Latest / Microsoft Cloud IT Pro Podcast / Episode 403 – Catching up on all the news from Microsoft Build 2025
Transcript
- 0:03Welcome to episode 403
- 0:05of the Microsoft Cloud IT Pro podcast
- 0:08recorded live 05/30/2025.
- 0:12This is a show about Microsoft three sixty
- 0:14five and Azure from the perspective of IT
- 0:16pros and end users, where we discuss a
- 0:18topic or recent news and how it relates
- 0:21to you. Microsoft build has come and gone.
- 0:24So today, we dive into some of the
- 0:26announcements from the conference.
- 0:27We'll catch you up on announcements around Microsoft
- 0:30three sixty five Copilot tuning,
- 0:33securing Copilot agents and enter ID, and a
- 0:36can't miss build session with Mark Russinovich and
- 0:39more. So let's dive into the show.
- 0:44It's been a week, Scott, for both of
- 0:46us. Time marches on. I I have not
- 0:48told you about my latest adventure. Your latest
- 0:51adventure? Not your latest toy, but your latest
- 0:53adventure? My latest adventure. Scott story time. Scott
- 0:56story time. So my latest adventure was
- 0:58my oldest graduated high school Congratulations. In the
- 1:01last couple weeks. Yeah. Thanks. Accomplishment. We will
- 1:03see. We need to get him out of
- 1:04the house still. As part of that, we
- 1:06threw him a graduation party. So we had
- 1:08people over to the house, and Yep. It's
- 1:10hot outside, but you can still hang out
- 1:12outside if you're in the shade.
- 1:14So we bought these
- 1:15pop up tents, like, 12 by 12 tents
- 1:18without sides. They just had a roof on
- 1:19them. Right. Like, the whole canopy thing, yeah,
- 1:21you see them at all the sporting event.
- 1:23Yeah. All that. Simple basic. We we we
- 1:25found some at a a bargain bin store.
- 1:28Like,
- 1:29there's a store in Jacksonville where this stuff
- 1:31just comes in on crates, and then they
- 1:33just mark it down to clear it all
- 1:35out. So it was whatever. They're they're normally
- 1:38a hundred $50,
- 1:39and so they were on sale for $60.70.
- 1:42I don't know. My wife went and grabbed
- 1:44a couple of them. I will say this
- 1:45is one thing I like about Florida because
- 1:47nobody has basements, nobody has places to store
- 1:49anything, so everybody gets rid of stuff if
- 1:50they don't need it. And I feel like
- 1:52because of that, you can find good deals
- 1:53and cheap stuff. Well, this was even brand
- 1:55new because this is, like, a wholesaler.
- 1:57So they're buying what's kinda like Amazon return
- 2:00bin kinda thing. They're buying a big
- 2:0340 foot shipping crate. They don't know what's
- 2:05in it, and then they're just bringing it
- 2:06into their store, and they're trying to clear
- 2:07it out and sell it. So so they
- 2:09must have had a pallet of these These
- 2:11canopies. Ready to go. Yeah. These canopies. So
- 2:14we bought a couple of them and stood
- 2:16them up for the day of the party,
- 2:17and it ended up raining.
- 2:19And so
- 2:20we ended up inside for a little bit,
- 2:22in, out, in, out. But I left them
- 2:24up after the party. So this was last
- 2:26week, and then we went through this week,
- 2:28and
- 2:29we've had a couple good rainstorms
- 2:32and squalls and things like that. So they're
- 2:34still up? You left them up? I did.
- 2:36I left them up. This wasn't, you know,
- 2:37hurricane stuff. I wasn't scared of something, like,
- 2:40blowing away and landing on a neighbor's house,
- 2:42things like that. So day one, they went
- 2:45up, party. It rained. It was it was
- 2:48fine. Like, all the water shed off the
- 2:50roof. Like, it was taut enough and and
- 2:52tight enough that
- 2:53all that stuff happened.
- 2:55At some point, it must have just loosened
- 2:58up or some of the some of the
- 2:59Velcro must have slipped or something that was
- 3:01holding that canopy onto
- 3:04onto the legs. So I went out there,
- 3:06one day after one of the rainstorms,
- 3:08and one of them just had this big
- 3:10bubble in a corner where it was all
- 3:12just leaning down.
- 3:14And we've got the lawn furniture and stuff
- 3:16underneath it, like, the patio furniture, and it
- 3:18was leaning down about to my head. So
- 3:20this is maybe a tent that's about eight
- 3:22feet high, and
- 3:23I'm I'm five seven. So it's coming down,
- 3:26like, two, two and a half feet and
- 3:28just about touching my head with this big
- 3:30bubble. And I'm like, oh, that's no good.
- 3:31So I'm in there and pushing it up
- 3:33and trying to get the water off it.
- 3:34And I take a look, and I'm like,
- 3:36oh, it came loose on on the side
- 3:38because the Velcro slipped. So I kinda, like,
- 3:41redid the Velcro. I went and pulled it
- 3:43all back down and tightened it back up.
- 3:45Fine. Great. Two more days go by. Rain.
- 3:48Rain. Okay.
- 3:50Yesterday,
- 3:50we had some massive kinda squalls come through,
- 3:54like sideways rain. I saw that. It was
- 3:57a little nutty. It it came down for
- 3:59basically
- 4:00two hours straight sideways.
- 4:02I was watching a little river form between
- 4:05my house and and my neighbor's house because
- 4:07there's kind of a natural divot there, and
- 4:09and that leads down to the preserve behind
- 4:11my house.
- 4:12And so I'm I'm looking. I'm looking. So
- 4:15I go outside, and I just I just
- 4:16open the door. I'm not I'm not going
- 4:18out in this weather, and I peek my
- 4:19head over, and there's not just one bubble
- 4:21coming down. Now there's two bubbles, and the
- 4:24whole thing is, like, leaning, leaning, leaning, leaning.
- 4:26And it's coming down, and it's almost touching
- 4:27the patio furniture. Oh, man. Patio furniture's
- 4:30only, what, three feet maybe? Right. Considering, like,
- 4:33the top of the back of the chair,
- 4:34three, four feet. So it's coming, like, just
- 4:36down, down, down. I'm, like, I'm telling my
- 4:38wife, I said, this might be it. Like,
- 4:39it it it's it's it's probably gonna go.
- 4:42Absolutely did. So it got so heavy with
- 4:44the rain
- 4:45on with those two bubbles on the one
- 4:47side, it actually tore
- 4:49the the entire top. Like, it just sheared
- 4:52itself, like, right off. And then when it
- 4:54sheared itself with the weight, it also bent
- 4:57all all the metal and things because this
- 4:59is like a big accordion. So now this
- 5:01thing will not go back together. It's probably
- 5:03the best $70 I've spent for, like, a
- 5:05week of just being able to sit outside
- 5:07and things like that. I was kinda telling
- 5:08my wife. It's like, you know what? Worth
- 5:10it. Maybe I should just go back and
- 5:11see if we can buy a couple more.
- 5:13Like, we'll we'll just keep them on the
- 5:15side of the house in in stock and
- 5:17and and ready to go for the next
- 5:19one. So that that that's my next adventure
- 5:21is after I'm off this call, I'm gonna
- 5:23go grab a hacksaw, and I'm gonna go
- 5:25out in the backyard. And now I gotta
- 5:26cut all these metal spider legs off this
- 5:2912 by 12 monstrosity thing because there's no
- 5:32way it's going back together with how bent
- 5:34it is and everything else. And then because
- 5:37it sat there over my patio furniture so
- 5:39normally in my patio furniture, it sits outside
- 5:41and we leave the cushions on sometimes, and
- 5:43it doesn't get soaked soaked. It's got, like,
- 5:46Scotchgard or whatever on it. The water all
- 5:47runs off. Uh-huh. Because this sat on it
- 5:49for so long, it soaked through the material
- 5:52on the tent, and it soaked into all
- 5:53the patio cushions. So when I picked the
- 5:55patio cushions up and I just turned them
- 5:57from horizontal to vertical, they've just been leaking
- 5:59out for the past hour. Like, yep. Still
- 6:01watching them drip. I will not lie. I
- 6:03thought for sure one of those tents was
- 6:04gonna, like, take off and blow away. I
- 6:06thought that's what was gonna happen. Because when
- 6:08it starts raining sideways, I've seen I think
- 6:11it was actually a hurricane. I've seen a
- 6:12trampoline, like, blow from one house into the
- 6:16neighbor's yard in some of the wind that
- 6:18we've had. It had a great, like, vent
- 6:20on the top that Okay. Would crosscut, so
- 6:22it really wasn't gonna take off or go
- 6:24up and down. And we had it all
- 6:25tied down and everything. I I didn't use,
- 6:27like, all the hurricane ties, but it it
- 6:30was definitely, like, attached to the house and
- 6:33and where it needed to be. So fun
- 6:35fun little adventure for the week. Wow. Way
- 6:37more adventurous than my week. I don't think
- 6:39I have any fun adventure stories. So between
- 6:41that and and my new terminal toy, which
- 6:43we'll we'll talk about later. We'll we'll do
- 6:45a toy episode and and annoy folks
- 6:47sometime in the future. But, anyway, that's my
- 6:50news for the week. Why don't we get
- 6:52into the news from the last couple weeks
- 6:55because we've had events like build
- 6:57and all sorts of other stuff. Yeah. And
- 6:59some of those events, they're definitely developer heavy.
- 7:03Like, it's all different things. But there's
- 7:06always news, I feel like, that relates to
- 7:08IT pros because us as IT pros have
- 7:10to manage our developers and keep them in
- 7:13line and make sure they're behaving. And as
- 7:15a result, we get different things, and let's
- 7:18face it, there's Copilot stuff everywhere, so there's
- 7:21always Copilot news to talk about as well.
- 7:23And we haven't done much of a news
- 7:24episode, so some of these I think everything
- 7:27we kind of put together was the last
- 7:28couple weeks. Some of it may have been
- 7:29a little before build, all around the build
- 7:32time frame. We really should do these more
- 7:34often. Although, I don't feel like there's been
- 7:36a ton of news either, or I've just
- 7:38missed it. There's been quite a bit. I
- 7:40think it depends It's kinda what you're looking
- 7:42for. Where you sit and where you play
- 7:44around. Yeah. Yeah. There's been a bunch of
- 7:46new stuff come out across
- 7:49M 365.
- 7:51I was looking the other day
- 7:53at some of the
- 7:54latest announcements that have come out for
- 7:58Azure Kubernetes service and and some of the
- 8:00things going on in AKS land.
- 8:02Tons going on over there.
- 8:05The the trick is kinda sifting through the
- 8:09AI, AI, AI to what's what's the what's
- 8:12the real thing, or if AI was your
- 8:15thing, then, hey, maybe there was something real
- 8:18for you there too. Like, I definitely saw
- 8:20some stuff with Copilot
- 8:22that piqued my interest, and and maybe we
- 8:24could start there. So
- 8:26last week, we we talked to Do you
- 8:28wanna start with that one? Yeah. So so
- 8:30last week, we chatted with AEC
- 8:32about declarative agents, and that was kind of
- 8:34a follow-up to some stuff that we had
- 8:36talked about previously with Copilot Studio and doing
- 8:39kind of
- 8:40well, for you and I, no code declarative
- 8:43agents, just in Copilot Studio,
- 8:46next, next, next in context of an m
- 8:48$3.65
- 8:49subscription, things like that.
- 8:51They're introducing
- 8:52fine tuning for Copilot,
- 8:55which kinda takes this a whole another level
- 8:57because some of the at at least from
- 8:59my perspective and the things that I do
- 9:01in my day to day, because what I
- 9:03do today
- 9:04in something like Copilot Studio with declarative agents
- 9:07is I'm relying on RAG and and retrieval
- 9:10augmented generation.
- 9:12And this is really being able to take
- 9:15a model that already exists and the weights
- 9:17in that model that are already deployed out
- 9:18there and exist, and then fine tune your
- 9:21own model on top of it
- 9:23or fine tune and refine that model so
- 9:25that you can drive
- 9:27specific business processes.
- 9:30And the demo I saw around this really
- 9:33got my head turning. So I'm a product
- 9:35manager.
- 9:36I live in a land where
- 9:39I'm a I'm a remote employee.
- 9:41Communication and particularly written communication is super important.
- 9:44So we spend a lot of time on
- 9:46product requirements documents, PRDs,
- 9:49justifications,
- 9:50specifications,
- 9:51designs,
- 9:52and we have formal templates for all this
- 9:53stuff. One of my big problems with it
- 9:56is
- 9:57people who take the template and they massage
- 9:59it into their own thing or they don't
- 10:01know what a good one is. So they
- 10:02they kinda look at the template that we
- 10:04have today, and and they just get like
- 10:07a deer in headlights and and scared about
- 10:09it.
- 10:09So I like, with this, I could take
- 10:12all our existing
- 10:14PRDs that we've done for, like, at least
- 10:16in my time for the past five years
- 10:18on my team.
- 10:19I could train
- 10:20and fine tune
- 10:22the existing model on my PRDs,
- 10:25and I could actually have it tune on
- 10:28top of my templates as well.
- 10:31So then I could make it super turnkey
- 10:34to go in and turn these things out.
- 10:36And that's a very simple scenario for me.
- 10:39I saw a great demo on the Microsoft
- 10:42Mechanics YouTube channel, and I'll I'll put a
- 10:44link in the show notes, and I recommend
- 10:46folks go out and watch that one. That
- 10:47was generating legal documents,
- 10:50and it actually did a pretty good job.
- 10:52Like, I'm not a lawyer, all all that
- 10:53jazz. Yep. It's just to couch all that
- 10:56and and have that caveat.
- 10:58But that said, I I think that one's
- 11:00really cool and and really exciting, and it's
- 11:02something I'll have access to in my tenant
- 11:05and totally intend to go out and play
- 11:07around with. So I don't know if you
- 11:09had a chance to either play with this
- 11:11in your tenant or watch any of the
- 11:14watch any of the introduction videos or anything
- 11:16like that that's out there, but we would
- 11:18would love to get your thoughts on this
- 11:20one as well. I did watch the Microsoft
- 11:22Mechanics video while I was doing other stuff,
- 11:24so I was kinda half watching it. This
- 11:26is a really cool feature. I like how
- 11:28you can fine tune it that way. I
- 11:30would say the part that I was interested
- 11:32in, kinda coming at this from
- 11:35the security model is in my
- 11:37head too when you see this announcement. You're
- 11:39like, okay. You're now fine tuning
- 11:42an LLM.
- 11:43What does this mean for
- 11:46data security? What does this mean for how
- 11:48is my data now being used if I'm
- 11:50fine tuning
- 11:51an LLM? Is it going into the LLM?
- 11:54Is it training the LLM? And they talked
- 11:56about this in the Microsoft Mechanics video, and,
- 11:58ironically, I don't I probably don't have the
- 12:00business use cases for this yet, although I'm
- 12:03starting to play with it more. But it's
- 12:05more thinking about this from a client perspective
- 12:08of questions they're gonna ask about their data,
- 12:11their data security. Some of the stuff we
- 12:13even talked about a couple weeks ago when
- 12:14I was talking about how a lot of
- 12:16clients are coming to me now and saying
- 12:18we wanna prep for Copilot, we wanna make
- 12:19sure data is secure.
- 12:21Now what does fine tuning this mean? And
- 12:23they did a pretty good job, I would
- 12:25say, in the Microsoft Mechanics video from that
- 12:27perspective of, like, you still have your LLM,
- 12:30and it's not and I I could not
- 12:32explain this from a deep technical level on
- 12:34how it actually works, but they actually take
- 12:36your data. And the way they illustrated it
- 12:38is they, like, attach it to the LLM
- 12:41for the fine tuning. It's not like your
- 12:43data's gonna go into the LLM and they're
- 12:44retraining the core LLM. It's like something between
- 12:48RAG and actually having your own model, where
- 12:51it's a model with your data attached to
- 12:54it to do the fine tuning. And it
- 12:56sounded like from that video that it's not
- 12:57even always attached to the LLM. Like, it
- 13:00attaches to the LLM to do the fine
- 13:02tuning, and then it kinda goes and stands
- 13:05separate from it after you've done the fine
- 13:06tuning. It only reattaches when you do it,
- 13:08and it still maintains, like, your data and
- 13:11your own tenancy.
- 13:13They're not taking your data to train the
- 13:15core LLM.
- 13:17All your access lists and everything, that is
- 13:19all still maintained.
- 13:20So it was that that part to me
- 13:23was interesting to try to understand a little
- 13:25bit more from that perspective
- 13:28on how was it doing the fine tuning
- 13:31of the model without actually
- 13:33absorbing the data into the model? The way
- 13:35you can think about it is
- 13:37fine tuning,
- 13:38and and that's really what we're talking about
- 13:40here is is kinda
- 13:42taking an existing model that's already been trained
- 13:45Yep. Let's say chat GPT four o, and
- 13:49then you're gonna take that, and that that
- 13:50that's a stage in that model's life cycle,
- 13:53and you can use it as it is
- 13:54with the weights that exist and and and
- 13:56all of that stuff. Now what you can
- 13:59then do so you do all that, and
- 14:00and that's kinda like training slash pre training.
- 14:03And then you take that pre trained model
- 14:06and you fine tune it. And fine tuning
- 14:09is a refinement process. So you're taking this
- 14:14very it doesn't have to be very small,
- 14:16but you're taking a smaller
- 14:18task specific
- 14:19dataset. So, like, in my example,
- 14:22PRDs and and justifications. Templates and yeah. You're
- 14:25doing this to
- 14:27optimize
- 14:28performance
- 14:29and
- 14:30and and drive a business
- 14:33process.
- 14:34So, like, you you you're right. You're not
- 14:36loading this stuff into the original
- 14:40pretrained model. You're kind of creating your little
- 14:43bit of, like, an your own offshoot of
- 14:46a model, but you don't have to go
- 14:47back and redo all the training and redo
- 14:49all the weights for the original model. So
- 14:51you've kind of augmented it with this set
- 14:54of sidecar weights that can then be used
- 14:57as well to
- 14:58go ahead and make your responses better.
- 15:02Now the the the purpose of this and
- 15:04and the difference
- 15:05versus,
- 15:07say, something like rag
- 15:09is
- 15:10RAG has to go out and you you
- 15:13have to chunk all these documents, you have
- 15:15to put them in a vector database,
- 15:16and you have to compute and and and
- 15:19run all that stuff every time. This is
- 15:21just having that set of
- 15:23kinda fine tuned weights out there for you
- 15:27and ready to go. So it's it's it's
- 15:29a click stop if you think about it
- 15:30as a series of stages. So pre training
- 15:33is all about, let me get the model
- 15:34out there, and that's
- 15:36general representation
- 15:38of whatever it was trained on. And then
- 15:40fine tuning is let me take and
- 15:43take this model to the next level by
- 15:46really augmenting it in a way I shouldn't
- 15:48say augment because of the whole rag thing,
- 15:50by
- 15:51adapting it. Yeah. Fine tuning it. We're fine
- 15:54tuning it. Right?
- 15:55We're fine tuning it. So you're going but
- 15:57you're going from general representations
- 16:00to specific tasks or Right. Knowledge domains.
- 16:04Processes. Yeah. Processes. So you could do this
- 16:06multiple times. So when you hear about maybe
- 16:09a company that's working on
- 16:11AI to solve cancer, well, guess what? They
- 16:14did the same thing. Right? They they took
- 16:15that general model that was out there, and
- 16:18then they took a bunch of cancer research
- 16:19that had already been done previously, and then
- 16:21they fine tuned that model so that they
- 16:23could have a purpose built model that they
- 16:26could go and and work with
- 16:28to
- 16:29attack cancer, attack your
- 16:31in your case, maybe your business processes
- 16:35around statements of work and how you put
- 16:38those together for clients. I should try one
- 16:40of those. You've done that a whole bunch
- 16:41throughout the years with things like document libraries
- 16:44and document sets, and this is I I
- 16:47I think you should give it a shot,
- 16:49like, if you're licensed for it because it's
- 16:50a little bit of a next level thing
- 16:52for you. I should absolutely go try that.
- 16:53I I didn't statements of work didn't click
- 16:56in my head until we were just talking
- 16:57about it. I'm like, yeah, master service agreements.
- 17:00And maybe it's because I don't put a
- 17:02ton in there. Statements of work, I put
- 17:04a little bit more in there. Master service
- 17:05agreements are a lot of addresses
- 17:07and that. But I have had clients ask
- 17:09can't even remember which client it was. Where
- 17:11they did, they're like, well, if I have
- 17:13a template of a document and I want
- 17:15to use AI to fill out the template,
- 17:19can you do it? And I'm like, well,
- 17:21not really.
- 17:22The way it was
- 17:24two months ago even or before they talked
- 17:26about this fine tuning, it's like, doesn't really
- 17:28do a great job at that. And this
- 17:30is where
- 17:32now I think as clients get this, again,
- 17:34licensed for it, willing to build it out,
- 17:36it's a lot more gets you a lot
- 17:38closer to being able to have templates that
- 17:40you can auto fill out. Not auto fill
- 17:42out, but you still as a user
- 17:45can yeah. Yeah. So so you can generate
- 17:47those
- 17:48and you as a user providing inputs. So
- 17:51the the way I think about it and
- 17:54the way I see a lot of folks
- 17:56use generative
- 17:57AI today
- 17:58is they kind of look at it and
- 18:02and I've definitely seen,
- 18:04particularly, like, my PRD example, I've seen some
- 18:06PMs that go out there, and they just
- 18:08say, hey. Write me the document.
- 18:10And this isn't, hey. Write me the document.
- 18:13This is write me help me write the
- 18:16document so that it conforms to this process,
- 18:19this knowledge domain,
- 18:21any anything like that, you know, that exists
- 18:24out there. So, yeah, it's it's a tool
- 18:26in the toolbox
- 18:28versus a do my job for me kinda
- 18:30thing. So if you take an agent,
- 18:33and I don't know that they talked about
- 18:35this, and I don't know if you've seen
- 18:37it, Could you potentially take like, let's say
- 18:39you go do the co the fine tuning,
- 18:42and, really, you're creating an agent. So that's
- 18:43the other aspect about this. This isn't fine
- 18:45tuning like your general
- 18:47Microsoft three sixty five Copilot responses. You're creating
- 18:50an agent that's fine tuned
- 18:53to do
- 18:54a specific task.
- 18:55Can you combine
- 18:57agents with actions yet where you could do
- 18:59a fine tuned agent going back to the
- 19:02proposals or statements of work where you're then
- 19:04using an action
- 19:05to prompt or to get certain information
- 19:08to feed into a fine tuned agent to
- 19:11generate these particular outputs? I haven't tried to
- 19:14chain them together like that yet. I I
- 19:16I think that's where all this stuff goes.
- 19:17So if you think about
- 19:19MCP servers and the whole model context protocol
- 19:22thing,
- 19:23that's where these agentic things are going. And
- 19:26you see a little bit of this today
- 19:28in
- 19:29in maybe in the declarative agents that we
- 19:31already talked about where there's a connector for
- 19:33SAP, there's a connector for Workday,
- 19:36things like that. Like, you have this agentic
- 19:39system on the side that it can reach
- 19:40out to and sidecar that knowledge in from
- 19:43that other system
- 19:44and bring it in. I always think it's
- 19:46super funny, at least in today's world where
- 19:48we're still
- 19:50very early days in this stuff, when
- 19:52you you see the funny two agents talking
- 19:55to each other kinda thing. I don't I
- 19:56don't know if you've seen the video. It
- 19:59came out a couple months ago, but it
- 20:00was like a robo calling agent called another
- 20:03robo caller. And then they both figured out
- 20:05that they were robo callers, so they just
- 20:07basically started talking to themselves in it sounded
- 20:10like an old, like, bot modem going off.
- 20:12Like I think I did see that. Yes.
- 20:14Back and forth because they knew there was
- 20:16no human there and, hey. Like like, let's
- 20:18go down a different path kinda thing. So
- 20:20they're still a little wild and off the
- 20:22rails. But, ultimately, I think that's where these
- 20:24things do go. So if I think about
- 20:27an
- 20:28agentic chain to build a PRD, well, great.
- 20:31So I've got the fine tuned
- 20:33process where I've come in and said, hey.
- 20:35Help me write this within these guard re
- 20:38these guardrails, or help me generate this thing
- 20:40within these guardrails and and these boundaries.
- 20:43But part of my PRDs would be customer
- 20:46stories.
- 20:47Well, maybe I wanna go build a little
- 20:49agent that does nothing other than reason over
- 20:52my customer feedback.
- 20:54And then there's another agent that I might
- 20:56wanna go build that reasons over
- 20:58the financial aspects of my business and and
- 21:01the cogs of it and and how that
- 21:03comes together. So all these little things have
- 21:05their own
- 21:07click stops along the way, and, eventually, I
- 21:08think, yeah, that that's where you end up
- 21:10is potentially a bunch of agents just chatting
- 21:12back and forth with each other, and one
- 21:14throws this over here, and then it throws
- 21:16it back this, then it throws it into
- 21:17a new one, and out the other end
- 21:20comes something else. And I I think that's
- 21:22the thing that we we all haven't, like,
- 21:25really figured out as Right. I saw the
- 21:27I think it was the CEO of Claude
- 21:29this week, last week. He was doing an
- 21:31interview, and he was talking about basically, like,
- 21:3450%
- 21:35of it was some crazy number, 20%, fifty
- 21:38%. It was really high of kinda low
- 21:40end knowledge worker jobs going away because of
- 21:43AI and and Gen AI. So nobody knows
- 21:46how this stuff's gonna go and where it
- 21:48comes out. I'm trying not to be, like,
- 21:50the doom and gloom person about it. Like,
- 21:52I very much do look to this stuff
- 21:55to augment my job today
- 21:57and help me be faster? Like, I don't
- 22:00know. IT Pro land, all we've ever been
- 22:02doing is trying to automate ourselves out of
- 22:04jobs. This is just another step in that
- 22:06journey is the way I kinda look at
- 22:08it. And then just go watch Mission Impossible.
- 22:10Have you seen the latest Mission Impossible yet?
- 22:12I haven't. Don't spoil it for me. I
- 22:14will not spoil it for you, but all
- 22:16I could think about was AI and Copilot
- 22:18during Mission Impossible.
- 22:25Do you feel overwhelmed by trying to manage
- 22:27your Office three sixty five environment? Are you
- 22:29facing unexpected issues that disrupt your company's productivity?
- 22:33Intelligink is here to help. Much like you
- 22:35take your car to the mechanic that has
- 22:36specialized knowledge on how to best keep your
- 22:39car running, Intelligink helps you with your Microsoft
- 22:42cloud environment because that's their expertise. Intelligent keeps
- 22:45up with the latest updates in the Microsoft
- 22:47cloud to help keep your business running smoothly
- 22:50and ahead of the curve. Whether you are
- 22:51a small organization with just a few users
- 22:54up to an organization of several thousand employees,
- 22:57They want to partner with you to implement
- 22:59and administer
- 23:00your Microsoft cloud technology.
- 23:02Visit them at inteliginc.com/podcast.
- 23:06That's intelligink.com/podcast
- 23:13for more information or to schedule a thirty
- 23:15minute call to get started with them today.
- 23:18Remember, Intelligink focuses on the Microsoft cloud so
- 23:21you can focus on your business.
- 23:25Alright. So Copilot Tuning, folks can go out
- 23:27and sign up for that. I believe you
- 23:29have to sign up for a preview for
- 23:30that one or or kinda put yourself Yeah.
- 23:33This one's still it's early access. About the
- 23:36Copilot Tuning early access,
- 23:38customers with more than so this is a
- 23:41high barrier to entry, so I probably cannot
- 23:43do this yet either. But it does say
- 23:45early access, you have to have more than
- 23:475,000
- 23:48Microsoft three 60 five Copilot seats to participate,
- 23:51EAP candidates undergo
- 23:53screening, make sure they have scenarios that align
- 23:55with it. Like, Microsoft is at that point,
- 23:58they're sounds like they really wanna learn from
- 24:00this before they just throw it out there
- 24:02to GA for everyone. This is the same
- 24:04thing you saw with Copilot before. So
- 24:07this shouldn't be a model that surprises folks
- 24:09at this point. They have a couple
- 24:11people that are already using it. Ernst and
- 24:13Young, McCarthy,
- 24:15I don't even know how to pronounce that
- 24:17word, Tyrolt,
- 24:18Land O'Lakes.
- 24:20They have some big companies using it. So
- 24:22if you're a large customer and wanna go
- 24:24play with us, early access program, if you're
- 24:27not, be patient. You'll get it eventually. Don't
- 24:29let that stop you from going out and
- 24:30watching some videos. Like, the mechanics video is,
- 24:33like, ten, twelve minutes. Totally worth your time,
- 24:35I think. This looks like a build video.
- 24:36I haven't clicked on this one. This one's
- 24:38probably, like, an hour long. Oh, no. This
- 24:41one's ten minutes. There's an introduction video too
- 24:43on the article, so definitely go out and
- 24:46learn from it. Start building agents.
- 24:48That's I've been playing more and more with
- 24:50building some agents as well, which that leads
- 24:52to another news article. This one's
- 24:54one that caught my eye is
- 24:57you now have
- 24:59Microsoft
- 25:00Entra Agent
- 25:02ID, Secure and Manager AI agents. So this
- 25:05one was, I think, right before
- 25:08Build, and I saw Merrill posted something about
- 25:12this. I saw that. I saw a few
- 25:13blog articles about it. Alex Simmons has a
- 25:16announcement on the Entra blog about it. But
- 25:19this was interesting to me because, again, we've
- 25:22been talking about we're building these agents. We're
- 25:24giving them access to
- 25:26SharePoint sites, to different data.
- 25:29And this is starting to look at, so
- 25:31how do we actually go in
- 25:33and start securing those agents, securing what those
- 25:37agents can do?
- 25:38And now if
- 25:40you're an admin and you're a tenant, you
- 25:42can go out to Entra, You can actually
- 25:44go out
- 25:45to Entra
- 25:46and go to your enterprise applications.
- 25:48And if you look at application
- 25:50types,
- 25:51you now have an agent
- 25:53ID
- 25:54type in your
- 25:56enterprise applications.
- 25:57And this is this is cool
- 26:00just from a pure
- 26:01reporting perspective right now, and that it will
- 26:04give you a list of every single agent
- 26:08that has been created in your environment
- 26:10by any users,
- 26:12whether it's an agent created in Copilot Studio
- 26:15or Azure AI Foundry that are registered a
- 26:19part of your Microsoft three sixty five tenant.
- 26:21This is one that there's some videos out
- 26:24there. I went and did it. My tenant
- 26:26saw all the agents there.
- 26:27Great for reporting perspective
- 26:29for right now. I would say from, like,
- 26:31a security perspective,
- 26:33there's not as much you can do, but
- 26:36there's a lot of stuff coming. So they
- 26:37have a whole part of this that's what
- 26:39what's next with these EntraID agents now that
- 26:41they're in EntraID, now that they're showing up
- 26:44as enterprise applications.
- 26:45It means that you can start going in
- 26:47and looking at, like, eventually what Graph API
- 26:51access they have or
- 26:53even creating conditional access policies for your agent.
- 26:55So now someone can interact with this agent.
- 26:57Being able to go in and do
- 27:00granular conditional access and all that detailed permissioning
- 27:05that comes from that perspective
- 27:06because,
- 27:08I mean, I mean, we're getting to the
- 27:09point with these things, like, going back a
- 27:11little bit to what we talked about earlier
- 27:12with
- 27:13agentic workflows and these things effectively talking to
- 27:16each other. Like, you almost are treating them
- 27:19as individuals within your business. So it makes
- 27:22sense from that perspective
- 27:24to secure them in in the same way
- 27:26that you would your users and other things
- 27:29that are out there. So this will give
- 27:30you additional operational controls
- 27:33and, frankly, like, comfort food around locking some
- 27:36of this stuff down when it comes to
- 27:38what do they access, when do they access
- 27:40it, how do they access it.
- 27:43The the the I think the granular graph
- 27:44stuff will be great. Like, hey, making sure
- 27:46this thing only has
- 27:48a read permission versus a read all, or
- 27:51does it does it have the ability to
- 27:53actually
- 27:55generate content in that environment? Like, oh, no,
- 27:57I wanna I wanna block rights for this
- 27:59one because I don't actually want it to
- 28:01go in and I don't know. Somebody goes
- 28:03and builds the agent to rename titles in
- 28:06the GAL.
- 28:07You you you might not want that that
- 28:09that thing
- 28:10running that way. Yep. It's also good just
- 28:12from a
- 28:13visibility perspective for these things. Like, it's another
- 28:16way to kind of track adoption in your
- 28:19organization, maybe go talk to some users,
- 28:22see where they're getting
- 28:24ROI out of out of these kinds of
- 28:26things,
- 28:27and better manage the life cycle of them.
- 28:29I still see a lot of places where
- 28:32you go in and you look at their
- 28:35their intro tenant, and it's just the graveyard
- 28:38of lost stuff that that has not been
- 28:40cleaned up over the years and hasn't been
- 28:42attributed to a user. Hey. What's this old
- 28:44enterprise app ID doing out here that has,
- 28:47like, basically, god permissions in your environment? Well,
- 28:50like, what did you do, and why was
- 28:52this a deity? Oh, that was created way
- 28:53back when five years ago when we spun
- 28:55up our tenant. Great. Let's clean it up
- 28:58today. Right. And I think some of those
- 29:00two going back to life cycle,
- 29:02above and beyond this announcement, permissions change too
- 29:05with Graph API. Right? Like, there's old applications
- 29:07out there that have full control over your
- 29:10entire SharePoint environment
- 29:11because
- 29:12granular
- 29:14site level API access in the graph wasn't
- 29:17there when these apps were originally created. So
- 29:20as things continue
- 29:22to evolve, permissions continue to evolve over time,
- 29:26being able to,
- 29:28see that life cycle, see what they're doing,
- 29:30see what they have access to, some of
- 29:32the auditing and monitoring you'll be able to
- 29:34do around
- 29:36logs and visibilities into what these agents are
- 29:38doing, what they're accessing.
- 29:40You've even seen that some with Copilot and
- 29:42some of the new stuff around,
- 29:45sensitivity labels and being able to do DLP
- 29:48and audit user interactions
- 29:51with Copilot,
- 29:52now being able to audit agent interactions with
- 29:55stuff. Enterprises, this is I feel like it's
- 29:57something you have to have when you start
- 29:59getting into AI. If there's legal questions that
- 30:02come up, questions that come up about how
- 30:04something was generated, being able to dive into
- 30:06these logs and see this does become super
- 30:08important. And this is one that's out there
- 30:10today. Like, folks can go play with this
- 30:12one. Yes. You can see them. Like, they
- 30:14have a whole list. Again, I could see
- 30:16it in mine. It's labeled as preview, so
- 30:17all the preview stuff, it's in preview.
- 30:20I'd say don't use it in production, but
- 30:22it's just a preview feature that shows up
- 30:23in production.
- 30:24So don't put production workloads with it, but
- 30:28then a lot of this other stuff that
- 30:30we talked about, the conditional access, the auditing,
- 30:32life cycle management,
- 30:34in this blog article from May 19, it
- 30:37says,
- 30:38this is all in the coming months. A
- 30:40lot of these capabilities
- 30:42around
- 30:42zero trust security posture,
- 30:45all of that is going to be rolling
- 30:47out to these
- 30:49agent IDs in Entra. So that's one I'm
- 30:52looking forward to playing with, keeping an eye
- 30:54on. What other one you wanna talk about?
- 30:56We maybe have time. One more announcement. Why
- 30:58don't we do the Russinovich session? So he
- 31:01he did a
- 31:03app build. Mark Russinovich did a session on
- 31:07inside Azure innovations.
- 31:09I don't I don't know if you had
- 31:10a chance to go and watch that one.
- 31:12I failed on my homework to watch that.
- 31:15So
- 31:16I am curious for you to fill me
- 31:17in on that one. I'll put a link
- 31:19in the show notes for everybody, but that's,
- 31:21b r k
- 31:23one ninety five.
- 31:25Slides are out there, recordings out there,
- 31:28all all that good kind of stuff, so
- 31:30folks can go and take a look at
- 31:32it. But, effectively, like, this talks about
- 31:35the internals
- 31:36of Microsoft and Azure infrastructure.
- 31:39So what's the latest in boost and DPUs?
- 31:44What's going on with FPGAs?
- 31:47How are these things being
- 31:49leveraged within the
- 31:52within the Azure fabric.
- 31:53One of the cool things
- 31:56that was in there, or at least I
- 31:57think it was cool more because, like,
- 32:00some folks on my team were
- 32:02were were involved in it,
- 32:04is there's a demo in there where he
- 32:07talks about blob storage and scaled accounts. And
- 32:11so one of the things that we've been
- 32:12doing in storage over the years is trying
- 32:15to enable these large
- 32:17AI training companies
- 32:19who are doing these large pre training runs,
- 32:21so OpenAI, so the world, things like that.
- 32:24And
- 32:25as these customers are bringing in petabytes
- 32:28and petabytes and petabytes of data, it's not
- 32:30like single digit, It's it's double to triple
- 32:33digit
- 32:34to exabyte scale because you got, like, the
- 32:36world's knowledge in there. Like, downloading the New
- 32:39York Times archive is a pretty big thing.
- 32:41Do that across a bunch of
- 32:43different sessions that exist out there, and
- 32:46it it gets quite a bit bigger. So
- 32:49we did a demo,
- 32:50and we've been on this journey with scaled
- 32:52accounts to ever increase the amount of throughput
- 32:55that we can give them. So ingress, egress,
- 32:57the amount of IOPS we can give them
- 32:59in the form of TPS, all that. So
- 33:01we did a demo where
- 33:03we were able to, on a single storage
- 33:05account,
- 33:06provision
- 33:07a workload
- 33:09that ran
- 33:10at 15 plus terabits a second of ingress,
- 33:1425
- 33:15plus terabits a second of egress, and this
- 33:18is all random
- 33:19random IO through
- 33:21through an HPC benchmarking
- 33:24suite called
- 33:25IIOR. Okay. We were able to just run
- 33:28this for hours.
- 33:29We didn't do this in a special place,
- 33:32like, just public fleet.
- 33:34This this is stuff our our customers
- 33:37can go in and do today.
- 33:39So so I thought that was a super
- 33:41cool one. And like I said, important to
- 33:43me because while my team doesn't work on,
- 33:45like, the perf and scale stuff, one of
- 33:46the key components we work on is AI
- 33:48and ML clients.
- 33:49So I own our my team,
- 33:52we we do our SDKs, client tools.
- 33:54BlobFuse is one of those. So BlobFuse was
- 33:57heavily used in that demo and ready to
- 33:59go. So we've done a bunch of work
- 34:01there just to be able to, like, spin
- 34:02up workload
- 34:03in BlobFuse and have it be able to
- 34:05take, like, full advantage
- 34:06of
- 34:07a VM's NIC and soak it all the
- 34:10way through. So I I would recommend everybody
- 34:12go watch that session. Like, if you're interested
- 34:15in the
- 34:16either the internals of Azure or even a
- 34:19little bit more about the internals of of
- 34:21the way some of these workloads compose. So
- 34:24there was some stuff in there about Linux
- 34:26Guard and kind of how code integrity is
- 34:29handled
- 34:30within Linux, some cool demos for that one.
- 34:33There was a bunch of stuff around
- 34:38virtual machines
- 34:39and kinda how we do host OS upgrades,
- 34:42hot patch, driver swaps,
- 34:44full reboots,
- 34:46and and how those things go within the
- 34:49within the infrastructure. There was some insights into
- 34:51how we do RDMA
- 34:53on the networking side, like, all sorts of
- 34:56cool stuff. And because it's a Recenavit session,
- 34:58it's also demo driven, so you actually get
- 35:00to, like, see it in action. It's not
- 35:02just a bunch of slideware. Yeah. I'm flipping
- 35:04through
- 35:05the slides here, and it looks like there
- 35:07were, like, 10 or 12 demos in there.
- 35:09And it is like, I always try to
- 35:11go watch Russinovich's.
- 35:13It has been a time constraint more than
- 35:15lack of desire to watch his yet. But,
- 35:17yeah, all this stuff, there's some stuff around
- 35:19that. A confidential
- 35:21compute,
- 35:22a bunch of stuff around Azure confidential
- 35:24GPU VMs.
- 35:26Yeah. His are always,
- 35:28like you said, super
- 35:30demo driven, but I would also say super
- 35:32nerdy.
- 35:33Like, for
- 35:35super nerdy or super technical. Maybe I should
- 35:37be politically correct. Super technical, like, the amount
- 35:40of detail that are in his sessions and
- 35:43the amount of I don't know if he
- 35:44just sits there and practices all these to
- 35:46get all these technical details down or if
- 35:48he just has this much technical knowledge floating
- 35:50around us in his head. Either way, I
- 35:53am always super impressed with the
- 35:56level of technical depth that he's able to
- 35:59go to in these discussions and these presentations.
- 36:02So I always find them fascinating. They're great
- 36:05to go watch.
- 36:06Mark is
- 36:07super deep in the platform
- 36:10and and does understand a bunch of this
- 36:12stuff. Like, he's not out there just
- 36:15spouting it off based on a script that
- 36:17I think somebody else wrote for him. So,
- 36:20yeah, I I encourage everybody
- 36:22to to to go and watch that one.
- 36:23It's it's it's worth an hour if you're
- 36:25into Azure in general, and you and you're
- 36:28kinda looking for what's the latest in hardware
- 36:30innovations there across really all the core services,
- 36:34so compute, networking, storage. Yep. And if you're
- 36:37only interested in one or two of those,
- 36:39skip around the session. You can watch it
- 36:40on YouTube, and it's got chapters and everything
- 36:43in there for you. What is this? The
- 36:44world's first analog optical computer
- 36:47for accelerating
- 36:48AI
- 36:49inference.
- 36:50Potentially a hundred times more energy efficient than
- 36:53GPUs.
- 36:54If that is at all intriguing to you
- 36:56about light doing massive parallel computations
- 36:59for AI, go watch the end of this
- 37:01session. Right? Go watch the end of it.
- 37:03So power consumption is important. Sustainability
- 37:05is important. Yep. I saw and and
- 37:08don't quote me exactly again on the percentage,
- 37:11but it's something like three to 4%
- 37:14of the world's power now is going to
- 37:16data centers,
- 37:17and it's been driven on this general uptick
- 37:19around
- 37:21more and more CPU and particularly more and
- 37:23more GPU usage because these things are so
- 37:25power hungry. So
- 37:27I I think it is important that all
- 37:29these companies that are working towards this stuff
- 37:31continue to strive towards sustainability
- 37:34and innovations
- 37:36in
- 37:37not only, like, capping power consumption, like, at
- 37:40some point, it can't keep growing forever,
- 37:42but also bringing it back down. So things
- 37:45like like analog
- 37:46optical computer AOC, not not the other AOC,
- 37:50this AOC
- 37:51fall into that category of let's make the
- 37:53world a better place. Very cool. Well, those
- 37:55are some fun announcements,
- 37:57interesting news. We spent, like, the whole time
- 37:59in, like, three news articles. You should see
- 38:01the number of blog posts I have to
- 38:02go through. I think we went through
- 38:04a a couple hundred of them that are
- 38:06out there. So the other thing that I'll
- 38:08put in the show notes for folks,
- 38:11the the book of news
- 38:13is out and available as well for that,
- 38:15and and that's publicly available. So if you're
- 38:17interested in maybe the more
- 38:20broad roundup of, hey, what were the bigger
- 38:23announcements
- 38:24at Microsoft Build, things like that, I would
- 38:27encourage folks to go out and
- 38:30take a look at the book of news.
- 38:32Maybe
- 38:33throw it into your favorite LLM and have
- 38:36it summarize it for you. Absolutely. And if
- 38:38you have any questions,
- 38:39topics you want us to talk about,
- 38:42news you want us to cover in future
- 38:43episodes as well, let us know. I would
- 38:45say either if you're part of the membership,
- 38:48let us know in Discord. If you're not,
- 38:50LinkedIn,
- 38:51I'm quickly leaning towards as being the social
- 38:54media social media probably social media, social network
- 38:58that I am spending the most time on.
- 39:00All the other ones
- 39:01Yeah. I feel very hit or miss. LinkedIn
- 39:04is probably the best place to get both
- 39:06Scott and I and get our attention,
- 39:08if you have topics or questions about the
- 39:11show. You can tag the podcast there. You
- 39:13can also go to the website and just
- 39:14use our contact
- 39:16contact
- 39:17contact us form.
- 39:18It's alright. It's Friday. It's time to
- 39:21call it a day at 11:00 in the
- 39:23morning and go relax for the rest of
- 39:26the day or go to meetings. I'll probably
- 39:28do the former. But as always, thanks, Ben.
- 39:30Appreciate the conversation. Thank you, and enjoy your
- 39:32weekend. We'll talk to you next time. Yep.
- 39:36If you enjoyed the podcast, go leave us
- 39:38a five star rating in iTunes. It helps
- 39:40to get the word out so more IT
- 39:42pros can learn about Office three sixty five
- 39:44and Azure.
- 39:46If you have any questions you want us
- 39:47to address on the show, or feedback about
- 39:49the show, feel free to reach out via
- 39:52our website, Twitter, or Facebook.
- 39:54Thanks again for listening, and have a great
- 39:56day.