builder.
Copilot in SharePoint runs on a Microsoft-managed reasoning model from OpenAI that you can’t change, but you can pick the model for an agent you build in Copilot Studio.
Autofill columns are library columns that work with managed metadata, and there’s no supported way yet to reuse one in other libraries or sites.
Restricted Content Discovery, which your Copilot licenses unlock through SharePoint Advanced Management, keeps Copilot in SharePoint and agents off a site without changing who can open it.
Copilot in SharePoint isn’t available in GCC, GCC High or DoD yet.

Getting Copilot in SharePoint in your tenant

How do I know when Copilot in SharePoint is available in my tenant?

Look for the Copilot in SharePoint button in the lower-right corner of your SharePoint pages and document libraries. General availability started rolling out on September 30, 2026, and Microsoft’s get started guide for Copilot in SharePoint says the capabilities may appear in your tenant after that date as the rollout progresses. So if you don’t see it yet, don’t worry. As Gokan said during the webinar, it’s still rolling out.

And yes, it’s the paid license. Each user needs a Microsoft Copilot license, and with it the everyday features are included at no extra cost, with no opt-in step. Copilot Credits don’t replace the license either. They pay for advanced work on top of it, which I cover below.

As for the pay-as-you-go option I mentioned in the chat, there’s no pay-as-you-go way to use Copilot in SharePoint without the license. SharePoint agents pay-as-you-go only applies to .agent files, which stop working after March 2027, and Microsoft’s upgrade guidance for agents in SharePoint says it doesn’t apply to Copilot in SharePoint. For autofill columns you set up with Copilot in SharePoint, new files are covered by the Copilot license, even when the library is connected to a pay-as-you-go meter. Pay-as-you-go isn’t gone from Copilot altogether: as the .agent files retire, unlicensed use of agents moves to Copilot app pay-as-you-go.

If you’re an admin, the preview PowerShell settings, KnowledgeAgentScope and its list of sites, are only honored until November 1, 2026. After that, you manage where Copilot in SharePoint shows up with the Site AI settings and Restricted Content Discovery. If you used those settings to limit Copilot in SharePoint to a pilot, or to turn it off, plan for it to appear on every site for every licensed user after that date. If you excluded sites, decide which of them truly need Restricted Content Discovery, because it also hides them from organization-wide search and Copilot responses across Microsoft 365.

Does Copilot in SharePoint work in GCC?

Not yet. Microsoft’s get started guide says GCC, GCC High, DoD, Office 365 air-gapped cloud environments and Microsoft 365 operated by 21Vianet don’t currently support Copilot in SharePoint. Message Center post MC1511052 says it’s coming soon to the government clouds, without a date. Copilot Credits aren’t an option there yet either, because usage-based billing isn’t supported in GCC or GCC High.

Microsoft says the agent changes I cover below also apply to its government clouds, and upgrading to Agent Builder is supported there, though Agent Builder isn’t currently available for declarative agents in DoD. Converting an agent to a skill has to wait until Copilot in SharePoint arrives.

What is Site AI?

Site AI is the settings panel where site owners decide how people use AI on their site, Copilot in SharePoint included. You open it from Settings > Site AI, and it’s the panel I showed in the demo for setting the site’s main agent. From there, a site owner can:

Choose which agent opens from the agent icon in the header, either Copilot in SharePoint or a custom agent.
Hide the Copilot button on site pages for site visitors only.
Turn on advanced document processing, which today means a site admin sharing their own Copilot Credits so autofill uses advanced processing across the site.

On my demo site, Copilot was the main agent and credit sharing was off, so autofill there uses standard processing, which the Copilot license covers.

Hiding the button only applies to visitors, so to take Copilot in SharePoint off a site completely, admins turn on Restricted Content Discovery instead. The main agent choice also has an end date: after March 2027, Copilot is the only main agent a site can have, and Microsoft will remove that setting.

What the Copilot license covers, and when you need Copilot Credits

Does creating a skill cost extra?

No. Creating and managing skills is on Microsoft’s list of everyday work that the Copilot license covers, so a skill like the expense policy check I built in the demo costs nothing extra to create. Whether it uses credits depends on the work it does when it runs. Microsoft calls this a hybrid model, and its get started guide lists the everyday work the license includes:

Asking questions, finding content and catching up
Creating HTML, Word, Excel or PowerPoint files
Creating and editing pages, libraries or lists
Simple rules, approvals and deterministic workflows
Renaming and sharing individual files
Creating and managing skills
Using autofill metadata

Copilot Credits, through usage-based billing, pay for the advanced work:

Working across large or unbounded content sets
Creating entire sites and solutions
Creating and editing files, pages, libraries and lists at scale
Multistep, complex AI workflows, like skills in scheduled AI automations
Real-time autofill metadata
Generating new images and editing existing ones
Analytics on how sites, pages, news and documents perform

That’s the same split Gokan showed on his licensing slide, with his advice on who needs what along the bottom:

The “No user should invest in Syntex” line at the bottom of the slide is only Gokan’s and my opinion. Microsoft hasn’t announced that Syntex is being deprecated, and its pay-as-you-go services are still available under a new name, document processing for Microsoft 365.

Microsoft’s rule of thumb is that credits apply when a job “spans more content than you can work through yourself, needs to run on its own, or has to keep pace as content arrives.”

The 10,000 files I mentioned come from Microsoft’s Copilot Credits Guide, where it’s one of the heavy examples. Its medium examples, like re-tagging the documents in a project folder of 500 files, need credits too. Microsoft doesn’t publish a file count where “at scale” starts. Instead, when a prompt crosses that threshold, Copilot asks you to approve credit use before it continues. If you don’t have credits, it completes what it can and asks you to start a new chat. If you missed my first skills webinar, you can watch the Getting Started with SharePoint Skills recording for the basics.

Your admin turns on the advanced work in the Microsoft 365 admin center under Copilot > Cost management. You add a spending policy, choose all users or specific security groups, select Advanced work in SharePoint under Agents and services, then set limits and alerts and a billing method. If you already have spending policies, review them, because a policy that allows new agents and services as they become available can pick up Advanced work in SharePoint automatically.

Until a spending policy covers you, you only get the everyday work. Even /cost, the command that shows how many credits a task used, isn’t available yet, as I found on my demo site:

Gokan’s advice was to license everyone for the everyday work and give credits only to the people who manage content at scale. Microsoft’s guidance agrees: it prioritizes content managers and site managers, and says most other SharePoint users likely don’t need credits.

Will the prebuilt SharePoint page and list agents need Copilot Credits?

Nothing Microsoft has published ties them to credits, so the answer Gokan and I gave live holds: they need the Copilot license. The SharePoint list agent left the Frontier program and became generally available in February 2026, and it requires a Copilot license. The SharePoint page agent, a Frontier preview, is being retired between mid-October and mid-November 2026, according to Message Center post MC1481325. Gokan mentioned live that Microsoft had deprecated one agent a few weeks earlier, and that notice went out on September 29, 2026.

The same notice says you can keep creating and refining pages and news posts with Copilot in SharePoint. That’s everyday work the license covers, since creating and editing pages, libraries and lists is on the included list. Only doing it at scale needs credits.

Can everyone use an agent I put on a site?

They need either a Copilot license or pay-as-you-go that your admin sets up. With the license, using agents grounded in your organization’s content is included at no extra cost. For people without one, it depends on how the agent was built:

A classic SharePoint agent, the .agent file, uses SharePoint agents pay-as-you-go. Your admin creates a billing policy, and only users in the security group assigned to it can use agents. Microsoft lists the pay-as-you-go price at $0.01 per message, and a successful interaction uses 12 messages.
An Agent Builder agent that uses SharePoint content is billed through pay-as-you-go, consumed as Copilot Credits. Agents like that are off by default for Microsoft Copilot Chat (formerly Microsoft 365 Copilot Chat) users.

As the .agent files retire, Microsoft consolidates unlicensed use under Copilot app pay-as-you-go. For SharePoint content, people need permission to both the agent and the content it uses, so an agent won’t surface a SharePoint file someone couldn’t already open. The exception is files a maker uploads into an Agent Builder agent: anyone who can use that agent can get answers from them.

Does testing an agent in the preview pane use credits?

Not if you have a Copilot license, which is how I tested the Zava Travel Assistant in the demo. Testing an agent is interactive use, and Microsoft includes interactive use of agents by Copilot-licensed users at no extra cost.

Microsoft doesn’t document whether the preview pane in SharePoint, or the Try it tab in Agent Builder, is billed for someone who builds on pay-as-you-go without a license. If that’s you, I’d plan as if a test that reads SharePoint content is metered like any other use.

The only place Microsoft spells it out is Copilot Studio, where testing an agent in the test chat doesn’t consume Copilot Credits for the agent’s own activity. The exceptions are AI prompts or models inside agent flows, and agents built on the GitHub Copilot harness, which bill while you test.

SharePoint agents are moving to Agent Builder

Is a SharePoint agent basically an Agent Builder agent now?

Yes, and from mid-November 2026 that’s exactly what it will be. Today’s SharePoint agents, which I called classic agents in the demo (my naming, not Microsoft’s), are declarative agents saved as .agent files on the site. Microsoft’s upgrade guidance says that starting in mid-November 2026, you’ll use Microsoft Copilot Agent Builder to create agents for SharePoint and OneDrive, and creating an agent no longer saves a .agent file. You still pick SharePoint and OneDrive content as knowledge, and Agent Builder adds sources like Teams chats and meetings, Outlook email and Copilot connectors.

Microsoft’s timeline, which I put on one slide in the webinar, gives you until the end of March 2027 to move your classic agents:

In mid-November 2026, new agents for SharePoint are created with Agent Builder, an Upgrade to Microsoft 365 agent button appears in the builder, and the convert-agent-file-to-skill skill arrives in Copilot in SharePoint.
In November 2026, the Agent Chat web part, which puts an agent’s chat right on a SharePoint page, starts its preview, with general availability planned for January 2027 on the Microsoft 365 roadmap.
After March 2027, .agent files stop responding in SharePoint, Teams and the Copilot app.

There’s no automatic upgrade, so plan each one. An agent that mostly answers questions about its own site can be replaced by Copilot in SharePoint. One that runs a repeatable task becomes a skill, and a named Q&A or retrieval experience becomes an Agent Builder agent. Permissions don’t carry over, because an Agent Builder agent is shared on its own instead of following SharePoint file permissions. Unless the original was shared with everyone in your organization, the upgraded agent starts out private to you, and the .agent file’s usage history doesn’t transfer.

It hasn’t been removed: .agent files keep working until the end of March 2027. My best guess is a change in what the agent icon opens. For Copilot-licensed users, Copilot in SharePoint replaced the ready-made site agent. Unless a site owner picks a custom agent as the main agent in the Site AI settings, Copilot is now the main agent for the site.

Two other things can hide it too: Restricted Content Discovery is turned on for the site, which removes the agent icon from the header, or you don’t have a Copilot license and your organization hasn’t set up pay-as-you-go for agents. And if you mean the Copilot app rather than SharePoint, Microsoft is replacing its Agents entry point with Plugins, and upgraded agents show up in the plugin store.

Does a SharePoint agent get a Microsoft Entra ID, and does it show up in Agent 365?

No to the first, yes to the second. Microsoft’s agent identity documentation says Agent Builder agents currently don’t use or require app registration IDs or Agent IDs. Of the tools in this post, it’s new agents you create in Copilot Studio that get a Microsoft Entra Agent ID automatically. Microsoft doesn’t mention one for classic SharePoint agents either, and their answers depend on each user’s permissions to the agent’s sources.

For admins, agents show up in the agent registry in the Microsoft 365 admin center, under Agents > All agents > Registry. It lists agents with or without a Microsoft Entra agent identity, and SharePoint is one of its agent types. For classic agents, the registry gives you an overview of the agents that have been used in Copilot Chat, and blocking one there only affects its availability in Copilot Chat.

Agent Builder agents, including the ones you’ll create for SharePoint from mid-November, are integrated with Microsoft Agent 365 automatically. Viewing the registry doesn’t need a specific license, while Agent 365’s observability and policy features need Microsoft 365 E7 or an Agent 365 license.

Can an agent answer only from approved documents?

For your readers, mostly yes. People who can read a library but can’t see drafts get the last approved, major version of a file, and SharePoint search doesn’t return files that only exist as a minor version. So a visitor who only sees the published version gets answers from the published version.

The exception is the library’s Draft Item Security setting. If it’s set to “Any user who can read items”, everyone can see drafts, including the account that crawls for search. Editors and approvers can see drafts in the library too, so test the agent with an account that can’t.

For policies, I’d still keep it structural: keep final, approved documents in their own library or folder, keep work in progress somewhere else, and point the agent only at the final location. Someone in the chat suggested adding “only published documents” as a condition in the agent’s instructions. That’s fine as extra guidance, but don’t rely on it, because instructions aren’t a security boundary.

Can I choose the model Copilot in SharePoint uses?

Not in Copilot in SharePoint. Microsoft’s get started guide says it currently runs on a Microsoft-managed reasoning model from OpenAI, that Microsoft selects and manages the model, and that the specific model might change over time. So my answer holds, and I still wish we could choose. Microsoft doesn’t document a model picker for Agent Builder either, because those agents run on the same models Copilot uses.

Gokan’s workaround from the webinar is the one that works today: build the agent in Copilot Studio, add your SharePoint site as knowledge, and select the agent’s model from the dropdown, which includes Anthropic Claude models.

External models have to be turned on in both the Power Platform admin center and the Microsoft 365 admin center, where the Anthropic setting is on by default for most commercial customers outside the EU, EFTA and the UK. Anthropic models are also currently excluded from the EU Data Boundary, so check with your compliance team. Then publish the agent to the Microsoft Copilot app, so you can put it on a page with the Agent Chat web part. Publishing it to the SharePoint channel creates a .agent file, which stops working after March 2027.

Autofill columns

Do autofill columns work with site columns and the term store?

The term store, yes. Site columns, not officially. Autofill works with managed metadata columns, so you can map an autofill column to a term set from your term store, as I showed in my video on autofill columns with managed metadata. Microsoft’s autofill documentation notes that it only uses the first 100 terms of a mapped term set, and when a preferred term or a synonym matches, it saves the preferred term.

Besides managed metadata, autofill supports text, multiple lines of text, number, yes/no, date and time, choice, hyperlink and currency columns. It doesn’t support person or group, location, image or lookup columns.

The autofill column itself is a library column. You write its prompt in that column’s settings, and the values it extracts are saved to the column, so like other column data they can be indexed and used to trigger workflows. As Gokan said, there’s no supported way to reuse an autofill column in another library or site today. The site column workaround I mentioned is real, and the credit goes to Kasper Larsen. He showed how to reuse autofill prompts across libraries that share a content type with PnP PowerShell, building on Leon Armston’s earlier work. It relies on settings Microsoft hasn’t documented, so treat it as unsupported.

Does turning on autofill add the column to every library on the site?

No, only to the library you’re working in. Autofill columns are created one library at a time: Copilot in SharePoint can configure document libraries only, and even when you start from a folder, the configuration applies to that whole library. So keeping autofill in one library is simply a matter of creating autofill columns only in that library, and each column has its own toggle if you want to switch it off later.

In Gokan’s demo, Copilot suggested the columns for the one library he had open, showed a value for each invoice, and asked which columns to keep:

There’s no per-library on switch, because everyday autofill comes with the Copilot license without an opt-in. The controls above the library are broader. Restricted Content Discovery takes Copilot in SharePoint off a whole site, and in the Site AI settings, a site admin decides whether to share their Copilot Credits for advanced autofill. Once that’s on, autofill uses advanced processing whenever it runs on the site.

Can I roll out autofill columns across hundreds of libraries?

Not with anything Microsoft supports today. Microsoft doesn’t document an API, PowerShell cmdlet, site template setting or content type hub option for autofill, so as Gokan said live, you set up the column in each library. In his words, if you have 10 document libraries, you copy and paste the prompt 10 times.

What scales differently is the processing. Standard autofill is included with the Copilot license, and new files uploaded to the library are processed whether or not the uploader has a Copilot license. Real-time autofill metadata, the site-wide processing that keeps pace as content arrives, needs Copilot Credits and usage-based billing turned on for the site. Microsoft doesn’t publish a file count for “at scale” here either.

If you have to do it now, the community has tools. Leon Armston automated autofill columns with PnP PowerShell, and the CLI for Microsoft 365 has a command that sets autofill on a library column. Neither is supported by Microsoft. Leon’s post warns that the API behind it isn’t documented and is subject to change, and he has since noted that the backend APIs changed, so his original script may not work as written. Test carefully before you run anything across a tenant.

If you need the same extraction to run across many libraries in a supported way, Gokan’s slide had the other options: AI Builder models with Power Automate and Power Apps, Copilot Studio, and Azure Document Intelligence in Foundry Tools.

Protecting your content

Can I use SharePoint Advanced Management to keep sites out of Copilot?

Yes. As soon as at least one person in your organization has a Microsoft Copilot license assigned, your SharePoint admins get the SharePoint Advanced Management features included with Copilot. Restricted Content Discovery is the one that does what you’re describing. Copilot reaches your content through Microsoft Graph and the semantic index, which Microsoft builds automatically from your SharePoint content, and it only returns content the person asking can already access. My post on how Microsoft Copilot accesses your work data walks through how that works.

Restricted Content Discovery keeps a site’s content out of organization-wide search and Microsoft Copilot responses, including files people recently worked on. Copilot in SharePoint and AI actions don’t appear on that site at all, and people can’t use the ready-made agent, create agents, or add that site’s content to other agents. It doesn’t change who can open the content, so people who already have access can still open it directly, search within the site, and ask Copilot about a document they have open. You turn it on in the SharePoint admin center from the site’s settings in Active sites, or with PowerShell:

Set-SPOSite -Identity <site-url> -RestrictContentOrgWideSearch $true

The other controls you mentioned are on the same list of features your Copilot licenses include:

Restricted Access Control limits a site to the members of up to 10 groups. People outside them can’t open the site, even with existing permissions or a sharing link, and Copilot and search honor it too.
The block download policy gives a site browser-only access, with no downloading, printing or syncing.
Authentication contexts connect a Microsoft Entra Conditional Access policy to a SharePoint site, which is my best guess at the limited access feature you meant.

Microsoft calls Restricted Content Discovery a temporary governance control and recommends using it selectively, and on sites with more than 500,000 items it can take more than a week to fully apply. My post on SharePoint agent governance shows how to apply it straight from the agent access reports.

Can admins sanitize the HTML that Copilot creates?

Not today, and I haven’t found any sign of it coming: nothing on Microsoft Learn or the Microsoft 365 roadmap mentions an option to sanitize generated HTML.

Microsoft’s support article on HTML pages in SharePoint says SharePoint renders HTML in a sandboxed iframe, with Content Security Policy rules for extra isolation, and blocks arbitrary fetch requests and outbound API calls. JavaScript only runs when it’s contained in the same HTML file, apart from Microsoft-managed versions of a few libraries like Chart.js and Mermaid. External links also need a site admin to add their domain to HTML Field Security. That’s probably why the JavaScript in my expense report didn’t behave in the demo.

The same article answers your tenant question directly: you can’t disallow or restrict HTML pages at the tenant level. Site owners can turn off Allow HTML files in the Site Pages library, and that setting matters more soon. HTML pages, which let you publish an HTML file as a SharePoint page, reach general availability between mid-November and late December 2026, according to Message Center post MC1479517. For the external sharing risk, use the controls you already have:

The site’s external sharing settings
A data loss prevention (DLP) policy using the “File extension is” and “Content is shared from Microsoft 365” conditions
The block download policy on sensitive sites

Sensitivity labels won’t help at the file level, because HTML isn’t one of the file types they support in SharePoint. If a tenant-level option matters to you, send it to the product team as feedback. In the video they recorded for the webinar, they asked us to keep it coming.

What this means for you

Gokan and I focused the webinar on using Copilot in SharePoint, and most of your questions were about what it costs and what’s changing with agents. Both come with dates, so if you’re an admin, put these in your calendar:

Before November 1, 2026, check whether you used the preview PowerShell settings to limit Copilot in SharePoint to a few pilot sites or to keep it off some sites. After that date, Microsoft stops honoring them and Copilot in SharePoint shows up on every site for everyone with a Copilot license, so turn on Restricted Content Discovery early for any site that truly needs to stay out.
Before people start working at scale, decide who gets Copilot Credits and create a spending policy for Advanced work in SharePoint scoped to them. Check whether your existing spending policies already picked it up, too.
Between mid-November 2026 and the end of March 2027, inventory your .agent files with the agent registry, the SharePoint Advanced Management agent reports or search. Then move each one to Copilot in SharePoint, a skill or an Agent Builder agent.

If you’re a site owner or a maker, the rule of thumb from the webinar still applies: use an agent when people need a subject matter expert that answers questions, and use a skill when you have a repeatable process you want Copilot to run. Licensing can decide it too. Skills run in Copilot in SharePoint, so everyone using one needs a Copilot license, while an agent also works in Teams and the Copilot app, including for people on pay-as-you-go:

Start with the classic agents you already have, because those are the ones with a deadline.

Ready to tackle the admin side of Copilot in SharePoint? This webinar was about using it, and on November 12, Microsoft MVP Drew Madelung and I are covering governance: who can create skills and agents, where they live, what they can reach, and which control governs each one. Save your seat for Governing SharePoint Skills and Agents.

How Microsoft Copilot Accesses Your Work Data and the Web

Vlad Catrinescu — Fri, 09 Oct 2026 11:59:45 GMT

Microsoft Copilot (formerly Microsoft 365 Copilot) sees the same data as your users, but how does it actually access and use it? Copilot doesn’t directly access your data, and it doesn’t just open up SharePoint and start reading files. The process involves several different systems and safety checks, and as an IT professional, understanding how it works gives you the mental model behind the protection and governance tools you’ll use with Copilot.

Key takeaways

Copilot doesn’t read your files directly. It grounds your prompt with Microsoft Graph and Work IQ before anything reaches the large language model.
Copilot checks that the user can access the content, respects Restricted Content Discovery and Restricted SharePoint Search, and applies your Purview labels and data loss prevention policies.
The prompt you type isn’t the prompt the large language model receives, because Copilot modifies it along the way.
When Copilot needs the web, it sends Bing a short, anonymized query, not your full prompt, your files or your identity.
Your work data stays inside your Microsoft 365 service boundary, and admins can turn web search off.

Why Copilot data access matters for IT pros

We already know that Microsoft Copilot can access your data. If you manage a Microsoft 365 tenant, you also need the technical details of how it gets to that data, especially your work data.

Microsoft also uses a lot of different product names here, some more technical and some more marketing, especially in the Copilot ecosystem. You’ll often see Microsoft Graph, Work IQ and even more advanced terms like the semantic index. I’m going to focus on what’s in scope for the AB-900 exam, so let’s start with two definitions.

Microsoft Graph and Work IQ explained

Microsoft Graph is the index of your tenant. Everything Microsoft 365 already knows about your users’ files, mail, chats, meetings and calendars sits in or behind Microsoft Graph. It’s really the catalog, and it has been around for a while: the Microsoft Graph API went generally available in November 2015, and most documentation today still references Microsoft Graph when it talks about how Copilot accesses data. If you’re a developer, you probably use the Microsoft Graph API every single day.

Work IQ is the layer that takes Copilot to the next level. It doesn’t only have access to the data, it also has memory and inference, so Copilot understands what matters to you. That means who you collaborate with the most, the topics you focus on day to day, the projects you actually spend time in, and the files you really open compared to the ones you needed on a previous project that aren’t relevant anymore.

Work IQ is much newer. It launched at Microsoft Ignite in November 2025, around the same time as the AB-900 exam. Think of Work IQ as the personalized layer on top of Microsoft Graph. They work together to give Copilot its power.

How Copilot accesses your work data

It all starts with the user, who writes a prompt in Microsoft Copilot. Copilot acts like an orchestrator, and the first thing it needs to do is ground that prompt, which means finding the information from your organization that helps answer it. Here it uses both Microsoft Graph and Work IQ, and the goal is to select the most relevant documents before anything is sent to the large language model.

This is important, because it’s part of the value Microsoft adds. The relevant documents, emails and information are selected by the orchestrator before they ever reach the large language model.

The safety checks before the large language model

Microsoft also checks several things about those documents before they get to the model:

Does the user have access to those documents?
Are there restrictions, like Restricted Content Discovery or Restricted SharePoint Search, that say this content shouldn’t show up in Copilot?
Are there Microsoft Purview sensitivity labels or data loss prevention policies that change how Copilot can handle the data it found?

After Copilot does all of that, it combines the information and sends it to the large language model as part of your prompt.

Your prompt isn’t the prompt the model receives

The prompt the user types isn’t the same prompt that gets sent to the large language model. Copilot modifies that prompt a lot before it gets there, including for responsible AI. That’s also why you sometimes get different results between ChatGPT and Microsoft Copilot, even on general knowledge questions, when they use the same model in the back.

The answer then goes back to the user. Since Copilot checks things like Purview, if content was generated from data with a sensitivity label, the content Copilot generates keeps that label. Tools like Microsoft Purview and Microsoft Defender monitor everything, so every interaction is logged and any suspicious prompts are escalated to the admin team. And everything that has to do with your work data remains inside your Microsoft 365 service boundary and respects your service agreement.

How Copilot accesses web data

Web data works differently. This is when Copilot needs to go to the public web to get information that improves the answer to your prompt, or when you ask it to research the web.

Everything starts the same way. The user sends Microsoft Copilot a prompt, the orchestrator decides it needs data from the web, and Copilot creates a short, anonymized query stripped of your identity, with no file information. It sends that query to Bing, which gets the information from the public internet. That part is outside your Microsoft 365 service boundary, so when it comes to web data, you’re leaving that trusted boundary before going back to the large language model.

Even when the query leaves your boundary, there are still safety guardrails:

Admins can turn web search off. Because data leaves the tenant, even anonymized, some organizations prefer to disable it.
Users can switch between work and web. In work mode, Copilot can use their work data as well as the web. In web mode, it uses the web and the context they give it, but not their work data.
The query is anonymized. No identity and no files leave your tenant.
A safety filter runs before the answer gets back to the user. The answer also comes with citations, so the user can verify where the information came from.

So yes, some information does leave your service boundary, but it’s only a short query generated by Copilot. It’s not the full prompt, it’s not any file, and the user’s identity stays safe inside the tenant.

Four things to remember for the AB-900 exam

There are four things I want you to remember for the AB-900 exam about how Copilot keeps your information safe:

Copilot reads only what the requesting user can read. No backdoors, no elevated service account, no admin shortcut. It will never surface data the user doesn’t have access to.
Copilot always respects your compliance settings, such as data loss prevention policies, sensitivity labels and data residency.
Your work data stays in your Microsoft 365 boundary, including your documents and identities.
Microsoft Purview and Microsoft Defender XDR make sure everything is logged, and anything suspicious is escalated according to the policies you have in place.

Microsoft sometimes groups those into four principles: data minimization, transparency, content filtering and auditability. My AB-900 study guide maps the rest of the skills measured.

What this means for you

So the work answers follow your permissions, and the web answers? Well, that’s a switch. Both are things you control as an admin, which is why this mental model matters before you roll Copilot out.

On the work side, Copilot only shows people what they can already access, so the protection and governance tools you already have, from permissions to Restricted Content Discovery to Purview labels and data loss prevention, are the tools that shape what Copilot can use. On the web side, decide whether your organization is comfortable with an anonymized query leaving the service boundary. The control is the Allow web search in Copilot policy, which Microsoft documents in its data, privacy and security guide for web search in Copilot.

If you’re preparing for the exam, those four points are the ones to know cold. I also shared what to expect on the AB-900 exam after I took it.

Ready to get AB-900 certified? This lesson is part of my AB-900 certification path on Pluralsight, which covers the core Microsoft 365 services, data protection and governance, and the basics of Copilot and agent administration. Later in the path, I show you exactly where to turn web search on or off in the Microsoft 365 admin center. Start the AB-900 path on Pluralsight.

AB-650 Exam Review: Where Your Study Hours Should Really Go

Vlad Catrinescu — Tue, 06 Oct 2026 02:57:07 GMT

I took the AB-650 exam while it was still in beta. If I had planned my study time around the exam’s title, I would have studied the wrong things, and that’s the main thing I want you to know before you book it. Of course I can’t share any questions because of the NDA which we should all respect, but in this blog post, I want to share with you where I’d put your study hours and what to expect on exam day.

Key takeaways

The biggest part of the AB-650 is Govern and secure Microsoft 365 tenants and workloads, at 40-45% of the exam, and it’s mostly Microsoft Entra and Microsoft Purview.
For Purview, you need to know how to configure things like DLP policies and sensitivity labels, not just what they do.
Settings for Microsoft Copilot (formerly Microsoft 365 Copilot), Microsoft Agent 365 and SharePoint Advanced Management change fast, so click through them yourself before the exam.
Expect PowerShell, including knowing which older modules Microsoft doesn’t recommend anymore.
If this is your first Microsoft exam, start with the AB-900 and the SC-900.
If you were planning on the MS-102, it retires on November 30, 2026, and Microsoft recommends the AB-650 instead.

What the AB-650 exam really tests

The exam is called Administering Microsoft 365 and AI Services, and passing it gets you the Microsoft 365 and AI Services Administrator Associate certification. Microsoft first announced it as the AI Services Administrator Associate, so you’ll see both names out there. If you read either name and go straight to booking, you’ll walk in expecting a Microsoft 365 administration exam with some Copilot on top.

That’s not really what this is. My exam had a lot of Microsoft Entra and a lot of Microsoft Purview, and not that much Microsoft 365 directly. Everything was still about things that affect Microsoft 365, so it makes sense once you’re sitting there. It’s just not where I would have put my study hours based on the title alone.

And I have to be fair to Microsoft, because the study guide does tell you this. Look at how the exam is split, and Govern and secure Microsoft 365 tenants and workloads is worth 40-45%. That’s the biggest area by far, almost half of the exam, and it’s almost entirely Entra and Purview, with Microsoft Defender for Office 365 in there as well.

Skill area	Weight	What’s in it
Configure and manage Microsoft 365 tenants and workloads	20-25%	Tenant settings, domains, licenses, Microsoft 365 Backup, mailboxes, teams and SharePoint sites
Govern and secure Microsoft 365 tenants and workloads	40-45%	Microsoft Entra identities and authentication, Microsoft Defender for Office 365, Microsoft Purview
Manage and secure AI services in Microsoft 365	35-40%	Copilot settings, agents and Microsoft Agent 365, AI usage and cost monitoring

Even the AI part brings Entra back in, since agents now get their own identities through Microsoft Entra Agent ID. You can see the full breakdown in the AB-650 study guide on Microsoft Learn.

So the information is right there. It’s the exam title and the AI Services Administrator name on the certification that pull your attention toward Copilot and agents, when about half of the exam is identity and data protection.

My advice is the same one I’ve been giving IT pros for years, for every exam. Don’t just look at the title, and don’t just look at the marketing around it. Look at the study guide and at how Microsoft splits the exam, because the two don’t always tell the same story. Study everything, but if you’re going to over-invest anywhere, over-invest in Microsoft Entra and Microsoft Purview.

What I would brush up on

Let’s get a bit more practical. I’m calling out these areas because a lot of Microsoft 365 admins don’t use them day to day.

Microsoft Entra ID

Start with Microsoft Entra ID, and that includes authentication methods and Microsoft Entra Password Protection. Don’t forget that most organizations still have hybrid environments, so you need to know how to configure and manage those too, not just cloud-only scenarios. Password Protection is a good one to practice for that, since you can extend its banned password lists to your on-premises Active Directory with agents on your domain controllers.

Microsoft Defender for Office 365

Next is Microsoft Defender for Office 365. Go play with all the different services inside it, and not just on their own, but also how they integrate with Microsoft Purview.

Microsoft Purview

Speaking of Purview, this isn’t the SC-900, where you just need to know what each product does at a high level. You need to know how to actually configure it: data loss prevention (DLP) policies, sensitivity labels, sensitive information types. Reading about them isn’t enough, so get into the portal and build a few of each.

Copilot and Agent 365 in the Microsoft 365 admin center

This is a big one for the AI part of the exam. You need to know where everything is in the Microsoft 365 admin center for managing Copilot, and for Microsoft Agent 365.

I know what you might be thinking, and I agree: that whole area is brand new and moving fast. That’s how Microsoft ships things today, so remembering a blog post you read three months ago might not be enough. Go click through the admin center yourself before the exam. Microsoft’s study guide even says the exam can include preview features when they’re commonly used.

SharePoint Advanced Management

And don’t forget SharePoint Advanced Management. This is another one where you need to look at the latest, because it moves fast. Restricted Content Discovery, or RCD for short, changed behavior recently. Microsoft announced the change in message center post MC1427972, and the rollout started in late July 2026.

With the new behavior, files you recently accessed on a site with RCD turned on no longer show up for you in Copilot or Microsoft 365 search. Before, if you had recently viewed, edited or shared a file, you could still find it. The update also removes the Copilot button, the AI menus and Create pages with AI from those sites. Some older documentation still describes the old behavior, so make sure you know what changed. I also go deeper on RCD in my post on SharePoint agent governance.

Learn where things live across the admin centers

A broader tip for the entire exam is to learn where things live across the different admin centers. For any given task, know all the places you could configure it, because some settings exist in more than one admin center. Creating a user is the easy example, since you can do it in the Microsoft 365 admin center or in the Microsoft Entra admin center. At associate level, knowing which admin center is best, or which one gives you the option Microsoft is asking about, is an important distinction.

How much PowerShell is on the AB-650?

Because this is an associate exam, you also need to know PowerShell. The PowerShell questions I got weren’t that difficult, but I also spend way more time in PowerShell than the average IT pro, so that might be why they felt easy to me.

At a high level, know which modules you have, how to connect to each service, and what a few commands look like for each module. If you want a quick refresher, these are the main ones:

Service	Module	Connect with
Microsoft Entra ID and Microsoft Graph	Microsoft Graph PowerShell	`Connect-MgGraph`
Exchange Online	ExchangeOnlineManagement	`Connect-ExchangeOnline`
Microsoft Purview (Security & Compliance PowerShell)	ExchangeOnlineManagement	`Connect-IPPSSession`
SharePoint	Microsoft.Online.SharePoint.PowerShell	`Connect-SPOService`
Microsoft Teams	MicrosoftTeams	`Connect-MicrosoftTeams`

You also need to know the past modules. You might get questions where more than one option would work, but some of them use modules that Microsoft doesn’t recommend anymore, like the Azure AD and MSOnline modules that Microsoft Graph PowerShell replaced. So know which modules to use today.

Where to start if this is your first Microsoft exam

If this is your first Microsoft certification exam, or you’re just starting with Microsoft 365 and Copilot administration, I would start with the AB-900 (Microsoft 365 Copilot and Agent Administration Fundamentals) and the SC-900 (Microsoft Security, Compliance, and Identity Fundamentals). They aren’t required for the AB-650, but they give you a lot of the foundation on Microsoft 365, Copilot, Purview and Defender, and you can build up to the associate level from there. My AB-900 study guide and SC-900 study guide have free resources for both.

What to expect on exam day

I can’t share the number of questions on the exam, but Microsoft says most of its exams have between 40 and 60, and you get 100 minutes for an associate exam, or 120 if it may contain labs. For an associate-level exam like this one, expect the more advanced question types too: scenarios, questions where you put steps in the right order (Microsoft calls those build list questions), and pretty much every other question type Microsoft uses. You can try them all in Microsoft’s free exam sandbox.

Is the AB-650 a fair exam?

If I had to give an opinion, I think it’s a really fair exam, and a really fair test of how the job of a Microsoft 365 administrator has changed in the era of AI. For a long time, a lot of us mostly needed to know how to manage SharePoint. Now we need to understand identity better, because identity is the first way people get hacked. We need to understand data protection and Purview. And now we’re governing agents that have their own identities, all on top of the workload administration we already do.

So I think it points people in the right direction, with the right knowledge. Just go in with the right expectations about where the weight actually sits. If you’re a SharePoint, Teams or Exchange admin and you haven’t really touched Purview or Entra ID, learn those before you sit the exam.

What this means for you

If you’re a Microsoft 365 admin, this is the certification that lines up with what the job looks like now. And if you were planning to take the MS-102, keep the date in mind: Microsoft retires it on November 30, 2026, and recommends the AB-650 certification instead.

To get ready, I would go in this order:

Read the AB-650 study guide and check yourself against the Govern and secure section first, since it’s 40-45% of the exam.
In a test tenant, build a few DLP policies, sensitivity labels and sensitive information types, and spend time in Defender for Office 365.
Click through the Copilot and Agent 365 settings in the Microsoft 365 admin center, and read up on the new RCD behavior.
Practice connecting to each service with the current PowerShell modules.
If this is your first Microsoft exam, take the AB-900 and the SC-900 first.

Ready to start studying for the AB-650? My AB-650 study guide covers every skill measured on the exam, with free study notes and resources. And if you want to go deeper on Microsoft 365, Copilot and Purview, check out my 100+ courses on Pluralsight.

SharePoint Skills FAQ: Your Webinar Questions Answered

Vlad Catrinescu — Fri, 25 Sep 2026 04:07:27 GMT

The Teams gremlins held my camera hostage for part of my Getting Started with SharePoint Skills webinar, but the questions kept coming, and I ran out of time before I ran out of questions. This SharePoint skills FAQ answers every question from the Q&A panel and the chat, including the ones I answered live, double-checked against Microsoft Learn and Microsoft’s latest updates. If you missed the live session, you can watch the full SharePoint skills webinar recording on my YouTube channel.

Key takeaways

A skill is not a saved prompt. It’s a shared, multi-step workflow that everyone on the site can trigger in plain English, and Copilot in SharePoint doesn’t have saved prompts anyway.
Skills run when someone asks for them in the Copilot in SharePoint chat. They can’t run on a schedule, on a file event, or from Power Automate today.
Personal skills live in your OneDrive and are for you only. To share a skill with your team, create it as a site skill on the team’s SharePoint site.
You protect skill files like any other SharePoint content: break inheritance on the Agent Assets library, and people with View access can still run the skills.
There’s no official application lifecycle management (ALM) for skills yet, so keep a copy of business-critical skills in a central repository and use PnP PowerShell or Microsoft Graph PowerShell for inventory.
A SharePoint agent answers questions about content. Copilot in SharePoint with skills takes action on that content.
Microsoft just announced that Copilot in SharePoint has hit General Availability and also clarified the billing model!

Skills vs Saved Prompts and Other Tools

Why Is a Skill Different From a Saved Prompt?

Microsoft 365 Copilot does let you save prompts in the Copilot Prompt Gallery from Copilot Chat, Teams, and Outlook, but Copilot in SharePoint doesn’t have saved prompts, so on a SharePoint site a skill is the way to package a request people can reuse. There is some overlap in the idea, since both save you from retyping the same instructions. The difference is everything a skill adds on top.

A skill lives on the site, so everyone with View access can run it. Nobody has to find the prompt in a Notepad file or a OneNote page. People just ask in natural language, and Copilot in SharePoint matches the request to the skill through its description. A skill can also chain several steps, load reference files, and call other skills, so it handles long-running tasks a single prompt can’t.

The biggest difference is consistency. A saved prompt behaves differently depending on who runs it, because each person’s memories and customizations shape the answer. A skill documents the process once, so everyone gets the same steps and the same output. Microsoft’s skills documentation says skills let Copilot in SharePoint perform tasks “more consistently than one-off prompts.”

When Should I Use a Skill Instead of a Power Automate Flow, a Copilot Studio Agent, or a SharePoint Agent?

Use a skill when the work stays inside SharePoint and benefits from judgment: reading documents, classifying them, checking them against a rule, and updating lists or libraries based on what it finds. Skills are no-code, multi-step, and use AI to understand intent, but today they can’t connect to external systems or run custom code. That first limit is about to loosen: Copilot in SharePoint is getting support for plugins from your enterprise store, and a plugin can bundle skills and Model Context Protocol (MCP) servers that connect to other systems. It’s in preview and scheduled to roll out worldwide in December 2026, according to the Microsoft 365 roadmap.

The other three options each fit a different job:

A Power Automate flow is deterministic. It follows the exact same path every time, connects to third-party systems through connectors, and handles formal processes like approvals through Outlook. When the process must never vary, or it has to reach outside Microsoft 365, use a flow.
A SharePoint agent answers questions. It’s grounded in your SharePoint content, but it doesn’t take action on that content.
A Copilot Studio agent is for scenarios beyond SharePoint. Custom engine agents built in Copilot Studio can reach other systems and even trigger actions automatically, without a user asking. They’re also more work to build, deploy, and maintain.

This is exactly the question my October 20 webinar goes deeper on, with scenarios for each option. The registration link is at the end of this post.

Running Skills

Can Skills Run on a Schedule, When a File Is Added, or From Power Automate?

Not today. A skill runs when someone asks for it in the Copilot in SharePoint chat, either with a matching request or by naming the skill. You can’t trigger a skill from a file event, run it on a schedule, or call it from a Power Automate flow.

If what you need is a simple reaction to a file event, SharePoint’s built-in workflows for lists and libraries now cover some of it. In its September 2026 update, Microsoft added workflows that post a Teams message when files are added or metadata changes, and workflows that route approval requests. Those are separate from skills, but they might cover the event-driven part of your scenario.

Is There a Limit to How Many Documents a Skill Can Classify, and Can It Handle 10,000 Files?

I haven’t found a published limit on how many files a skill can process, and I haven’t tested a 10,000-file library myself. I’m checking with the product team and will update this post when I hear back.

What is documented is the usage limit. During the preview, according to the Copilot in SharePoint get-started guide, Copilot in SharePoint applies daily and weekly usage limits to each user. The limits are individual, not shared across your organization, and when you hit one, Copilot features pause until the limit resets automatically. A classification run across thousands of files does a lot of work, so plan for it:

Test the skill on a few selected files first, then on a single folder.
Run large libraries in batches, folder by folder, instead of in one request.
For files that keep arriving, look at autofill columns in SharePoint, which extract metadata as files are uploaded, so you only run the skill once on the backlog.

In the blog post from September 25th, Microsoft has announced the ability to work with Copilot in SharePoint to operate across large content collections, but I do not know yet what large means!

Can a Skill Create Site Pages From a Template and Add Quick Links Web Parts?

Creating pages, yes. Copilot in SharePoint can create pages, and a skill chains Copilot in SharePoint’s built-in capabilities, so page creation can be one of the steps. Whether it can also add and configure a Quick Links web part that points to related pages depends on the page tools available in your tenant.

The quickest way to check is to type --agenttools in the Copilot in SharePoint chat, which lists every action Copilot in SharePoint can run on that site, and look through the page-related tools. The list of tools differs from tenant to tenant, so check yours rather than relying on someone else’s screenshot. Then test the skill on one page before running it across a site.

A skill you create in OneDrive is a personal skill, and personal skills are designed to be yours alone. That’s why you got the message that the skill doesn’t support sharing with everyone who can access the folder. So yes, it comes down to OneDrive instead of SharePoint, though the real distinction is personal versus site scope.

Personal skills are saved as Markdown files in your own OneDrive, under Documents, Copilot, Agent Assets, Skills, and they follow you across every SharePoint site and OneDrive. They’re perfect for your own way of working, like how you want long documents summarized. They rolled out in preview in August 2026 and roll out worldwide in December 2026. Microsoft announced them in its September update, but the Microsoft Learn skills article only covers site skills so far.

To share a skill with your teammates, create it as a site skill instead. Open the Copilot in SharePoint chat on your team’s SharePoint site and create the skill there. It’s saved to that site’s Agent Assets library, and everyone with View access to the site can run it.

How Do I Protect the Skill Files in the Agent Assets Library?

You protect them the same way you protect any other SharePoint content. By default, anyone with Edit permission on the site can create a skill, and anyone with View permission can run one, which means every member of a team site can change the skill files. If you want a tighter model, Microsoft’s documentation points you to breaking permission inheritance on the Agent Assets library and applying more restrictive permissions.

In practice, that means giving members Read on the Agent Assets library and keeping Edit for the owners and the few people who author skills. Members can still run every skill, because running a skill only needs View access. Retention, sensitivity labels, and auditing apply to skill files too, and the library keeps version history, so you can roll a skill back if someone breaks it.

You already spotted the trade-off: once members can’t edit the library, they can’t create skills on that site anymore. The same library holds your SHAREPOINT.md file, so locking it down protects your site context too. The library also holds a Plans folder, so after you change the permissions, sign in as a member and run a skill to confirm everything still works the way you expect.

Governance and Lifecycle Management

Can I Maintain SKILL.md Files With Azure Pipelines in Azure DevOps?

Not officially, and I’ll be honest, this is one of the more disappointing parts of the preview. Skills are managed at the site level, and Microsoft doesn’t offer a supported way to deploy or update them from a pipeline.

In the meantime, a few things help:

Version history. Skills are files in a document library, so every change is versioned and you can restore an earlier version.
Built-in skills that improve your skills. Microsoft ships built-in skills to create, check, and improve skills. In September 2026 they added measured evaluations, where Copilot suggests tests, evaluates the current results, and recommends changes before you share a skill more broadly.
A preview of what’s coming. When you type /skills, the list includes a section for enterprise skills. It was empty in my tenant during the webinar, but it shows where Microsoft is heading for centrally managed skills.
Plugins from your enterprise store. A plugin can bundle skills, and an admin can install it tenant-wide, so once plugin support rolls out you’ll have a governed way to distribute the same skills across sites.

Until then, my recommendation is simple: when a skill becomes business-critical, keep a copy in a central repository you control. That gives you a source of truth even without an official process.

Until Enterprise Skills Arrive, Is PnP PowerShell the Way to Govern Skills?

Yes. Since skills are Markdown files stored in each site’s Agent Assets library, PnP PowerShell works well for building an inventory: loop through your sites, look in the Skills folder of the Agent Assets library, and each subfolder with a SKILL.md file is one skill. Microsoft Graph PowerShell works too. Auditing also applies to skill files, the same as other SharePoint content.

I can’t wait for Microsoft to give us proper governance for skills. Governance for SharePoint skills and agents would make a great future webinar topic too.

SharePoint Agents vs Microsoft 365 Copilot Agents

Every SharePoint site comes with a ready-made agent that answers questions based on that site’s content. People with Edit permission can also create custom agents, which are saved as .agent files in the site’s Site Assets library, scoped to the sites, folders, or files you choose, and tailored with your own instructions.

Custom agents don’t automatically appear in any list, so you share them like any other file. People can open the .agent file, and you can also share a SharePoint agent in Teams group chats and meetings. Site owners also decide what opens when people select the agent icon, either Copilot in SharePoint or a custom agent, and access always follows the site’s permissions.

Keep in mind that SharePoint agents answer questions, while Copilot in SharePoint is the experience that runs skills and takes action. My step-by-step guide to creating SharePoint agents walks through building one.

Is a SharePoint Agent the Same as an Agent Created in Microsoft 365 Copilot?

No. A SharePoint agent lives in its SharePoint site and only uses SharePoint content. An agent you build with Agent Builder in Microsoft 365 Copilot can use knowledge from SharePoint, Teams, Outlook, public websites, and Microsoft 365 Copilot connectors, and you can share it across the organization or submit it to your organization’s catalog.

Where you build it depends on where the content lives and what the agent needs to do:

Build it in SharePoint when the content lives on one site and the audience is that site’s members.
Build it with Agent Builder when the agent needs content from several places, like Teams and Outlook as well as SharePoint. My guide to building Copilot agents with no code in Agent Builder covers the basics.
Build it in Copilot Studio when the agent needs to take actions in other systems or run without a user asking.

Agent Builder agents can do a lot more, but they’re also a bit more work to maintain and deploy.

What This Means for You

Most of these questions come down to the same thing: skills are still a preview feature, and the parts around them, like scheduling, lifecycle management, and governance, haven’t caught up yet. That doesn’t make them less useful today. It means you should be deliberate about where you use them.

If you’re an admin, decide who can edit the Agent Assets library on your important sites before skills spread, and start keeping business-critical skills in a central repository now. If you’re a builder, pick one repetitive task that stays inside SharePoint, build it as a site skill so your whole team can use it, and test it on a small set of files before pointing it at a large library.

Ready to take SharePoint skills further? On October 20, Microsoft MVP Gokan Ozcifci and I are running a free webinar on automating content in SharePoint with Copilot, covering more advanced skills, SharePoint agents, and autofill, and when to reach for each one. Save your seat for the October 20 webinar.

AB-620 Exam: What to Expect, Difficulty, and Study Tips

Vlad Catrinescu — Wed, 19 Aug 2026 04:59:33 GMT

The AB-620 exam is the path to the brand-new Microsoft Certified: AI Agent Builder Associate certification, and after taking the beta a few hours ago, I want to share my honest first-look review. If you are deciding whether to sit for this exam, what to study, or whether you are even the right audience for it, this post covers everything I can share without breaking NDA.

Key takeaways

Passing the AB-620 exam earns you the Microsoft Certified: AI Agent Builder Associate certification.
Despite being tagged as an app maker exam on Microsoft Learn, the AB-620 is really built for developers and consultants integrating Copilot Studio with enterprise systems.
Microsoft scrapped the original AB-610 Copilot Studio Agent Maker exam and consolidated everything into the AB-620.
Expect heavy coverage of Azure AI Search, Microsoft Foundry, Microsoft Fabric, the agent-to-agent protocol, and Power Platform pipelines for ALM.
Testing agents and application lifecycle management makes up 20 to 25 percent of the exam, which is enough to make or break a pass.
If you are a Power Platform app maker, the Copilot Studio Applied Skills are likely a better fit than this exam.

What is the AB-620 exam?

The AB-620 is a Microsoft associate-level certification exam currently in beta. Pass it and you earn the Microsoft Certified: AI Agent Builder Associate credential. The official exam name is Designing and Building Integrated AI Solutions in Copilot Studio.

The exam tagline tells you most of what you need to know: Develop and integrate advanced, scalable AI agents using Microsoft Copilot Studio, Power Platform, and enterprise technologies to deliver robust solutions for organizations. In plain language, the AB-620 exam is about putting agents into production, not about the basics of agent building.

Why the AB-620 is really a Copilot Studio developer exam

Here is the most important thing to understand before you register: even though the exam is tagged as an app maker exam on Microsoft Learn, in my opinion that tagging is misleading. The AB-620 is built for developers and consultants who spend almost every workday building agentic solutions that integrate Copilot Studio with Microsoft Foundry, Azure AI Search, Microsoft Fabric, and more.

If you are a citizen developer who occasionally builds a topic in Copilot Studio, this exam will feel out of reach. If you live in Copilot Studio and routinely wire it up to enterprise services, you are exactly the audience Microsoft built the AB-620 for.

Having spent the last decade helping IT pros prepare for Microsoft certifications, I have seen plenty of exam mismatches between the listed persona and the actual content. The AB-620 is one of the more notable examples I have come across recently.

What happened to the AB-610?

A few months ago, Microsoft announced two Copilot Studio certifications: the AB-610 Copilot Studio Agent Maker and the AB-620 Copilot Studio Agent Developer. That two-track approach made sense. The Agent Maker exam would have covered the low-code persona, and the Agent Developer exam would have covered the integration-heavy persona.

Microsoft has since scrapped the AB-610 and consolidated everything into the AB-620. Instead of keeping the original Copilot Studio Agent Developer name, which would have been honest signaling, Microsoft renamed the exam to AI Agent Builder Associate. That broader name suggests both makers and developers belong here, but the content still skews heavily toward the developer side.

For the full backstory on these two exams, see my earlier post on the original AB-610 vs AB-620 Copilot Studio certifications.

What the AB-620 exam actually tests

The meat of the AB-620 exam is about production-grade agents and the advanced scenarios an enterprise will ask about before it ships an agent to production. Do not expect questions about turning generative AI on, what a topic is, or who should be authoring agents. That foundational content is not what this exam is measuring.

What you will see is heavy coverage of how to connect a Copilot Studio agent to enterprise services. Specifically, expect to be tested on:

Connecting a Copilot Studio agent to Azure AI Search
Connecting a Copilot Studio agent to a Microsoft Fabric data agent
Connecting a Copilot Studio agent to a Microsoft Foundry agent
Connecting to partner agents outside your tenant using the agent-to-agent protocol
Designing multi-agent solutions that span those systems

Real production agents today are rarely built entirely inside Copilot Studio. The AB-620 reflects that reality by testing your ability to design solutions that span multiple Microsoft AI services.

Testing agents and ALM with Power Platform pipelines

A second major area of the AB-620 exam covers testing agents and application lifecycle management. This bucket is marked at 20 to 25 percent of the exam, which sounds modest on paper but is more than enough to be the difference between pass and fail.

You need to know all the ways to test Copilot Studio agents and how to implement application lifecycle management for agents using Power Platform pipelines. The exam expects you to be able to take an agent from a development environment, through any required validation, and all the way to production while following a proper ALM process.

The takeaway is that the AB-620 does not assume you are only a builder. It assumes you own the full lifecycle of an agent, from build to release.

How to study for the AB-620 exam

The single best piece of advice I can give is to do every scenario in the exam objectives at least once before sitting down. Go to Microsoft Learn, pull up the AB-620 study guide, and work through the tasks hands on:

Connect a Copilot Studio agent to Azure AI Search.
Connect a Copilot Studio agent to a Microsoft Foundry agent.
Connect a Copilot Studio agent to another Copilot Studio agent.
Experiment with the agent-to-agent protocol so you know what it is and why it exists.
Build a deployment flow for an agent using Power Platform pipelines.

Why does hands-on practice matter so much? Because every associate-level Microsoft exam, including this one, leans heavily on order-list questions. You will be asked, “you want to do this, put the following steps in the correct order.” If you have not configured the scenario recently, those questions are punishing. If you have walked through it once, they become almost free.

The other point worth flagging is the question weighting. Microsoft did honor the study guide in the sense that nothing showed up in my beta that was not on the published exam objectives. However, the percentage of questions from each bucket felt off compared to the weights listed on the study guide. I cannot share which buckets were over-represented or under-represented due to NDA, but my recommendation is the same as it always is for a beta: study every skill area equally and do not lean on the weights as a planning tool.

Exam format, questions, and Microsoft Learn access

A few practical notes about the exam itself:

The number of questions is always between 40 and 60. I cannot share the exact count for the AB-620.
You have access to Microsoft Learn during the exam, just like every other associate-level exam. I did not end up using it, but it is there if you need to verify something.
Expect associate-level question types: scenario-based questions, order lists, and multi-step problems. This is not the difficulty level of a fundamentals exam.
Labs may show up. Microsoft does not publish whether a given exam includes hands-on labs, and I cannot share what was on my exam, but be prepared for the possibility.

AB-620 vs Copilot Studio Applied Skills

If you are a Power Platform app maker who does not spend all day in Copilot Studio, the AB-620 is probably too advanced for where you are right now. The better path is the Copilot Studio Applied Skills. There are two of them available, and they validate hands-on Copilot Studio skills without the integration-heavy enterprise scope of the AB-620.

If you want a deeper look at the lab-based credential for Copilot Studio, see my APL-7008 study guide for creating agents in Microsoft Copilot Studio.

The AB-620 is still a great goal if your job is to become a full-time Copilot Studio agent builder, or if you are already a developer integrating AI services across the Microsoft stack. For everyone else, the Applied Skills path will give you a recognized credential without the steep developer-focused prep.

What this means for you

The AB-620 is a high-value certification for a narrow audience: developers, consultants, and architects building production-grade agents that span Copilot Studio, Azure AI Search, Microsoft Foundry, Microsoft Fabric, and Power Platform. If that describes your day-to-day work, this exam validates the skills you are already using and is well worth the prep time.

If you are a citizen developer, business user, or IT pro who builds simple Copilot Studio agents on the side, this exam is not the right next step. Start with the Copilot Studio Applied Skills instead. They give you a recognized credential without asking you to master enterprise integration patterns you may never use.

Before you register, pull up the AB-620 exam objectives on Microsoft Learn and do an honest gap analysis. If you can do each task on the list without looking it up, you are ready. If half the objectives are unfamiliar, build hands-on time with the missing scenarios before scheduling the exam. Beta exams take longer to score, so plan accordingly if you are aiming for a specific certification deadline.

Ready to build your Copilot Studio skills the right way? I have over 100 courses on Pluralsight covering Microsoft 365, Power Platform, and Copilot, including hands-on agent building. Browse the full course catalog on Pluralsight to find your next step.

Copilot Studio Applied Skill APL-7008 Is Retiring June 30

Vlad Catrinescu — Thu, 25 Jun 2026 14:01:16 GMT

Microsoft is retiring the Create Agents in Microsoft Copilot Studio Applied Skill, known by its code APL-7008, on June 30, 2026. If you have been meaning to add this Copilot Studio Applied Skill to your transcript, you have until the end of June to earn it for free, and after that it is gone. Below is what the credential covered, what the retirement actually means for your transcript, and which Copilot Studio credentials you can still earn once it is gone.

Key takeaways

APL-7008, the Create Agents in Microsoft Copilot Studio Applied Skill, retires on June 30, 2026.
It is free, it is not proctored, and it runs as an interactive lab, so you can earn it in a single sitting.
Applied Skills never expire. Once you pass, the credential stays on your Microsoft transcript permanently.
After June 30, the Enhance Agents with Autonomous Capabilities Applied Skill and the AB-620 AI Agent Builder Associate certification are your remaining Copilot Studio options.
This is one piece of a much larger shake-up. We are not even halfway through 2026, and Microsoft has already retired or replaced 14 certifications and 13 Applied Skills.

What the Copilot Studio Applied Skill (APL-7008) covered

APL-7008 validated your ability to build a custom agent from start to finish in Copilot Studio. It was a well-rounded credential that touched almost every core part of agent building, not just one narrow feature.

The assessment covered creating the initial agent, working with topics and trigger phrases, and configuring the different nodes in a conversation. It also covered managing variables and entities, setting up the generative AI side with knowledge sources and actions, and pulling data from Dataverse to make the agent more useful.

As someone who earned this Applied Skill myself, I can tell you it was a genuinely practical credential. It mapped closely to the work you actually do when you build an agent for a real business scenario.

If you want the full background on the credential, I covered it in detail when this Applied Skill first launched.

Why retiring this credential is a surprise

I will be honest: this retirement caught me off guard. Copilot Studio is exactly where Microsoft is pushing companies right now, so I expected an update to this credential rather than a full retirement.

There is a reason behind it, though. Copilot Studio is going through a massive change as a product. The interface is changing, topics have been deprioritized and tucked away in the new experience, and the platform now leans on skills and more AI capabilities. A refresh would have made sense to me. A complete retirement, with less than two weeks’ notice, still surprised me.

This is not happening in isolation either. The pace of credential changes has been relentless this year, which I broke down in my roundup of every Microsoft certification change in 2026.

What an Applied Skill retirement actually means

Here is where Applied Skills work differently from regular certifications, and it is the part most people miss. Applied Skills do not expire. If you pass one, it shows up as an active credential for two years after the retirement date, and then it moves to your historical credentials.

More importantly, it stays on your Microsoft transcript forever as a credential you have passed. Retirement does not erase it. It simply means no new learners can earn it after the cutoff date.

That is why earning APL-7008 before June 30 is worth doing, even though it is going away. You keep the credential permanently, and right now, you can still take it for free. If you want the bigger picture on how these two credential types differ, I wrote a full comparison of Applied Skills versus certifications.

The Copilot Studio credentials you can still earn

Losing APL-7008 does not leave you without options. Two Copilot Studio credentials are available today, and they cover different audiences.

The first is the Enhance Agents with Autonomous Capabilities Applied Skill. It goes deeper and more niche than APL-7008 did, focusing on the autonomous side of agents, but it is still firmly in the Copilot Studio world, and it is still free.

The second is the AB-620 AI Agent Builder Associate certification. This is a brand new exam aimed at developers and advanced builders creating production-ready agent solutions. I took it myself last month, and I will warn you that it is not an easy one, certainly harder than the title might suggest. I compared it to the maker-focused option in my breakdown of the AB-610 and AB-620 Copilot Studio certifications.

What this means for you

If earning a Copilot Studio credential has been on your list, the next few days matter. APL-7008 is free, unproctored, and you do not need to schedule it. You simply open the assessment on Microsoft Learn and complete the interactive lab, usually in a single sitting. After June 30, 2026, that option disappears.

So the practical move is simple. If you want APL-7008, earn it before the deadline and keep it on your transcript for good. If you miss the window or want to go further, the Enhance Agents with Autonomous Capabilities Applied Skill and the AB-620 certification are both still available.

Whatever you decide, do not assume you can come back to APL-7008 later. Once it retires, there is no way to earn it again.

Ready to earn APL-7008 before it retires? You can take the Create Agents in Microsoft Copilot Studio Applied Skill for free on Microsoft Learn until June 30, 2026. Start the assessment on Microsoft Learn. If you want to sharpen your agent-building skills first, my Pluralsight courses walk through Copilot Studio step by step.

AI Agents vs AI Chatbots: What Makes Them Different

Vlad Catrinescu — Wed, 29 Apr 2026 12:42:52 GMT

AI agents vs AI chatbots is one of the most confusing topics in technology right now. Most people use the terms interchangeably, yet they describe two very different kinds of tools, and the difference matters if you actually want AI to help you get work done. This post explains what AI agents are, the four capabilities that separate them from chatbots like Microsoft Copilot or ChatGPT, and the three main types of agents you should know about.

Key takeaways

AI chatbots respond to your prompts, but they cannot reliably execute tasks across your business systems on their own.
AI agents add four new capabilities on top of a chatbot: role-playing, memory, tool assignment, and execution loops.
There are three main types of AI agents: retrieval agents, task-based agents, and autonomous agents.
Task-based agents are the most common in business because they actually complete work, not just answer questions.
You do not need to be a developer to build an AI agent thanks to a growing range of no-code and low-code tools.

Why AI chatbots hit a wall

For the past few years, the spotlight has been on AI chatbots. The list keeps growing: ChatGPT, Microsoft Copilot, Google Gemini, Anthropic Claude, Grok from X, and Meta AI. Most of us use at least one of them every day, and they are excellent for general questions, drafting, and exploration.

However, chatbots have real limitations once you try to push them past simple Q&A.

The first limitation is action. Chatbots can suggest what you should do or describe a procedure, but they cannot perform tasks inside your systems. They will happily tell you how to submit a vacation request in your HR tool, but they will not actually submit it for you.

The second limitation is memory. Chatbots do not reliably remember your preferences, your previous work, or your ongoing tasks. As a result, you end up repeating context every time you start a new conversation, especially when the relevant information lives in another system.

The third limitation is that they are generic. They do not automatically follow your organization’s processes, terminology, or formatting rules. You have to spell all of that out in the prompt, every single time.

The fourth limitation is execution. A chatbot responds once and stops. It does not plan a multi-step task, decide which tools to use, review its own results, and keep going until the work is finished. That last gap is exactly where AI agents come in.

What is an AI agent?

An AI agent is an AI tool that can understand a goal, use tools, and take action to complete tasks on your behalf. Instead of just talking to you, an agent works alongside you. It interprets your intent, interacts with the systems where the data and processes live, and follows through until the job is done.

Having spent the last decade training IT professionals on Microsoft 365 and authoring 100+ Pluralsight courses, I think this is the single biggest shift since the launch of ChatGPT. We are moving from “AI that answers” to “AI that does.”

The 4 capabilities that set AI agents apart from AI chatbots

There are four specific capabilities that elevate an AI agent above a traditional AI chatbot. Together, they explain why AI agents vs AI chatbots is more than a marketing distinction.

1. Role-playing

Each agent has a defined role. You can give it instructions on how it should behave, what its responsibilities are, and how it should respond. This keeps the agent focused on a specific purpose, whether that is helping with research, customer follow-ups, or an internal HR process.

2. Memory systems

Agents do not respond based on a single prompt in isolation. They can remember past interactions and use that context for future ones. They retain your preferences, ongoing tasks, and previous conversations, which makes responses noticeably more relevant over time.

3. Tool assignment

This is where agents start to actually take action. Instead of only providing information, an agent can use tools and connect with other systems to do the work. That includes retrieving data, creating or updating records, and triggering workflows in line-of-business applications. To go deeper on this, the customizing Microsoft 365 Copilot concepts and terms guide is a great companion read.

4. Execution loops

Traditional chatbots respond once and stop. Agents can plan, use tools, review the results, and keep going until the task is complete. That loop is what enables them to handle real multi-step work end-to-end, not just provide a single answer.

The combination of role, memory, tools, and execution is what moves AI from answering questions to genuinely getting things done.

The 3 types of AI agents you should know

Not every agent is built for the same job. There are three main types, and they represent increasing levels of capability from simple to more advanced.

Retrieval agents

Retrieval agents, sometimes called single-turn assistants, focus on pulling information from connected data sources, summarizing it, and answering your questions. They help you find and understand information faster, but they do not take action on your behalf. Think of them as a smarter, more grounded search experience.

Task-based agents

Task-based agents go further. In addition to retrieving information, they can take action: create records, update systems, run workflows, and complete repeatable tasks. These are the most common types of agents in business today because they actually move work forward, not just answer questions. You might also hear them called tool-using agents or workflow-based agents.

Autonomous agents

Autonomous agents do not always wait for you to ask. They can plan and execute tasks based on events, make decisions using context, and coordinate multiple steps to achieve a goal. This allows them to handle more complex scenarios with less direct input from the user.

You don’t need to be a developer to build an AI agent

A common misconception is that building agents requires deep development skills. That is no longer the case.

At the simplest level, no-code tools let you build agents through visual interfaces without writing a single line of code. If you want to see this in action, my guide to building Copilot agents with no code using Agent Builder walks through the experience step by step. If you need more flexibility, low-code tools let you customize agents with additional configuration and integrations, which is especially useful for task-based agents that connect to existing business systems. And for developers, pro-code options allow fully custom agents to be designed, integrated, and shipped into production scenarios.

Whatever your skill level, there is now a path into building agents. That is a meaningful change for IT pros and citizen developers alike.

What this means for you

If you have been using AI chatbots for the last couple of years and getting decent results, AI agents are the next step, not a replacement. Chatbots are still excellent for general questions, drafting, and exploration. Agents are where AI starts to take real work off your plate, and understanding AI agents vs AI chatbots will only become more important as both technologies mature.

For IT admins, this is a great moment to start mapping which internal processes are good candidates for a task-based agent. Anything that involves repetitive lookups, status updates, ticket triage, or cross-system actions is worth a look. For citizen developers and Power Platform builders, the no-code and low-code agent tooling means you can prototype quickly and prove value before involving a developer.

If you want to formalize your skills, Microsoft is also rolling out new agent-focused credentials. The APL-7008 study guide for creating agents in Microsoft Copilot Studio is a great starting point if you want a structured path to validate what you build.

The honest reality is that the organizations getting the most out of AI in the next 12 months will not be the ones running the most chatbots. They will be the ones who deploy agents that actually do the work.

Ready to start building AI agents in your organization?

If you want a deeper, structured walkthrough with real examples and demos, including a sales RFP agent, an HR support agent, and a custom agent build, the full course is available on Pluralsight. Watch the full AI Agents for Productivity course on Pluralsight to take the next step from understanding agents to actually building them.

Microsoft 365 Conference 2026 Recap: Top Announcements

Vlad Catrinescu — Wed, 29 Apr 2026 02:32:36 GMT

The Microsoft 365 Conference 2026 in Orlando wrapped with two back-to-back keynotes from Jeff Teper and Ryan Cunningham, packed with announcements across SharePoint, Microsoft Teams, Microsoft Copilot, and Agent 365. This recap cuts through the marketing buzz and focuses on what actually matters for IT pros: the SharePoint Skills public preview, Agent 365 going GA on May 1, the new Shadow AI pane, channel agent workflows in Microsoft Teams, and the multi-model story between Microsoft and Anthropic. Here is what was announced and what it means for your day-to-day job.

Key takeaways

SharePoint Skills are now in public preview and included in Microsoft 365 and SharePoint AI licensing. Jeff Teper called this his favorite announcement of the week.
Agent 365 hits general availability on May 1, priced at $15 per user per month standalone or included with E7. A new Shadow AI pane helps IT block unsanctioned AI use through Intune policies.
Microsoft Copilot now offers a true multi-model experience: Claude models, OpenAI models, or Auto. The Researcher Critique feature pairs a GPT class model with a Claude class model for higher quality output.
Channel agent workflows in Microsoft Teams can trigger Azure DevOps, GitHub Copilot, Asana, and Jira automations from a single chat message or emoji reaction.
The new SharePoint UI announced at the SharePoint 25-year party is now generally available, and Viva Connections plus Viva Amplify are being folded back under the SharePoint brand.
Vibe coding for the enterprise launched on vibe.powerapps.com, with NASA as the headline customer running 47,000 monthly active Power Platform users on the same stack.

Microsoft 365 Conference 2026: Two Keynotes, One Story

Jeff Teper, the father of SharePoint, opened the Microsoft 365 Conference 2026 with a keynote covering Microsoft Copilot, Microsoft Teams, and SharePoint. Ryan Cunningham followed immediately with a session on apps, agentic AI, and governance. Because the Power Platform team and the Microsoft 365 team are now one combined organization at Microsoft, both leaders represented the same product family on stage. With no break between the two sessions, the experience felt more like one extended keynote than two separate ones.

What Was New for Microsoft Copilot

From a usage point of view, Jeff Teper shared that over 90 percent of the Fortune 500 has now deployed Microsoft Copilot. That is a significant adoption milestone. From a feature point of view, however, this keynote was less about brand new announcements and more about cementing the Microsoft and Anthropic partnership and showcasing recent investments.

Multi-Model Support With Claude and OpenAI

Microsoft is doubling down on the multi-model approach in Microsoft Copilot. As a user, you can now choose between Claude models, OpenAI models, or let Copilot pick automatically with Auto mode. If the multi-model story is new to you, my guide to customizing Microsoft 365 Copilot concepts and terms walks through the building blocks. The most memorable Copilot demo of the keynote was Dan Holm using Claude inside PowerPoint to take a boring finance slide and turn it into something visually polished. For PowerPoint work, Claude has been particularly strong, and that is a case where I would skip Auto and pick Claude directly. The multi-model story also extends into Copilot Studio, where makers can now build agents on top of Anthropic models like Opus 4.7. This is a meaningful shift for anyone building custom agents because it removes the assumption that maker tools are locked to a single model family.

Researcher and the Critique Agent

The Researcher Critique feature is not technically new. Microsoft blogged about Council and Critique a few weeks before the conference, but seeing it on stage made the value of multiple models in Microsoft Copilot click. Researcher now runs a GPT class model and a Claude class model in tandem. The first model handles the planning, research, and drafting. The second model steps in as an expert reviewer, evaluating and refining the draft before you ever see it. The result is that the first thing you read is already higher quality than a single model pass. Worth flagging for IT admins: in the keynote demo, Researcher applied the right sensitivity label and security policies to its output automatically, so the AI governance story is built in rather than bolted on.

Copilot Cowork Versus Claude Cowork

Microsoft also showed Copilot Cowork, which launched a few weeks before the conference. Copilot Cowork takes the underlying technology behind Anthropic’s Claude Cowork and adds Microsoft’s two enterprise ingredients on top: Work IQ for grounding in your organizational data, and Microsoft’s security and compliance controls. For a deeper look at Work IQ, I covered it in detail with Microsoft’s Adam Harmetz on knowledge agent licensing and Work IQ. Both products share the Cowork name, but they are not the same tool.

Copilot Cowork runs in the Microsoft cloud, so you don’t need a dedicated mini PC sitting in a closet to run overnight tasks. The trade-off is that it can reach fewer services and data sources outside the Microsoft 365 boundary. Claude Cowork runs locally on your device, which means it can access your PC directly, run PowerShell and Bash scripts, and plug into a much wider range of third-party services. The catch is that you need a PC running 24/7 for overnight tasks, and connecting it to enterprise data may raise concerns with your security and compliance team.

Personally, I run a mini PC at home dedicated to overnight Claude Cowork tasks. Most IT pros will need to figure out which Cowork is the right fit for each scenario before rolling either one into production.

Microsoft Teams Updates From the Microsoft 365 Conference 2026

Microsoft Teams got a healthy mix of fundamental improvements and new AI capabilities at the Microsoft 365 Conference 2026.

Fundamentals: Loop Pages, Threaded Discussions, and Catch Up

Jeff Teper reminded the audience that loop pages inside channels and Slack-style threaded discussions both shipped recently. If those are features you have been waiting for, you now have multiple ways to use them inside Microsoft Teams.

The brand-new feature on the fundamentals side is Catch Up. On mobile, you can now swipe through your notifications and either mark them as read or keep them unread for later. The idea is to triage on the go and answer back from your desktop where the real work happens.

Customizable Meeting Controls

The meeting toolbar in Microsoft Teams is being redesigned with customizable controls. If you have ever clicked Leave when you meant to stop sharing, this update is for you. It is a small change, but it is a good reminder that Microsoft is still investing in the fundamentals and not putting every engineer on Microsoft Copilot.

Facilitator and Channel Agents

On the AI side, the facilitator agent keeps the meeting agenda visible at the top, takes continuous notes, and merges context from the meeting transcript, the chat, and the live web to answer ad hoc questions. You can ask it to produce a one-pager directly from a chat conversation.

Even more interesting is the channel agent, which can drive real business workflows. The keynote demo showed a feedback message in a Microsoft Teams channel triggering an automation that created a bug in Azure DevOps, assigned it to GitHub Copilot, opened a pull request, and tagged the engineering manager for review. The entire chain ran in a few minutes from a single Microsoft Teams message, and the trigger could even be an emoji reaction. Channel agents can also integrate with Asana, Jira, and other systems.

If you run any kind of triage process, this is the kind of automation that can save real hours. It also lowers the bar for IT pros who never built a Power Automate flow because the channel agent can be configured with an inline prompt.

SharePoint Got the Biggest News at M365 Conf 2026

SharePoint, dare I say the best tool in the stack, got the biggest chunk of real announcements.

New SharePoint UI Goes Generally Available

The new SharePoint UI announced at the 25-year party in March is now generally available. That includes the new portal, the cleaner navigation for Discover, Publish, and Build, and the unified OneDrive experience that is now consistent across Microsoft Teams, SharePoint, and Outlook. Microsoft also confirmed during keynote week that a new Outlook mobile experience is on the way, so the Outlook side of that story is not standing still either.

Viva Connections and Viva Amplify Rebrand to SharePoint

Announced just before the conference, Viva Connections and Viva Amplify are being rebranded back under the SharePoint name. Viva Connections Mobile will literally be renamed to SharePoint, and you will see that experience inside Microsoft Teams Mobile with a combined feed of news articles from across your organization.

I have to laugh a bit at this one. When Viva launched, half the internet declared it a SharePoint killer. A few years later, the apps are coming home to SharePoint where they belong. Having watched the SharePoint platform evolve for over a decade, the brand has only gotten stronger. Ten years ago people were asking if SharePoint was dying. Today nobody asks that, because SharePoint sits at the center of Microsoft Copilot grounding.

AI Comes to SharePoint Lists

Microsoft finally fixed one of the biggest pain points for SharePoint power users. AI is now built into SharePoint Lists. You can ask natural language questions on your list data, create new items by describing them, and format columns without writing a single line of JSON. The same experience now works on lists, document libraries, news posts, and sites.

SharePoint Skills Public Preview

The biggest SharePoint announcement at the Microsoft 365 Conference 2026 is SharePoint Skills, now in public preview. Jeff Teper called this his favorite announcement of the week, and I agree completely.

A quick but important note on licensing: SharePoint Skills are included in your Microsoft 365 license and the SharePoint AI license, which is included in Microsoft Copilot. They are available in public preview today, so you can start getting value from them right now.

The workflow is simple. You describe a SharePoint process in plain English, something like “Create a deal with these four folders inside, these two templates, and this metadata.” SharePoint saves it as a reusable skill that lives in your site. Anyone with access can invoke the skill by asking “Create me a new deal folder” or “Create a new deals workspace.” The skills are markdown files stored directly in the site, and they are versioned, governed, and editable using the SharePoint controls you already know.

I could spend hours talking about SharePoint Skills, and I already made a full walkthrough video on how they work, including the file structure and how to build them for your team. The link is below in the CTA section.

One detail many people missed in the keynote slides: the SharePoint Skills UI shown on stage included a grouping of “Sales skills” with multiple skills underneath, plus controls to see, create, and view site skills. None of that grouping UI is in the public preview yet, so Microsoft was clearly teasing capabilities that have not been formally announced.

Agent 365 Hits General Availability on May 1

Ryan Cunningham’s keynote opened with the biggest agent news of the week. Agent 365 reaches general availability on May 1, only a week after the conference.

Agent 365 Pricing

Agent 365 launches at $15 per user per month standalone, or it is included as part of E7. I am not going deep on Agent 365 licensing in this recap, but those numbers will help you start the conversation with your procurement team.

Agent 365 is your control plane for every agent in your organization. It provides a single registry across first-party Microsoft agents, agents built on the Microsoft 365 SDK, and even agents brought in from Amazon Bedrock or Google. The keynote demos confirmed that more platforms are on the way.

The New Shadow AI Pane

A new Shadow AI pane in Agent 365 caught my attention because I had not seen it before. The keynote demo showed a user running an unsanctioned AI client connecting to company data from a Windows PC. Agent 365 surfaced that activity, and IT pushed an Intune policy from the same pane to block it. If you are building governance around generative AI, this is a meaningful addition.

Onboarding Agents With Governance Templates

The Shadow AI pane is the discovery side of Agent 365. The onboarding side is where IT admins will spend most of their time. Agent 365 introduces governance templates, which are bundled groups of policies spanning Microsoft Entra, Microsoft Purview, Microsoft Defender, and SharePoint. When someone in your organization publishes an agent, you decide who gets access, whether the agent should be pre-installed for those users, and which template applies. Once you accept the permissions, the agent goes live with the guardrails you set.

Templates matter because they let you scale governance instead of hand-configuring every agent. As your organization moves from a handful of agents to hundreds, that consistency is what keeps Shadow AI off and innovation on.

One Billion Agents by 2028, and Microsoft’s Internal Number

Microsoft and IDC have shared the projection that there will be over one billion agents worldwide by 2028. I was a non-believer when I first heard it. Then Ryan shared a statistic that made the projection feel a lot more realistic. Microsoft internally manages more than 500,000 agents today, against 228,000 employees globally. That is more than two agents per employee, and agents have been a real category for less than a year. Awesome and a little scary at the same time.

Vibe Coding and the NASA Power Platform Story

Ryan Cunningham’s keynote also introduced vibe coding for the enterprise on the new vibe.powerapps.com surface, paired with one of the strongest customer proof points of the week: NASA running real workloads on Power Platform.

What Vibe Coding Looks Like in Power Platform

The premise is simple. You describe the app you want in plain language, the agent asks clarifying questions, and you get back a plan covering the UX design, functional requirements, data model, and design direction before any code is written. After you accept the plan, the agent generates the app, including sample data, a responsive mobile layout by default, and the underlying Dataverse tables. Inline edits let you switch between visual point-and-click changes and prompt-driven changes inside the same surface.

Two integrations are worth calling out for IT pros. First, Work IQ is built in, so generated apps can pull context from Microsoft Teams meetings, email, and other organizational signals without manual wiring. Second, the data layer ships into Dataverse by default, the same platform that backs Dynamics 365, secured by Microsoft Entra. The app you generate is not a sandbox toy. It is publishable across your organization with the governance you already use.

You can layer a custom agent on top via Copilot Studio. In the keynote demo, Clay Westner stood up an agent powered by Anthropic Opus 4.7 with access to vendor agreements in SharePoint and the Dataverse and Power Apps MCP service, all without re-explaining context the underlying app already knew.

NASA on Power Platform: 47,000 Monthly Actives and a Real Artemis Use Case

Chris from NASA joined Ryan on stage to talk about the agency’s Power Platform program. The numbers are striking. NASA has roughly 51,000 employees with full premium access to Power Apps and Power Platform, and approximately 47,000 of them are active every month. That is essentially saturation, with hundreds of deployed applications and 800 solutions in production.

The headline use case was DWACR, the Display Work Authorization Package Operations and Maintenance Systems Requirements Compact Unique Identifier. Despite the acronym-of-acronyms name, the app consolidated three legacy systems into a single Power Platform solution that gave 20 organizations across NASA, all the way up to the administrator’s office, a real-time view of testing and prep work from T-minus 45 hours down to T-minus zero on the recent Artemis launch.

The story I appreciated most was Chris’s framing of where AI rollouts go wrong. He compared it to surfing: the worst place to be when the wave is coming is in the break, getting battered. The play that worked at NASA was a data SWAT team to clean up source data first, followed by an AI enablement team to connect the dots, applied to a focused subset of the petabytes of data NASA holds. That is a much healthier rollout pattern than “give everyone Microsoft Copilot tomorrow.”

My Top 3 Picks From the Microsoft 365 Conference 2026

If I had to call out the three announcements with the biggest day-to-day impact in the next few months, here is my list.

First, SharePoint Skills. This will change how teams use AI in SharePoint. It will help you automate boring tasks with less code, and it is in public preview already.

Second, channel agent workflows in Microsoft Teams. This makes it easier to automate the things many of us were too lazy to build in Power Automate. With an inline prompt inside a channel, you might actually build the automation this time.

Third, Agent 365. The IT pro in me is genuinely excited to get hands-on. Microsoft has been showing the same Agent 365 slides and demos for months. With GA on May 1, it is finally time to test what is included, what is missing, and what works in real environments.

What This Means for You

If you manage Microsoft 365 or SharePoint, three of these announcements should hit your radar this quarter. Start by enabling the SharePoint Skills public preview in a test site so your team can experiment before the GA wave. The licensing is already covered if you have Microsoft 365 or Microsoft Copilot, so there is no procurement friction.

For Agent 365, line up a conversation with your procurement and security teams before May 1 so you can either pilot the standalone $15 SKU or fold it into an E7 bundle. Pay specific attention to the Shadow AI pane, because it is the first solid governance answer to the BYO-AI problem most organizations are quietly facing.

For Microsoft Copilot, take the time to understand which model excels at which task. Auto is fine for general work, but for visual tasks like PowerPoint design, Claude is worth selecting manually. If you support knowledge workers, share the model picker with them and explain the difference. They will get better results.

Want to Build Your First SharePoint Skill?

Ready to put SharePoint Skills to work? SharePoint Skills is the announcement that will reshape how you automate inside SharePoint, and the public preview is open right now. Watch the full walkthrough where I break down the file structure, show how to build skills for your own team, and demo what is possible today: SharePoint Skills tutorial on YouTube.

Microsoft Copilot | Vlad Talks Tech

Power Platform Just Hit 80 Million Users: PPCC 2026 Keynote Highlights

Inside the Power Platform Community Conference 2026

The Power Platform now has 80 million monthly active users

A new, unified Microsoft Maker Community

Better tools: the first demo had no AI at all

Power Apps: a pop-out formula bar and faster controls

Power Automate: inline action cards and virtual network connectivity

More makers: build Power Apps from Microsoft Copilot

No cliffs: low-code apps are code underneath

The Copilot Managed Runtime fixes the localhost:3000 problem

Copilot Studio workflows and the GitHub Copilot harness

Agents get their own identity, mailbox and OneDrive

The first governance demo on a PPCC keynote stage

What this means for you

PPCC 2027 moves to Chicago

How Agent Approval Works in Microsoft 365

Agent approval in Microsoft 365 depends on the type of agent

SharePoint agents don’t have an approval process anymore

Agent Builder: the sharing default that surprises most admins

Copilot Studio agents need an admin to approve them

How each approval flow looks in practice

Set the main agent on a SharePoint site

Share an Agent Builder agent

Submit a Copilot Studio agent to the org catalog

Approve or reject the request in the Microsoft 365 admin center

What this means for you

Copilot in SharePoint FAQ: Your Agents, Skills and Autofill Questions Answered

Getting Copilot in SharePoint in your tenant

How do I know when Copilot in SharePoint is available in my tenant?

Does Copilot in SharePoint work in GCC?

What is Site AI?

What the Copilot license covers, and when you need Copilot Credits

Does creating a skill cost extra?

Will the prebuilt SharePoint page and list agents need Copilot Credits?

Can everyone use an agent I put on a site?

Does testing an agent in the preview pane use credits?

SharePoint agents are moving to Agent Builder

Is a SharePoint agent basically an Agent Builder agent now?

Why did the agent in my navigation disappear?

Does a SharePoint agent get a Microsoft Entra ID, and does it show up in Agent 365?

Can an agent answer only from approved documents?

Can I choose the model Copilot in SharePoint uses?

Autofill columns

Do autofill columns work with site columns and the term store?

Does turning on autofill add the column to every library on the site?

Can I roll out autofill columns across hundreds of libraries?

Protecting your content

Can I use SharePoint Advanced Management to keep sites out of Copilot?

Can admins sanitize the HTML that Copilot creates?

What this means for you

How Microsoft Copilot Accesses Your Work Data and the Web

Why Copilot data access matters for IT pros

Microsoft Graph and Work IQ explained

How Copilot accesses your work data

The safety checks before the large language model

Your prompt isn’t the prompt the model receives

How Copilot accesses web data

Four things to remember for the AB-900 exam

What this means for you

AB-650 Exam Review: Where Your Study Hours Should Really Go

What the AB-650 exam really tests

What I would brush up on

Microsoft Entra ID

Microsoft Defender for Office 365

Microsoft Purview

Copilot and Agent 365 in the Microsoft 365 admin center

SharePoint Advanced Management

Learn where things live across the admin centers

How much PowerShell is on the AB-650?

Where to start if this is your first Microsoft exam

What to expect on exam day

Is the AB-650 a fair exam?

What this means for you

SharePoint Skills FAQ: Your Webinar Questions Answered

Skills vs Saved Prompts and Other Tools

Why Is a Skill Different From a Saved Prompt?

When Should I Use a Skill Instead of a Power Automate Flow, a Copilot Studio Agent, or a SharePoint Agent?

Running Skills

Can Skills Run on a Schedule, When a File Is Added, or From Power Automate?