Episode 439: Azure SRE Agent vs. the Azure Copilot Troubleshooting Agent
38 minAudio
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. Scott’s mental model is that the Troubleshooting Agent is the free, human-in-the-loop entry point built into the Azure portal. It runs as you, works one problem at a time, and opens a support request when it gets stuck. The SRE Agent is a paid operational platform with its own managed identity. It watches your logs, alerts, and repos, forms hypotheses, and can act in review or autonomous mode, much like a new member of your team. They also work through the SRE Agent’s Azure Agent Unit (AAU) billing model, the built-in guardrails worth copying into your own skills (no deletes, no Key Vault, respect for resource locks), and the VNet integration that lets the agent reach private endpoints and on-premises systems. They close with why you might spin it up for a few days just to see what it finds, and their ongoing wish for one place to manage every Microsoft agent. Show Notes Azure Copilot Troubleshooting Agent Agents in Azure Copilot Overview of Azure SRE Agent Azure SRE Agent portal Evaluate Azure SRE Agent (30-day trial) Pricing and billing for Azure SRE Agent Azure SRE Agent pricing calculator Run modes in Azure SRE Agent Execute mitigations in Azure SRE Agent (safety guardrails) Azure SRE Agent network integration Configure network controls for Azure SRE Agent Prior podcast context Episode 438: Taking your WorkOS and LLM Wiki to the Next Level Episode 434: Reading the Spec Sheet on Microsoft’s Agentic SOC Episode 432: AI-Era Consulting Billing for Microsoft Cloud Delivery Episode 419: Security and AI: Security Store, Security Copilot, and Agents 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 - Would you like to become the irreplaceable Microsoft 365 resource for your organization? Let us know!
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 439 of the Microsoft Cloud IT Pro Podcast, recorded live on October 19, 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. In this episode, Ben and Scott compare two new AI agents for Azure, the Azure Copilot Troubleshooting Agent, a free helper in the Azure portal, and the Azure SRE Agent, a paid service with a portal of its own.
They get into why the two have such different names, the different use cases for them, what the SRE Agent's usage-based pricing might really cost, and the guardrails and network setup that decide how far you can trust an agent with production. Can I start sending agents to conferences to speak for me? Probably. There's a guy who does these, he posts like the AI interviewers, and he just messes with them completely. I don't know that I've seen that.
I'll have to find some and put them in the show notes or things like that. But he'll hop on with somebody and they'll go like, what have you done? He's like, I'm Chick Bongo, and I'm a salesperson, blah, blah, blah. He had one where he brought an AI agent to, and these are all like video, like real-time, like video agents kinds of things, like IVR++ when you're on a phone tree. But he brought a virtual agent to an interview, so he was being interviewed by a virtual agent.
He brought his own virtual agent, and then he was there. And then the two agents just started talking to each other, and his virtual agent, like the one that was interviewing, said, what salary are you looking for? And the salary range of the job is $120,000 to $150,000 US, something like that. And his virtual agent, who's like his representative, who's supposed to be negotiating on his behalf, goes $60,000. It's a great starting point. And he's like, no, no, no, you can't do that. That's funny. I'll pull some together, you'll get a kick out of them.
I should actually go speak at conferences and let my agents do my work for me. I was going to say, you should probably flip it or reverse it if you could, while you still can. Or, you go to the conference, you just hang out, and you put your little virtual agent up on the screen or whatever. It's all good. Like, you know, they can scan you, it'll look just like you, it'll sound like you-ish, maybe. Can you make it talk Upper Peninsula? That would be pretty sweet. Well, I sent you the one because Riverside let you do that, right?
I sent you the, did I send you the Riverside agent or the Riverside, I can't remember what they call it in Riverside, where it can take all of our recordings and create our own virtual agents, essentially, our own virtual people. And you can just type in what you want it to say and post to social media. No. It was kind of creepy. I showed it to my family. I showed it to my kids for an example of, this is why you don't just trust anybody calling anytime.
Because this is not something I said, this is something Riverside made from recordings of me. And I just typed in what I wanted it to say. Love it. You can absolutely do it. It was, it did a really good job. It, like, I could tell, my wife could tell it wasn't me. Just some of the, the mannerisms, the phrasing of how it, how it articulated certain things. Like, there were still, still signs, but it was, it was kind of scarily good.
If you weren't paying attention super close or weren't looking for it, it could easily be interpreted as actually being me. Okay. Yeah. I'll have to go make one. I'll send you one. Looking out for, for that one too next, huh? But we are not here to talk about Agent Ben or Agent Scott. No? Unless it's not really us recording the podcast. Unless it's our agents recording the podcast pretending to be us. No, there's a couple new agents in Azure that we were going to talk about. Because, what else would we talk about other than AI?
I'm going to find some topics, Scott. That's my new goal, find a non-AI. Good luck. So I was going to say, like, very, very much, like, good luck to you. Good luck to that. It's a, it's, it's a goal, I guess. It can be a goal. We got some more agents today. So we chatted a little while back. Well, I mean, last time we chatted, we were chatting about just LLM wikis, right? And how we kind of wrap these things and a little bit what you can do with skills and harnesses.
I think you could apply a lot of what we talked about there to what we're going to talk about today. Because really, it's just that manifestation of going beyond single skills and into more of these composite workflows with agents. And, you know, there's still the, I think, the crisis or maybe not a crisis, but like, what's the definition of an agent and how does that all come together? But I think of them as just composition of skills and we can run these things on a schedule or we can interact with them or put them in a harness and have them run to some kind of outcome.
But we're going to talk about two other agents today. So we also talked in the past about project perception and kind of what comes with some of the agent side with kind of red teams, blue teams, green teams, things like that. That was a little bit more Defender-based, like M365, Microsoft Defender, Security World. Today, we're going to talk about Azure mostly and kind of focus on two things. So the Azure SRE agent and the Azure Copilot troubleshooter or the Azure Copilot troubleshooting agent, I guess is the other way you could think about it.
Can I ask a question? Why is it not the Azure Copilot SRE agent? Why is it the Azure SRE agent and the Azure Copilot troubleshooting agent? I feel like there's some naming mismatch there. I think somebody, you know, just had to think about ways to break them up. So I don't know if this is the right mental model, but I'll give you a little bit of my mental model for these two things. Okay. My mental model for Azure SRE agent and the Azure Copilot troubleshooting agent is you can think of the troubleshooting agent as the free version of the SRE agent.
So there's a set of capabilities that kind of exist across both of them. But the thing you have to keep in mind is with the, and I hate to get into licensing already, but with the Azure SRE agent, it is a consumption-based offering. So it's paid versus the troubleshooting agent, which is effectively your lowest friction entry point, no additional cost. You know, you can just walk right in the front door kind of thing along the way. So you've got this simple troubleshooter native built into experiences like the Azure portal.
It's ready to go for you. Or you've got this other thing, SRE agent, which kind of does the same things the troubleshooting agent does, but it actually does more than them. So it's, like I said, it's consumption-based. It's a paid platform. It's an operational platform. So where the troubleshooting agent is coming in and that's just you asking a question like, hey, I think this network security group rule, this NSG rule is broken, can you help me fix it? Versus SRE agent being more like operational platform side of things.
It's got its own managed identity, its own operational context. It can run CLI in human-in-the-loop, basically like a review mode, or it can go fully autonomous for you. So it's these degrees of maturity along the way. I don't know why they couldn't just call it all SRE agent and find a way to bundle that experience. Like that is kind of weird and bifurcated today. And even from the point of where I started putting the note together for this one a couple weeks ago,
a month, two months ago, whenever it was, the experience has changed. Like SRE agents started in the Azure portal and now it's got its own website and its own experience. We should probably talk about this at some point too, how product groups are moving away from the Azure portal as their grounding experience for their Azure products. That's kind of a weird one to me too. But yeah, you basically have these two agents. They're named two different things, but my mental model is they're all kind of the same thing.
You're basically looking at here's a set of free capabilities versus here's a set of metered or paid capabilities. Does that help? That helps. It almost looks like, too, as I was looking at this, that the SRE agent, and you can correct me if I'm wrong, it looks like it pulls in more context than troubleshooting. Like troubleshooting is relatively basic. Like you said, go look at this firewall and figure out what's wrong with my network path or something like that,
where the SRE agent is looking at diagnostic logging, and it can look at log analytics, and it can look at Defender. And it's pulling in a lot more of your context to give you a much more, I would say like a much more holistic or complete using a much broader set of data than maybe this troubleshooting agent is. I don't know if this analogy lands or this way of framing it either. One is the troubleshooting agent, the Azure Copilot troubleshooting agent. I would argue that's the more reactive one.
That's like there to help you solve something point in time when it goes wrong. So very human in the loop, like I'm going to use this as an extension of me kind of thing. It doesn't even have its own identity. Its identity is you. So that Azure troubleshooting agent can help you solve a case. I have this point in time issue, and I want to go solve this. Can it help you do that? Yes. The other side of that is the Azure SRE agent. And like you said, it can be autonomous, and it knows more about your environment just based on the context and the permissions and having its own identity and things.
So versus troubleshooting agent where I'm saying, hey, this is an extension of me. Azure SRE agent is an extension of your team, right? It's another identity that's out there. So one is being reactive, help me solve this case, or help me solve this problem right now. The other one is, it's just an operator. It's a member of your team. Granted, it's a member of your team who is on a spectrum of autonomy, like how much you grant it and what you give it and what you trust it with.
So there's definitely a whole trust angle and trust boundary thing that comes into play there. But arguably, like the latter, the Azure SRE agent, far more actually an agent than the troubleshooting agent, which is more, I guess, akin to a skill, right? I'm going to go fire this thing up point in time. It's got some grounded knowledge about Azure. It knows how to, in context of me, go look at an NSG rule, a firewall rule, something like that. So again, it's all on the spectrum.
It'd be way easier if they just named it Azure SRE agent. And you said, here's Azure SRE agent and the set of clearly human in the loop capabilities. And here's Azure SRE agent, the autonomous side of things, rather than having to say troubleshooting agent and SRE agent when it's all very much kind of in the same vein. Yeah. And the troubleshooting agent to get started with that is super simple, too. Like you said, you just go into Azure. You can open up Copilot there in the upper menu bar by your search.
And troubleshooting agent just shows up there, pop it open, and just start chatting with it. And it gives you a few sample prompts, can't connect to my virtual machine, help me understand this Azure error, what could be causing performance issues. Like that's really all there is to it. Versus like you said, the SRE agent has an entirely separate site now, but also a significantly, there's a little bit more to do to get it set up and things to pay for. There is. I mean, it's like any other fully agentic thing.
It's got its own identity, and that means you have to onboard it basically like you would onboard a person. You're going to go and give it a set of permissions and capabilities within your environment, all that good stuff. Oh, and you can create agents within the SRE agent. Or is it, no, it's just creating my first SRE agent. So I have not gone through the onboarding yet. So I just went to the website, and I can get my 30-day trial at sre.azure.com. And then it's create my first Azure SRE agent.
Another weird one. Like I said, you're into this operational flow of, I'm going to start in the portal, and it's immediately going to kick me out to another place. And it looks like I can create multiple SRE agents, which I guess makes sense because you're going in and connecting them to different data sources. You can put them in different subscriptions for different billing models because it is the pay-per-go. If you've played with security Copilot, like this is very security Copilot feeling with opening a new portal and having billing and then going in and doing all the configuration within that separate portal, where Azure is really just a way to bill based on usage more so than anything else.
Yeah, so I think from the thing, like you said, it's like other agents that are out there. I think that experience is fairly consistent. You have to give these things an identity. There is a consumption model there for billing, so you've got to go bootstrap it and stand all that up. And as you're doing that, like I said, you should think about how you bootstrap these things as bootstrapping members of your team. So in your case, if you were going to onboard a new contractor, effectively that Azure SRE agent, it's a new contractor for you to come in and take a look at your environment and be an extension of you and the things you do.
So have you looked at, because this always gets me with these, the pricing and billing for these yet. I was looking at it a little bit. And again, I'm going to draw parallels to security Copilot because that's what I've used. Where security Copilot, we had security compute units that were what you paid for that were always on, SCUs. The Azure SRE agent uses AAUs or Azure agent units, which is a standardized measure of agentic processing used across all pre-built Azure agents, except for apparently security Copilot.
Well, is security Copilot an Azure agent, or is it a security Copilot agent, or is it part of Defender? That's not even what defines an agent. That's just like what org ships this thing. Yeah. Which part of the world, yeah, which part of the org, like I said, is the thing that pushes this one through. Yes. So we do have a new measure of agentic processing, and then it goes through all of these.
I mean, I looked at this. I don't even know how you'd start pricing something like this out. Like, there's scenarios on here for quick questions, and then you have input tokens and output tokens and cash read, cash write. Whether you're using Claude Opus or GPT, a whole bunch of math that at the end of the day, it feels like they say just spin it up and monitor your cost. And that's how you price this out is you start seeing how much something costs you.
Because they don't even, I don't even see dollar amounts on here. They just give you guidelines of like how many tokens each of these are going to take and what your billing impact is, how to set spending limits, how to monitor your usage. But there is not a single dollar amount on the entire pricing and billing page. No, this is definitely one of those that you need to go and kind of POC. So you've got to get out there and say, okay, hey, I'm going to run this through a couple of workflows and see what it does.
And it's kind of a crappy answer, but I think it is part of the state we're at, too. Like, it's very hard to predict when you stand these things up. Number of input tokens, number of output tokens, things like that. Especially considering like the models that you can run behind them between things like Opus and GPT and things like that. Because you might have a, given your environment, a better experience even with one model over another. Just based on like the grounded system prompts and things that are out there.
So it is kind of pay-per-use, input tokens, output tokens, and a little harder to figure out in that world. So you've got to go kind of POC it and see if it's there. Arguably not for everybody. Like, if you're looking for the, I know this thing is going to cost me $30 a month and it's going to be fixed costs. Like, this just wouldn't be the offering for you. I would also argue that that's kind of hard to find these days. That doesn't exist. I even see, like we talked about this a while ago, too, right?
The shift in kind of like billing and consulting and things like that. Like, folks bill around outcomes, not necessarily hours. So, you know, how much does it cost you to get to that outcome is a variable thing depending on the issue that you're trying to solve for. Scott, while you were talking about that, I did go just type into Claude. Don't tell anybody. I just asked how much an AAU is for the Azure SRE agent and it did give back some numbers. So again, precursor, this is just Claude spitting it back at me.
Was that it's $0.10 per AAU pay-as-you-go rates. Microsoft, similar to Security Copilot, recommends four AAUs, so $0.40 per hour, just base cost to have it up and running. It's $0.10 per agent per month, which works out to about $300 per agent per month, just to sit there. So this is not actually doing any work. This is like that based cost of sitting there. But then, I mean, expensive, maybe?
I mean, you compare what you pay for Claude and some of that. Like it's not, that is not the obscene cost to me that like Security Copilot was. Security Copilot was a little over the top. This feels attainable by most organizations. I think given some of the things it does, especially, so the way Microsoft describes some of these things, like they'll walk you through where they sit on that spectrum and what they do, right? So I think troubleshooting Copilot, or sorry, Azure Copilot troubleshooting agent,
like ACTA, you know, needs a, has a pretty definitive workflow for it. So I have a problem, I'm a human operator, I step in, I describe my problem. That problem's going to be within a given scope. Typically a resource, but it could be at a higher level, like a resource group or subscription. So that's what I mean. When I say scope there, it's going to diagnose at whatever scope it has access to. It'll recommend a fix. And if a fix can't be recommended, it'll let you escalate it and create a support request.
Like, fairly simple, straightforward, covers, you know, VMs, VM scale sets, AKS, all that kind of stuff is there. The SRE agent, like, way richer than that. So rather than me coming in and saying, hey, I'm trying to solve a case, it's kind of out there always looking, or it can be always looking. So that means it can look at things like, you mentioned diagnostic logs. Like, Azure Monitor Resource Logs, Diagnostic Logs, App Insights.
If you have Log Analytics spun up, it can run Kusto queries for you against that Log Analytics workspace. It's integrated with Azure Monitor Alerts, all those kinds of things. It can also monitor GitHub repos or Azure DevOps repos. So that's interesting. If you think about, like, N10, CI, CD, and having this agent be able to, like, troubleshoot as you're doing deployments, things like that, it can step in and do things like that along the way. And then, you know, it's got all sorts of, like, lifecycle hooks and other things that are there.
So in that vein of, like, it does more, to your point of $300, like, how much do you pay an MSP to come in and have some level of access to your environment? Most folks are paying more than $300 for, you know, a full MSP to be sitting there, kind of ready to go. So that ROI could be there if it can kind of inflict some of the value that you're looking for within your organization. Absolutely.
And I was looking at some of this pricing. This pulled up from a third-party article, too, that compares a whole bunch of different SRE agents. So take some of that pricing stuff with a grain of salt, because it was not pulling all from 100%. 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. No, this is... I'm with you. That pricing is relatively inexpensive.
It feels like it can do a lot and... On one hand, it sounds expensive when you say it. On the other, it's actually like really aggressive when you talk about potentially tearing down a whole industry that sits behind it, should these things take off in the way that they're being positioned. It's a double-edged sword, right? The agent doth come for us all. It'll also be interesting to see what this will do. I would be interested to spin this up, and I may end up doing this for a month or two in my tenant, because it does... To your point, it talks about...
You're looking at things like diagnostic logs, like at Log Analytics. It talks in the Microsoft articles about incident response plans or other automated triggers. How close can you get to even some of the security Copilot stuff with the SRE agent in terms of... If it's just Azure you're talking about, I think you can get darn close, because you can even inject your own custom skills and things like that in there. So it can run Azure CLI. It can execute Python scripts. It's all like, what level of things do you give to it?
So it'd be interesting to maybe take a step back. I'd have to go look at and run it through. I was certainly doing it for this one. But one of the things that I find helpful is when you're trying to compare these things, like they all sit on a spectrum, is just what's the operating model for them? Or what's the boundary of the operating model? So we talked about the troubleshooting agent is really trigger, what scope am I at, diagnose, resolve, and if it's not resolved, escalate. It's simple, right? Five steps that are there.
The Azure SRE agent can do the observability part. So rather than trigger, it's observe. So I'm not triggering it, it's out there observing for me. It can correlate, it can hypothesize, which is another interesting thing, like that's a different mental model. And then it can propose, or it can execute, or it can do both, depending on what you've chosen to do. It can actually verify what it did, and then it retains that context for the next time. So should that issue or that particular thing crop up again, it knows, knows, big air quotes, it knows.
But it can at least go back and re-eval in context and get it going. So you think about that operating model of observe, correlate, hypothesize, verify, retain context, like, yeah, the security agent was kind of there. The perception stuff was totally different, right? That was a mix of agents that were coming through and kind of running up. So then maybe the other thing to think about is, could you wire all this up yourself? Sure. Like, this is a set of connectors, right? It's MCPs and CLIs, and at the end of the day, it's skills that are being composed as an agent,
and kind of an agent in a box to you as a consumer of it. Could you build all that yourself? Sure. How long would it take you? Like, how many tokens would you burn to get there? Maybe you can do it for, let's say, $100, but then you run it. Does it only cost you $300 a month to run? Or does it cost you more than that? It all depends. It's very hard and very fuzzy to figure out. But if you're like, I don't think Azure SRE agent is the type of thing that's there for,
you know, you have a super well-known, super constrained workload. Like, it's all on Rails. And I think for a lot of that stuff, the troubleshooting agent's fine, right? You stand it up, you deploy it, infrastructure is code, you know what all your firewall rules and NSG rules and your VMSS configuration and everything is from the start. So you just fire that out. It's there for when something probably more goes wrong on the Azure side than on your side kind of things. Versus the other, I think the Azure SRE agent would be way better fit
for like a mixed environment. I run lots of workloads in here. I have multiple teams doing deployments. They all need to kind of live within the same boundary, but that boundary is a little fuzzy sometimes, right? When I say like, hey, go ahead and deploy into my fully governed landing zone subscription over here. And it's got its own set of constraints that come with it. That can be a little bit harder for teams to kind of figure out and stand up and get going. Even the fact that the troubleshooting agent, and they highlight it in some of the documentation,
is one of the secondary steps of the troubleshooting agent is it can go open a support ticket for you. Troubleshooting agent is a little bit more like that first line of, it's almost like a first line of tech support, kind of like you said with SRE, it's much more operationalized, a member of your team, that type of thing, which is probably it's going to use more compute. That's why you're going to pay for it more. But I mean, let's be honest, the troubleshooting agent probably is going to help Microsoft lessen the number of support tickets that come in as their hope,
because it's able to help people solve issues, yes, before that support ticket hits a Microsoft support desk. I mean, at the end of the day, it boils down to ad hoc troubleshooting, troubleshooting agent, versus repeatable incident response SRE agent. You know, you got to kind of like pick the mental model that works for your org, and what you're trying to solve for along the way. You do have these kind of options that sit in front of you. Very cool.
I was looking through more of the connecting sources and spinning up agents and onboarding teams into it. There's a whole bunch in there that it can do. There's a bunch of learned documentation that we'll have links to in the show notes as well for, here's how you even go start setting up this SRE agent. Again, very similar to how you would set up security Copilot, that type of thing in terms of connecting data sources and setting up activities and skills,
and you can connect MCP servers to it and all of that. It would be interesting to me, I'm going to harp on this over and over again, Scott, if they could consolidate these. Like, I don't want, if I'm a large organization running security Copilot, I want to run the Azure SRE agent. I can see an argument for not having this stuff in the Azure portal, for having a separate agent or a separate AI website where all these things are. I don't want 10 different websites
where all these things are. I want, like, whether it's, you can somehow fold all this into copilot.microsoft, or you can have agents.microsoft, something like that where me as a end user can go and I can go connect to these different agents, have these different skills. Wouldn't it be great if it was just part of, like, maybe, like, that whole agent 365 thing or something, right? Like, single pane of glass,
here's all my agents, here's what they do, all that, yeah. Welcome to, at some point, you're always going to see somebody ships the org. And, hey, here's the org chart manifesting out in the real world. 100%. What is going on here? We'll see. Azure just had a big reorg last week, so now it's all under one banner. One banner? I've started seeing emails, too, about my pricing changing in Azure. Your pricing? My pricing. That's another topic we can talk about another day.
Speaking of Azure, I just popped up because you said Azure reorg, and I was like, Azure reorg and Azure pricing and all the Azure things. Yeah, another topic. Another day. There was some new Copilot stuff and OneDrive stuff. Oh, yeah, we have more topics we can talk about. I'm not going to make any promises. Yeah, more topics. But we should close this one out. So, let's see, what else with this thing? So, SRE agent, much richer, even if you're not going to deploy these things. So, to your point of, could you build the skill yourself and could you run it and do it?
Absolutely. I do think it's helpful to look at how Microsoft or even AWS GCP as they do their agents for their clouds kind of thing manifest is what are the permissions that come with them? What can they do? What can they not do? So, if you look at Azure SRE agent, it's got some interesting guardrails. It can't delete or remove anything just out of the gate. That's all blocked. It can't get at your key faults. All those commands are blocked. So, when I'm saying blocked, they're blocked in the CLI.
Agent can totally figure out how to run a REST request if you want it to kind of thing. But at least from the CLI that's integrated, delete and remove blocked, easy key vault blocked, respects your management locks or your resource locks. So, hey, this resource group has a do not delete lock on it kind of thing. It'll validate your subscription IDs. It always logs all the actions and results of CLI things it takes. So, even if they said, even if you're not going to use it, you're going to go build your own skill,
it'd be great to pump these rule sets into your own things, right? Give them a little bit of guardrail and make them a little bit more constrained out of the gate. Oh, this is interesting too, even with VNets. Operations agents that cannot reach private resources. VNets are another weird one. So, that was like an extra capability that came. That is, because it looks like the SRE agent can always limit it to only public management plane signals. It can't go through private resources or private things that are just private links on a VNet.
I'm reading that right? Which is funny, because depending on a resource, I'm colored because I live in storage land. So, a storage account is never really private until you go into the firewall and crank it down. It's always got a public endpoint, even if you stand up a PE on it. And there are other resources out there like that too. So, yeah. So, the VNet stuff was an add-on. It was like a GA after GA capability, I guess, that came to things. So, by default, yes, it does look to just the
public management plane or control plane signals from Azure Resource Manager. And with this VNet integration, you can start to bring it into VNet mode. And then it can reach your private endpoints. And it can even reach out to on-premises systems. So, if you're hooked up through like a site-to-site VPN or ExpressRoute and your policies and routing allow for that, you can get going with some of that stuff.
It does require, it does have some own networking requirements on its own side. Maybe similar to like Bastion when you stand that up. Like, it needs its own network segment to stand in. So, you need a slash 27 or larger subnet that you can delegate out there. Subnet has to be in the same region as the agent that's out there. It's also egress only. So, the agent talks out, but you can't do some kind of like private inbound connectivity from that VNet
to that agent kind of thing. But it lets you do some, the biggest thing for me would be like internal package repos or like private package repos. So, lots of orgs block like public NPM. You might publish like NPM packages that are vetted. Same thing with like PyPy. You know, Python being so popular with these agents. You don't want like a weird like man in the middle or like supply chain attacks. So, you might have your own PyPy registries, things like that that are there. So, it lets you kind of navigate that entire system
on the network side of things. Ooh, that's another topic, Scott. Add this one to the list. Supply chain attacks in Microsoft 365. Did you see the news last week about that? Different topic. We're not going to talk about it here. I did not. We should talk about this here on a future episode. For now, we can wrap this one up. I like that we're coming up with ideas. So, folks should go check these out. I think, like I said, the low-hanging fruit of getting started with this stuff is that Azure Copilot troubleshooting agent. It's kind of like a no harm, no foul.
It's there for you. It's in the portal for everybody. You might want to try it out because if it helps, it helps. And if it can deflect a support ticket, that saves you money. So, good. I would say go try out the SRE1 too. Like, again, 300 bikes for a month. If you don't even spin it up for a month, like, this is one of those. It does have a trial that you can get started with. I have a 30-day trial, but you can also, similar, I've done this with Security Copilot, which is why I'm bringing it up. You can go spin it up because you're spinning up these AAUs and even try it
for two or three days. And then just delete it. Like, it's a resource that you can just create and delete. If you want to try it out for two or three days, it costs you, what, 10 or 20 bucks to try it out for a couple days. See how you might be able to use it in your own environment and let us know if you've used it, what your experience has been with the SRE agent. Absolutely. With that, are we going to go attack our days or let agents attack our day for us? We're going to do something. Do something like that. All right. Well, thanks again,
everybody. Again, let us know SRE agents, LinkedIn, probably the best way to let us know how you're using these, how you want to try them out. And Scott, if people are going to be in Dallas, I'll be in Dallas the beginning of November for M365 conference down in Dallas. And I'll be out in Scottsdale in January for the Workplace Ninjas conference. Back to some IT Pro cloud show on the road. Oh, folks should go check that out. Those were all posted up on LinkedIn
and things like that. We'd love to get your feedback there as Ben's out traveling the country. Do you want to see more live streams when Ben's running around our defense? He did get that nice board behind him printed up. Yeah. Yes. I have to think Jay Leask did most of the work on that one. That was a Jay Leask initiative, but it did turn out nice. We had fun with that. Jay and I were at that conference in Branson last week and did a whole bunch of interviews.
I might try to do some of those in Dallas. Jay and I will be doing it again in Scottsdale. So like you said, Scott, go check those videos out on YouTube. Let us know what you want to hear. And then Scott, we need to get you at a conference. Maybe, I don't know where. We got to get to a conference. We got to get you to a conference. The problem, yeah, what I've hit. I think I might submit for one of the live 360 events in August. They're going to be in Redmond, so it's an interesting way for me to get out to campus. Maybe I'll see if I can pitch
a couple sessions around Azure storage development or something like that. Yeah, we'll figure out one of these conferences. Yeah, we'll figure one out, Scott, so you and I can do it with a fancy banner behind us. Yeah, we'll just have to run one here in Jacksonville. We'll have everybody come to us. We can just do a live, our own live. All right. All right. Well, thanks for chatting agents with me today. I know it's more AI beat down, but it's what's in front of us. It's what's out there now. Yes, absolutely. You're welcome. Enjoy your week and we'll talk to you later.
Thanks, Ben. 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 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...
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 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...