Episode 441: Microsoft Ignite 2026: Autopilot, Workspaces, and Spot Plus
48 minAudio
Welcome to Episode 441 of the Microsoft Cloud IT Pro Podcast. In this episode, Ben and Scott take a first pass at Microsoft Ignite 2026, starting with a question a lot of IT pros are asking: is the pace of change wearing everyone out, or is it just us? Ben sums up the event in one word, agents, after counting 31 named agents across the Ignite news, 21 of them built by Microsoft. Scott then digs into Azure Compute Spot Plus, a public preview that targets a best-effort first hour for Spot VMs, waives the compute charge if Azure evicts the VM inside that hour, and gives interruption-tolerant jobs a little more warning before eviction. On the Microsoft 365 side, they cover Copilot Workspaces (the new name for Copilot Notebooks) and Autopilot, the cloud agent formerly known as Scout. Autopilot gets its own identity, runs on pay-as-you-go credits, and does some things as itself and others with its owner’s access, which leaves a split audit trail and security controls like DLP that have not caught up while it is in Frontier. They close with the SharePoint and OneDrive Agent Access Heatmap, which shows where agents are reading and writing content, and Agent Rewind, a private preview for rolling back changes after an agent goes off script. Show Notes Azure updates Microsoft Ignite 2026 Book of News Public Preview: Azure Compute Spot Plus Azure Spot Plus VMs: Public Preview Overview Copilot at Microsoft Ignite: Expanding what’s possible with intelligence built for work Copilot Notebooks Is Now Copilot Workspaces Autopilot (Frontier) overview Configure Autopilot (Frontier) for your organization Autopilot (Frontier) FAQ Microsoft Purview DLP for Microsoft 365 Copilot and Cowork Microsoft Agent 365 overview Agent access heatmap overview (preview) Microsoft 365 Copilot and AI Agent readiness and resiliency with SharePoint Content Governance Overview of Microsoft 365 Backup Prior podcast context Ep 430: Scouting out Microsoft Scout Ep 431: Agent Governance Is the New App Governance Ep 427: Copilot Cowork Hands-On Experiences Where to find Ben Workplace Ninjas US: Ben will be in Scottsdale, Arizona, January 11 to 13, 2027, at The Scott Resort & Spa, where he and Jay Leask plan to record Cloud Pro on the Road. Registration is open. James on Stellantis, Conver on AI, Stegink on Microsoft (Daily Dealer Live): Ben joined the Car Dealership Guy team on November 16 to talk about the recent Microsoft 365 outage and what it means for dealerships. About the sponsors Nasuni is a leading unstructured data platform for enterprises where file data is mission-critical for both people and AI. Nasuni powers the operational file layer where work happens - helping organizations manage, protect, and activate data so teams can work smarter, reduce costs, and operate securely without limits. Intelligink designs, secures, and modernizes Microsoft 365 and Azure environments for organizations of every size, working alongside your internal IT team or MSP. From security audits and Copilot readiness to migrations and ongoing fractional cloud services, it’s the same production-tested advice Ben shares on the show, applied to your tenant. Book a conversation
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
Welcome to episode 441 of the Microsoft Cloud IT Pro Podcast, recorded live November 18, 2026. This is a show about Microsoft 365 and Azure from the perspective of IT pros and end users, where we discuss a topic or recent news and how it relates to you. It's Microsoft Ignite Week, and that means Ben and Scott dive into some of the latest announcements from Microsoft Ignite and provide their thoughts and perspective on them.
So today, we take a look at Azure Compute Spot Plus, Copilot Workspace, Autopilot, the Copilot agent, not the Intune service, the SharePoint agent access heat map, and Agent Rewind. Take me back to studio. It's like we're in technology and things change. Technology changes every day. The fatigue is real. I was just going to ask that. Do you find yourself, I've been debating this internally, and I feel like every night now, I go downstairs from work because I have that long commute from upstairs to downstairs.
And I tell my wife, I'm like, I'm tired. Like, do you feel like the pace of change is just exhausting? And is it the fact that we're getting old? Or is it the fact that, to your point, it's speeding up? And it feels like every day you have to learn something new as opposed to like every week or every month or back when we were in the on-prem days and it was every three years. Like, do you find yourself getting more exhausted? Does some aspects of it energize you? I'm curious. This is a side topic before our main topic.
The cognitive load is very real. Like, it's constantly there to think about ways to, at least with kind of the role I'm in and my employer, you know, continue to accelerate yourself through agents and using AI to augment and move faster. So that's an interesting pivot, I think, in general. For a lot of folks, and especially for ICs, when you're basically being told like, hey, you're kind of like a manager now, but you manage a team of agents.
That's a whole different just way of operating and having to wrap your head around. So one, you know, getting into that modality and then figuring out how do I manage a team of technical players tends to be a part of it. The other thing I noticed for me is the stuff that I do during my day job also can closely align to my hobbies at night or on the weekend. And so something like hacking away in the past, like hacking away at a script to like organize a bunch of photos or something like that.
Like, hey, let's look at all the metadata and organize those photos. You're kind of doing the same thing during the day where you're hacking away at your job and then you're still hacking away at the personal projects at night. So the same things that I'm doing during my day job where it's like, oh, hey, I see this toil work that can be repeated. So let me go and generate a skill or an agent or figure out the right harness to wrap it in, things like that. And then at night, it can turn into the same thing. Like, you know, I've noticed with a bunch of stuff from be it like Home Assistant, like where I would have built a dashboard by hand in the past and kind of hacked away at it.
Now it's just going back and forth with an agent and saying, well, make it look like this, tweak this, let's move this over here kind of thing. So I'm not getting like hands-on with Home Assistant anymore in the way that I was in the past. And I'm sitting in the same harness and the same tooling, be it like Claude or GitHub Copilot or whatever it is. I'm sitting in the same thing at night that I am during the day. So I think that's a contributing factor, right? Like, are you ever actually shutting off and kind of touching grass and getting out there? All to say, I probably need some new hobbies.
Get me out of the house some more and don't contribute to that load and that kind of mental burden that comes with it. I think I'm feeling it more from maybe a little bit the consulting side of things where I'm getting asked questions and trying to keep up with this stuff. Well, your world is changing just as much as mine from the way of work to customers and how they interact. Yeah, and being able to accurately answer customer questions, like going into the portal. And I feel like every time a question gets asked, I'm like, oh, let's go look at this.
And something's new and something's changed. And I feel like it's a full-time job with the rate of change to keep up with stuff, to be able to talk to clients, answer clients, all of that. It is a full-time job. You got to have like a process to get through these things, right? So today we're going to talk about Microsoft Ignite and some more changes, changes, new things that are coming. So you have to kind of come up with that process. And even for me, like this was kind of a, this was a different year to go through Ignite news.
You know, things like blog posts and the Azure updates feed coming out later, the book of news coming out later, things like that meant like a different way of work this year to consume it, to turn around and be ready for something like this. Whereas in the past, what we would have had like a little bit of time to go and consume this stuff and figure it out. It's just bang, bang, bang, pretty rapid. I don't know if folks watch like the Azure updates feed. I wasn't watching the M365 one really hard yesterday, but on the Azure one, it was like 120 plus updates just jumped, like just dropped within like two to three minutes.
Lots of stuff, some little stuff, some bigger stuff, some little stuff that was kind of wrapped up into bigger things, but then disseminated as 12 updates when really it was just kind of one thing or one like unit that was out there. I have a summary, Scott. I looked through the book of news. Claude looked through the book of news. Claude looked through other news sources. I have decided if I was going to sum up this Ignite in one word, it is agents.
Agents, agents, agents. Agents and more agents. I actually asked, so right before this. Agents and connectors, yes. Agents and connectors. I went out and said like, go look at all the Ignite news and count the number of agents that are mentioned in Ignite news. It came back with 31 named agents in total across like local coding agents. Microsoft built named agents was 21 agents that were mentioned in Ignite.
Some of them are previews. Some of them was frontier stuff. Some of it was generally available, but yeah, it's like agents. And then there's like agent-like experiences that Claude pulled out and then partner agent-like platforms that were mentioned. It's just agents everywhere. It's interesting, even with the number of, I think, kind of first-party Microsoft-released ones, there's always this push-pull. And it's probably more prevalent now during events like this. But you can see kind of where Microsoft wants to become a little bit of that connective tissue for everything, but also go at the same time and own a little bit of their own destiny when it comes to putting some of this stuff out into the world.
So I can have a Salesforce connector, and if you're a Salesforce connector or a Salesforce CRM customer, that's great. That doesn't mean that Dynamics isn't still going to be getting pitched at you or sitting there with some set of differentiated capabilities that hopefully sway you as a customer one way or the other kind of thing to come in and consume it. So it's a brave new world of everybody just trying to go out and ultimately get folks to a new way of working while at the same time thinking about new ways of work and where you want to go.
And for you, especially being in consulting, where your customers want to go with their businesses and what are their goals and outcomes that they're trying to achieve. And then figuring out how all this stuff plugs in and lines up, it's an exercise. It's that full-time job, right? It's that full-time job we talked about. Yes, yeah. Do you want to talk about some of the announcements? Agents and otherwise, I went through and there's some Microsoft 365 stuff that actually intrigued me that was non-agent and non-AI. I was pleasantly surprised with a couple of announcements.
I think there's some interesting stuff out there. There's some that I want to go dig hands-on or get hands-on with a little bit first before I really kind of come back here and talk about them. I saw, I don't know if you caught this one, there's a whole new templating and deployment mechanism for VMs. And I was like, oh, hey, what does ARM do? How is this different than Bicep and the existing automations that are available to us? Or maybe if I'm in a Windows environment, stamping that down. There was a bunch of things there, even down to VM management and image management and things like that.
But yeah, on the non-agent side of things, I think there was a bunch of goodness for customers when it comes to compute and having more choice about compute and how they consume it. There's very real capacity constraints and trying to understand what VM series and when to use what can be hard, when to buy a reservation, how to be cost conscious, and maybe leverage things like Spot has become harder to do with some of the capacity crunches and more aggressive evictions, things like that.
So maybe one of the fun ones that I found was the Azure Compute Spot Plus, which is out in public preview. So this is for customers who run batch jobs on Spot VMs today. You might be out there and you're already doing that, but now they've got this Spot Plus thing. So with Spot, Microsoft is coming in and saying, hey, we'll kind of target your first hour on a best effort basis.
And then VMs can be evicted within there. But it gives customers a little bit more flexibility along the way. So it says, hey, for that first hour of Spot spinning up, you're going to be allowed to run a little bit more. You still might get evicted, to be very clear. You still could get kicked out. But it's nice to see the optionality there to say, I want to consume Spot. There's kind of like maybe a little middle ground between a Spot Compute VM price versus a full VM versus a Spot Plus VM and the best effort model that exists within there.
So that was a fun one. It's got a couple different signals within health of the service that are there for customers. So you can kind of understand when rebalancing is going to happen or when reallocation is going to happen along the way. The same API surface, which is nice. If you're already using Spot, it's effectively an extension to that. You're going to have this new Spot Plus API that comes out alongside the other API sets for Spot, but available with single VMs, VMSS Flex, VM scale sets, all that kind of stuff sit out there for them.
I think that was a good one to kind of give customers, like I said, a little bit more choice, maybe a little bit more predictability. We'll have to see how it goes. Like if it's actually worth paying a little bit more for a Spot VM to still have it evicted at like 30 minutes. Like how does that 60-minute best effort thing come into play? And then how does it come into play across Spot Plus VM series? Like there might be some VM series that get to running that 60-minute threshold a little bit more than others kind of thing.
So that might be one for folks to watch if they do use Spot compute. Sweet little area, like I said, between regular Spot and then kind of full VM. Hopefully it brings a little bit of that predictability. The other cool thing they did there was, you know, you can really tell like it is like really trying to target like that best effort thing. So the billing model for a Spot Plus VM, like if you know that job needs to run for 45 minutes and you're really banking on it not being evicted, but you're looking for a little bit of something, like if it does get evicted,
there's this billing protection that's built into the business model for Spot Plus. So if the Azure fabric decides, or like the compute fabric, you know, evicts a Spot Plus VM to reclaim capacity for something else, and that eviction happens within the first 60 minutes, the VM's usage isn't charged, which is nice. So you don't get charged for the compute. I imagine like it wasn't very clear in the documentation or the posts that I read. There's going to be other associated charges there, like networking, things like that.
I'm sure you're going to, I'm pretty sure you're going to continue to get charged for that stuff, but the actual compute costs are good. So it just occupies a little bit of a different space that's out there. It's available in preview today. So, you know, you've got to be on kind of like the latest REST API versions to get going with it. There is an Azure feature enablement flag that you have to enable. So, you know, you have to go like register a feature against your subscription and just wait for that feature to register and come through.
Other than that, it looks super easy to onboard too, because like I said, if you're already using Spot, it's the same deployment surfaces that are there for you. So if you're doing single instance VMs, VMSS, Flex, or Uniform, it all just works there. What is a use case that you can think of? And I was thinking like if you're going to use this for like a DevOps runner or a GitHub runner where it's going to run like a quick seven, 10 minute job right when it goes up. But those machines you don't necessarily boot when the runner goes, they're always sitting there.
I was trying to think through like workloads where maybe you'd go to that Spot plus VM versus a Spot. Because it looks like it's really targeted for those. You spin it up, run something right away, and then shut it down again? Or I was kind of working through this in my head. You're targeting things where you're looking for some level of determinism in the runtime. So let's say I had a, this VM boots up and it's got a script that it goes and runs, right?
It iterates over something and maybe does some like transformations on logs or something like that. And I know that thing takes 30 minutes, 40 minutes between VM bootstrap plus run the script and go and do it. And then before it's done, yeah. So I think it's great for those kinds of things where the determinism is in the runtime that you need from bootstrap to this unit of compute has done the thing. Done its job. So if you're looking for something that's, you know, more than a couple minutes and up to an hour, Spot plus would kind of fit that.
That's where this is going to fall. Would fit that bill and be in there. It is still that best effort kind of thing, but at least you do get the credit back if that capacity is reclaimed for anything where you're not charged for that usage. So you'd have to rationalize through and see what that looks like, right? But it's very different than, like that billing protection doesn't exist in regular Spot today. There is absolutely no minimum runtime target. So there's no like best effort. We're going to get to 10 minutes with Spot. We're going to get to 60 minutes with Spot plus kind of thing.
So if your workload is just super resilient, you're just looking for the lowest cost, when it gets done, it gets done, I don't care if it gets evicted, not the thing for you. But if you're not as resilient or you're still in a place where you are, how does Microsoft phrase it? They phrase it as interruption tolerant. So if you are interruption tolerant and you can kind of recover, like, hey, that thing that was supposed to run for 30 minutes, it did get evicted, but we got the money back for it, we're just going to go spin up another one and start it again.
And you're looking for that runtime consistency, then it's just sitting out there ready to go for you. And kind of like you said, even though you get the cost, the money back, you still want to make sure that workload's going to be interruption tolerant in case you do get evicted, you don't want that to break something. You very much still need to be in. Have resilient workloads that it's doing, yeah. You need to be in that modality. So it's not for, like, every workload kind of thing. Again, back to my example of, like, hey, maybe I'm going to, like, iterate over a log and compact it or do some transformation.
If that thing's just kind of checkpointing progress as it goes, do you really care if the VM gets evicted? Because you'll just kind of pick up with the next one, wherever the previous job. Right, wherever it left off. Yeah, left off. There's also a little bit of, you get a little bit more of a grace period for eviction. So it looks like with Spot Plus, I'd have to go test this out and see, but it looks like you get, with regular Spot, like a 30-second notification. If that, and then you're just booted out, you get a 30-second to two-minute notification.
So you do get a little bit more of a notification window if the compute fabric is coming in and getting ready to evict you or rebalance things. I'm going to go into agents and AI just because this was a bigger one, I would say. And one we've talked about is the whole Copilot Workspaces and Autopilot. We could kind of combine these into one. This is, I would say this was one of the big, this was probably one of the larger Copilot announcements.
I would say there's some other interesting ones in here. But essentially, our Copilot notebooks that we used to have are now called Copilot Workspaces. So if you're going through, you're seeing some of the stuff from Ignite about Copilot Workspaces, this is a bit of a rebrand from Copilot notebooks. It's essentially giving you a shared place to work on a team or work with your team to keep working within Copilot. So this also is going to bring in, you can bring in things like files, pages from SharePoint, your shared work context, and bring these into these Copilot workspaces to be able to essentially collaborate with team members within Copilot.
This is a little different, and then notebooks are going to automatically become workspaces. So you are going to want to keep an eye on those, just automatically flipping over to workspaces. This is another one of those Frontier things, I think. Workspaces experience on Frontier. Yep. January 2020, I guess 2027, 2027 is the GA for those. And kind of with that, like, Autopilot got announced alongside of it, and you're going to be able to bring Autopilots into workspaces as well?
We should maybe take a step back real quick. Should we talk about what Autopilots are? Before we do that. Or workspaces? Workspaces are effectively what Notebooks was, with some new functionality on top of them. So if you were already using Copilot Notebooks, congratulations, you will feel right at home in Copilot workspaces. It's going to feel the same. It's going to feel the same. But we've always had this kind of, like, projects and Notebooks thing. Or especially if you've been in other providers, like if you say, like, Claude and their desktop app, or ChatGPT, Codex, you've had this concept of projects within those.
And then they do projects as individual things or as shared things. Microsoft's approach seems to be a little bit different here, where you still have projects for personal context. And then you'll have workspaces, artists formerly known as Notebooks, for more team collaboration and bringing that through. So that modality, still a little bit weird. Like, why not just have a project that can be an individual project or be a team project? Or why not just have a workspace that can be an individual workspace or a shared workspace?
Who knows what it's bootstrapping in the back end to make it all work with some of the kind of plus-plus functionality that's there with Copilot workspaces. But they're being very intentional in positioning them as workspaces being the place for team shared work with something-something Copilot within there. So let's put all the documents for our team project, a collection of SharePoint pages, anything like that in there.
At least Notebooks are going away. Like, it's all just going to be transitioned over to workspaces. So you're not going to have projects, workspaces, and Notebooks. You're just going to be left with projects and workspaces that are out there. I'm curious to see... Can I upload? That's files. Attach Cloud files. So I... Yeah, this is all going to get weird. I'm just going to say it's going to get weird because I have workspaces, and then I can toggle my workspace between a workspace. So there's a workspace where it's only going to use content from this workspace, and then you also have workspace and web, where it's going to use content from the workspace and the web.
And then you can still go in and add your work content, so you can add plugins, skills, people's files. But I'm not sure where plugins and skills come from. Because I have plugins and skills in my OneDrive for, like, co-work. But they don't show up in a workspace. I don't see settings. Okay, so I can upload plugins. So it looks like in the settings of the workspace, I can maybe add skills and plugins and all those. I wonder if they're going to go across workspaces.
Or if I'm going to have to... If I want certain skills in there, like, am I going to have to upload skills to every single workspace that I want them in? I would bet, for now, that it's probably the former. The latter, arguably. It's the latter. The latter. The former being the better experience, right? If I don't have to upload this every time. If you think about, like, co-work and what's sitting there, that's a little unit of compute. That, like, little mini VM that's sitting out there. And, yeah, it can dynamically inject skills from your OneDrive, things like that.
That's going to be a lot harder in, like, a shared team context. I guess, what's the analogy I would use? Oh, like, you and I. Like, you and I have the LLM wiki for the podcast that we share. That wiki has a set of local skills that are in it so that we can both pull it up and those kind of follow us around. Versus, like, Scott's global skill that he has installed for, you know, organizing photos from his camera. That kind of thing that you never want to see in yours that's out there.
So, that whole skills thing and Copilot is kind of, like, messy or co-work is kind of messy right now anyway. So, I don't know that I miss it too much. I still see a lot of Copilot as folks who are out doing the prompt engineering thing and not so much the other ways of interacting through skills for more determinism, anything like that. If you're managing file storage today, you've probably felt the pressure. Refresh cycles are harder to plan. Costs and complexity are increasing, especially as file workloads scale across regions.
That's why Microsoft customers trust Nasuni. As an Azure-certified software, Nasuni provides a global file system designed for enterprise-scale environments, keeping file data secure, governed, and accessible across Azure environments without added infrastructure complexity. It's the foundation to keep your unstructured data stable, even as your infrastructure evolves. To see how much you could save while keeping your initiatives on track, visit nasuni.com slash TCO.
That's N-A-S-U-N-I dot com slash TCO. I still want, on my wish list, I don't think I've seen it from Ignite, is like a plugins, skills, marketplace, library. Like, this could be a SharePoint site that you stand up, that you put all the markdown files in, or something to that effect. Or GitHub. Similar like Claude, I can pull stuff from GitHub into any of my environments. Why can't we have something like that in Copilot? Where I can go stand up a central skills plugin repository in my tenant.
Then when I go to like a workspace, and I want to add a plugin or a skill, it can just pull from my marketplace. Or when I go to Copilot, it can pull from my marketplace. Or when I go to co-work, it can like, yeah, off my soapbox. Some day. Wish list, Microsoft, can you please do that for me? Now, the funny thing is, you could go create a plugin marketplace today, but that plugin marketplace will not work with M365 Copilot. It will absolutely work with Codex, it will work with Claude, it will work with GitHub Copilot, but it will not work with M365 Copilot.
I'm not going to rant on that one. You mentioned autopilot. Yeah, we should talk about autopilot. Much like we're having the artist formerly known as here with notebooks and workspaces, autopilot is the artist formerly known as Scout. Welcome to the new cloud version of Scout. We talked about this on another episode when it was first discussed, but now we have a lot more information about it. So this is an agent that is in preview today, or should be in preview today, is starting to roll out in preview.
I do not have it in my tenant yet. I went and looked. So the way to get this in your tenant and for users to start using it is you go into the admin center, and you go into your all agents in your Microsoft 365 admin center. And then you can go in and kind of add it, or I can't remember what the exact term is they use there, whether it's add or install or enable or one of those. And then pick, do you want all users to be able to use Autopilot, or do you want just select users to be able to use Autopilot so you can scope it?
And then as part of that enablement of Autopilot, it will also walk you through a cost policy to say how many credits you're going to give people that are going to use Autopilot. Autopilot because, lo and behold, this is one good thing. It doesn't look like Autopilot requires my GitHub Enterprise account like Scout did. I haven't seen anything about that, but it is a pay-go agent. So you're going to want to work through those cost policies and costing settings so nobody goes out and gets Autopilot and blows through thousands of dollars.
In a month, or hundreds of thousands, depending on the size of your org and how many people you give it to. Also, not clear how many. Do you see a proliferation of these things? Back to our jobs are becoming managing swarms of agents and teams of agents. Do you have one Autopilot who's really good and is targeted at helping you manage your calendar and your inbox, things like that? Maybe you have shared team Autopilots that get more broad context because these things, they're units of compute.
They're no different, I think, on the surface than some of the other agent experiences that are out there, like personal agent experiences. So if you think about a muse for meta, dots, duet, things like that, very similar in the mindset. It's just having access to the M365 ecosystem and being able to kind of follow along alongside you for your work is Microsoft superpower in that they have this big graph of knowledge that exists about everybody. This is going to get interesting, Scott.
So I was curious to dig into some of this, even from that administrative perspective and looking through how some of this works. Because one of the things I pulled out is this agent, this Autopilot agent, is going to have its own agent ID and it's going... Its own, effectively, like mailbox, teams identity, all sorts of things. Right. Folks should go and read the docs, even like the IT admin docs for this one.
And really read the docs. Even as like an end user, just to get a sense of this world of like active conversation, recurring background work, all the connectors and tools and things that can come into play here. So Autopilot is kind of meant to be that intersection for you, right? It's meant to sit there and be a personal agent that can help you prepare your work for some kind of review. Maybe that's having it help you draft your emails or draft documents, review a presentation, things like that.
It can do things because it's running constantly with like its own compute in the background. It can kind of run things on a schedule. Microsoft talks about it following commitments. So, you know, going in and asking it for unresolved decisions, upcoming deliverables, stuff that you can do today with Copilot. But it's a very like disconnected experience to try and figure out like when and where to ask and how to prompt and maybe even like what model to use. All those kinds of things that are out there. And because these things are identities, you're going to be able to pull them in and they are autonomous in that way.
Like you're going to be able to pull them in and say like you and I are joined in an intelligent workspace because we wanted to do some stuff. We like if we both had Autopilot agents, we could also invite those Autopilot agents in and give them access. And then they would have that context and interactivity there. And one of the things that I found interesting with this is when it starts acting as its own identity, like you said, it has its own name, its own email address, is it's doing certain things as itself. So if I'm interacting with it, I could, for instance, say, go look at this particular information, go look at the information about the upcoming podcast, send a note to Scott that gives him a summary of what we're going to talk about in the next podcast.
So now from an audit perspective, and from a conditional access perspective, from some of the DLP perspective, there are certain things it does as itself on behalf of me. But in audit logs, they're going to show up as Autopilot did this or Autopilot changed this or it can, it's getting a little goofy there. And I started looking through some of this and best I could pull is there are absolutely things it's going to do as itself that in audit logs are going to show up under the Autopilot username ID.
But then there are things like when it goes and accesses content that it is going to, it's going to work like a delegated app where it's going to be based on what I have access to. So it's not that Autopilot's going to have broad access to a bunch of content that it's going to be pull, that it's going to pull, that I can't. It's still going to respect that. So you end up with a little bit of a fractured audit log even for certain things, which will be, it'll be interesting to sort through that. It's all going to come under like agentic governance.
You're going to have like M365 agent governance and things like that, that I think will, or agent 365, that will try and bridge that gap. And get it to where it needs to be. The one thing I will say about Autopilot, you know, I've seen people kind of thinking about the already, like what they're going to name their Autopilots because you get to, like it's part of the provisioning experiences coming in and picking a name. Yeah, guess what? Your Autopilot, like the thing that you go provision, is getting an identity. It's getting a team's identity. It's getting a mailbox identity, all these kinds of things.
So it's going to be visible in Outlook. It's going to be visible in Teams. It could potentially join you in your meetings. So you don't want to make it like something starky or anything like that to come along with you, which is another weird thing. I will say, like, I've been, we kind of started out, like, do you get more tired and fatigue and things like that? One of the things I've been struggling with lately is as I'm directing more and more of these things around, I am starting to anthropomorphize them and treat them as humans. And I'm like, hey, you go do this, you go do this. I'm not treating them as just like the computer from Star Trek anymore.
Like, computer, you know, tea, Earl Grey, that kind of thing. I think this is going to contribute to that too as well, though, where you start to see more of these things showing up as people when they're not people. Like, yeah, they'll be delineated and you'll have that there. But it's not going to be like, hey, computer, go do this for me. It's going to be, you know, hey, Joe, Sally, whatever somebody names their Copilot to go and do that. So that's another weird slippery slope to put folks on the path of these things are going to, just by virtue of their names and the way we interact with them, become more integrated into our day-to-day and feel more and more human, which is a little weird.
The other thing I want to mention, because I ran into this with co-work, is some of these frontier models that are, I mean, it's essentially preview, right? Not all of the security controls that you might expect to be their work. We had that conversation about with co-work when DLP wasn't working right out of the gate. I'm seeing the same thing. That was one of the first things I went and looked with this. I went and looked at the DLP coverage list from Microsoft, Autopilot's not mentioned in there, and pulling out some of the documentation is like DLP for prompts with Autopilot are being tested in some tenants.
Sometimes you'll get generic errors. Sometimes prompts will, you'll be able to keep going with a prompt. Sometimes files might not get labeled. Like, you should absolutely treat some of these as that preview functionality and not all of your traditional security controls that you may be... I think people think, oh, it's just part of Copilot. Everything that's turned on for Copilot is going to work for this. That's not necessarily the case based on what I have seen with these and what I've experienced testing some of these is be careful rolling these out.
Don't just turn on Frontier for everybody and expect all of the security, governance, et cetera, to work for this the way it does for something like Microsoft 365 Copilot. I think it's going to take a little while to fuzz through the system and figure it out. So it is a unit of compute, like it is a persistent unit of compute, like a micro VM or a mini VM that's running 24-7 that also has an identity associated with it and can reach out and do these things on a customer's behalf or autonomously.
Like, if they've told it to go out and do that. I mean, it's really like, if folks remember, like back to the world moves so fast, if folks remember like OpenClaw, this is OpenClaw in the cloud and running through and doing what it needs to do, right? It has memory, like a memory.md. It has an identity, like a soul.md, that have been put into some of those harnesses like Hermes and OpenClaw, things like that that are out there. So brave new world with that one.
I think it'll be telling to see like how customers pick it up and how they integrate along the way with it. It's getting weird out there, right? For like, what's the repeatable work that folks are doing, where they're choosing to exert their big air quotes, like taste on these things and how those fall in. I also wonder with some of this stuff, like how much worse off are we going to be on the other side? Because I see a lot of folks who, I felt this and had to pull myself back from it, who offload cognition and cognitive load to these agents because they do start to really treat them as people and think about them as like, oh, this isn't just my autopilot.
It's not just my scout. It's like a co-worker or a set of co-workers that are there for me kind of thing. But autopilot will be nifty. It has some differences from scout because it is a cloud VM that's running. Things like local disk access, like some of that feels like a regression with some of that going away. They do have some options for, I guess, computer use light. So they can do like web browsing. There is the ability when you're going in and you're looking at like admin controls for your users, you can turn on a capability where you can allow autopilots to interact with like local edge sessions, things like that.
So it'll have computer use light and a couple of things like that along the way. The governance game is going to take a while to suss out, I think, just to see how organizations use them. 100%. Your organization is different than mine, is different than the next one. Everybody's going to have very, very different needs when it comes to all that. Do you want to tackle another item from Ignite? Yeah, let's keep going. I don't know how much time you have or when you have meetings or any of that. I got to stop at the hour and then on to the next. Pick one. We're going to have to come back because there's a few more in here I want to talk through and we're never going to make it through in seven minutes.
Do you want to stay on agent stuff? Can I steal one of your M365 ones? Sure. While we're talking about agents? I want to know which one you're going to steal. I want to talk about the SharePoint and OneDrive agent to access heat map. Oh, I liked this one. I'm glad you picked this one. This one I am actually a big fan of. So what is this one? You know, you've got agent 365 out there. You've got just this proliferation of connectors and other things that are running around for you. So this gives you a heat map. It gives you a grid of where rows in the grid are your agent platforms, columns are your business groups or your site, like your SharePoint site or OneDrive sensitivity labels.
And then you can go and drill in based on kind of the heat map, red, green, yellow kind of thing, of identifying the agents, owners, activities at like a site level, which is really cool. So you'll start to be able to see some of this stuff bleed through the ecosystem a little bit more. So effectively, you get like a statewide, tenant-wide, SharePoint tenant-wide visibility into agent usage in a heat map view, and then understand like where the hotspots are and where people are using these things the most within your organization.
Yeah, this one is, I really like this one because I think that is one question. I don't know that clients have verbalized it to me, but I feel like this is a question that a lot of people have is, well, when I'm going and asking these agents questions, and I have these agents connecting to different sites, and I have everybody building different agents connected to different data sources, what content is it pulling? And some of the scenarios they give with this, where it's questions that people might have is, from something as basic as what content are agents accessing the most?
When people are building agents, when people are building agents, what are those files, what is that site that all these agents are going out to and using regularly? To even things like, to your point, looking at sensitivity types, or sensitivity labels, sensitive information types, is, are there agents that are actually going out and connecting to your most sensitive information? Give me a heat map of sensitivity labels that agents are accessing, or sensitive information, I don't know if it has, it says sensitivity labels, maybe it will do sensitive information types too.
I think it's just labels for now, so if you go out and you look at the docs for this one, and the announcement, things like that, there's a couple fundamental questions that this is meant to help with. So one is, which content are agents using? So you'll be able to see which SharePoint sites, which OneDrive's account, and which of your O365 group, or I guess M365 team connected sites, are being accessed by agents? And which kinds of agents are interacting with them?
So those could be agents that are coming from GitHub Copilot, or from Claude, they could be coming from another external platform that was given access to yours via connector, things like that. So that's question one, is which content are agents using? Question two is, whose content is it? So what is the business group or the operating entity within your organization? And should those agents actually have access to that? Because in the age of SharePoint and permission sprawl and everything else, it is very easy for that stuff to get out of whack.
The sensitivity thing is, how sensitive is it? So you'll be able to see a grouping of agent access by sensitivity label. The last question is, do agents only read it, or do they also update it? Which I think is an interesting one, like back to autopilot, which we just talked about, which is going to be able to go out and execute actions on your behalf, or because you tell it to. Well, it's going to be going out and it's going to be writing things. It's going to be impacting and interacting with the environment. I think it's also going to be a way for folks and admins and organizations to rationalize human time versus agent time in your org.
If you all of a sudden see that, let's say you have a team site for your monthly or quarterly business report, and all of a sudden you see that agents are the only things that are accessing that, and they're reading and writing in there, your MBR or your QBR might have just turned into AI slop, and you didn't know it because you weren't paying attention kind of thing. So I think it'll help with that. It'll also help customers understand where the highest impact items are, both from a business process perspective or just from an information leak perspective.
You might find sites that are super lit up on the heat map that you never even knew that these things were going out there and interacting with. Super quick, did you also see, this was not in the book of news, but it's another one I pulled out. And because you mentioned agents editing content, there is also a new feature that they came out with that's in private preview now is Agent Rewind, where you're going to be able to essentially pick a point in time and roll back to before an agent made the updates.
So you pick a point in time, roll Microsoft 365 resources back to it to undo if an agent went rogue and edited a bunch of things and made a bunch of changes to your files. I did not see this one. I can post you a link. There's a session about it. There is an Ignite session about it called Agent Rewind. And then there's a private preview link, sign up for it. Yeah, that's interesting. Is it using backup, like M365 backup in the background or versioning?
Because I can see Rewind for documents and things like that. Right. I think the issue there is, you know, you can take a snapshot now, but just with like the proliferation of these things and people interacting with things, like you have like 10 people open a Word doc at the same time in a SharePoint site. That's not just 10 changes going through. Sometimes it's hundreds of changes going through in minutes rapidly as folks are clicking around and accidentally hitting the space bar, the inner key or whatever that is along the way. It says, today, Agent Rewind, blah, blah, blah, blah, blah, you can rewind unintended actions, pick a timestamp which you want to rewind your resources, Agent Rewind will take care of bringing that known good state.
It does not say... And I guess because it's framed in context of rogue agents, it's more like a data protection play. Right. It would sit alongside any kind of like anomaly detection. It would be like ransomware protection, mass deletions, it's trying to protect you from that. That's what it feels like because they do bundle it in with a lot of that. It's bundled in with like the Microsoft 365 backup updates, the enter backups. It's a lot more resiliency.
The heading it's under in the announcement page too is resilient AI, business continuity when agents go off script. So whether it's a rogue agent that somebody puts in your tenant or, I mean, I'm an admin. I've written rogue PowerShell scripts before. It's possible people unintentionally write rogue agents in your environment that you need to go undo some stuff. That's where it's bundled more versus something like versioning where you just need to go back a version or two. This feels like more a point in time snapshot.
So I guess from that perspective, Agent Rewind, really nothing new because we've had M365 backup. There might be some extensions, the API surface, things like that. Right. We've had the library restore the snapshots. What it looks like it can tell you is maybe a little bit around like when to restore to based on signals within the platform like anomaly detection. Like I saw a bunch of agents or an agent come in and just make a bunch of changes at this timestamp in this site. So therefore, you might want to roll it back to here kind of thing.
And it doesn't look like the whole site. They have like a little animation in the blog announcement where it shows you can go look at all your agents. So it doesn't necessarily go site by site, but it can see the resources that it changed across SharePoint, Teams, Exchange, and then recover those resources, just the resources that an agent taught.
So it is scoped to agent changes, not necessarily a whole site or something like that. And it's interesting that Rewind itself is an agent. Yeah, because it looks like it's part of the, again, going off the screenshot. Copilot experience, yeah. Copilot experience. It's M365 governance agent. Governance agent. Governance, yes. Same thing. I thought that one was interesting in light of kind of the autopilot stuff and what we had just talked about with the heat map about is it reading stuff or is it editing stuff?
I think we'll be taking a look at some of these announcements as they kind of fan out over the next couple months and mature a little bit. I did see a lot of previews that are out there. So who knows how this stuff changes between preview and GA. Lots of gated previews as well. So I think that's another sign of the world is moving faster. Microsoft wants to put things in front of us as customers quicker, but they still need to kind of meter out the access list and make sure that it's kind of controlled for capacity and usage reasons along the way.
Yeah, I think it'll be fun times here for a little bit, but just based on the volume of announcements across Azure and M365 here, I imagine between now and what, March to May-ish, whenever build and kind of the next set of conferences is, you're going to see a whole bunch of stuff just slam out in a GA fashion, which will be fun. It will be. And I'm excited to get my hands on some of these previews. And like you said, we'll play with them and have some more updates. All right. Well, thanks as always, Ben. Always appreciate it. Not a problem. Thank you.
Enjoy the rest of your afternoon and we'll talk to you next time. If you enjoyed the podcast, go leave us a five-star rating in iTunes. It helps to get the word out so more IT pros can learn about Office 365 and Azure. If you have any questions you want us to address on the show or feedback about the show, feel free to reach out via our website, Twitter, or Facebook. Thanks again for listening and have a great day.
Welcome to Episode 440 of the Microsoft Cloud IT Pro Podcast. Recorded live at the North American Cloud and Collaboration Summit in Branson, Missouri, this MS Cloud on the Road conversation has Ben sitting down with Jay...
Welcome to Episode 439 of the Microsoft Cloud IT Pro Podcast. In this episode, Ben and Scott dig into two Azure agents with confusingly different names: the Azure Copilot Troubleshooting Agent and the Azure SRE Agent....
Welcome to Episode 438 of the Microsoft Cloud IT Pro Podcast. Back in Episode 429, Ben and Scott introduced LLM wikis. In this episode, they come back a few months later to compare what those wikis turned into after...
Welcome to Episode 437 of the Microsoft Cloud IT Pro Podcast. In this episode, Nick Burling, Chief Product Officer at Nasuni, joins Ben and Scott to talk about getting unstructured file data ready for AI. Nick digs into...
Welcome to Episode 436 of the Microsoft Cloud IT Pro Podcast. In this episode, Ben and Scott dive into Microsoft Entra Tenant Governance. Almost nobody runs one Microsoft Entra tenant. Mergers, divestitures, dev and...