The End of the 3 Year Implementation: Lessons from the AI Front Line

Speaker A: Foreign.

Speaker B: Good afternoon and welcome to episode 81 of the Leading It Podcast. The podcast for Australian technology leaders who want to cut through the noise on AI cyber security data, uh, cloud infrastructure strategy and leadership. I'm your host, Josh Rubin, CEO, uh, at empyrean it. And I'm joined by CIO Tom Laden. And we tackle the fast moving IT landscape from both sides of the fence with pragmatic advice you can actually use by popular demand. In this episode, we are privileged to have Andrew Welsh, the global AI technologist, return as our guest to continue our discussion about AI in the enterprise and what he's seeing in the market. G', day, Tom. G', day, Andrew. Welcome back.

Speaker C: G', day, Josh. G', day, Andrew. And g', day, Josh's dog in the background as well.

Speaker B: Yes. Sorry, I've told him to shut him up, but no, I love you.

Speaker A: So, guys, I was. Episode 80 and episode 81. It must be a slow news day. It must be a slow news day in the world.

Speaker B: You're not the first returning. You're not the first returning guest, but you are the first that we've done back to back because the conversation was so good that we thought, and we didn't get through everything we wanted to talk about, so we thought we.

Speaker A: Now there's been more.

Speaker B: Yes, there's more. But wait, plenty more. So today is, by the way, it's King's birthday in Australia, Andrew, if you remember that, that King's birthday. So it's actually a public holiday here.

Speaker A: Is that the. Sorry, is that the current HM King or is that some other king that he's

Speaker C: got? You guys got rid of him a long time ago.

Speaker B: But we.

Speaker C: Mainly because we got off out of it.

Speaker A: Yes, I did. So. So I did. I lived in England for four and a half, five years and I actually went to see. This was pure coincidence. I went to see the show Hamilton on the day of Charles's coronation, which was very. I don't know if you guys are familiar with the show, but it is, it mercilessly mocks King George iii, though actually in a way that is very catchy and made him one of the most popular characters on the show. But it was very, it was sort of surreal. On the day of the coronation, going to, uh, to see Hamilton for the first time. But anyway, neither here nor there. Happy birthday, King.

Speaker B: Thank you. And, uh, Andrew, you've been traveling, so you've had a couple of conferences and some events. Johnna, maybe some of them are relevant to, uh, discussion. John, maybe talk about.

Speaker A: Yeah, well, The I'll focus on the one that is open invite that, you know, some of the other travels I've had recently have been specific to specific organizations I work with. But I mean, listen, since we last talked we had Dynamics, uh, Minds, which is the. I think it's pretty. Amongst those who are aware of dynamic Minds. It is, I feel like pretty well agreed. The best Microsoft. The best Microsoft Orbit event of the year. There was a lot of talk this month or this year it happens every week at the end of May in Portoro, Slovenia. And this was the fourth one. So there was a lot of talk I felt this year of those of us who have been to all of them where the first year we all kind of went because we were, we thought it was weird, right? Like this was this odd conference happening two and a half hours from any. Away from any major airport in a country that almost nobody has been to. It's one of the smallest countries by population in Europe. And we all kind of went. A lot of us went because we thought, well, it's never going to happen again, right, let's go see this strange event. And I think everyone who was there the first year, we had different moments when we thought, oh, this is going to happen again. For me it was walking out into this terrace bar that they have on the top floor of the hotel, the hotel there, and seeing the Adriatic Sea. And from that vantage point you can see straight out to sea. The sun sets over the water, you can see Slovenia where you're standing, you can see Croatia on a clear day, you can see Italy. And thinking this is so worth the two and a half hour journey from the closest major airport. It's just a, uh, it's just a phenomenal. It's a fabulous three or four days

Speaker B: and it covers so D365AI, all the core Microsoft.

Speaker A: Yeah, so. So it covers. Traditionally it has been a uh, business applications focused event. I actually think if I were going to give the organizers any feedback, I would say, listen, take the tracks, right? Have a track for business applications, have a track for AI, have a track for Azure, maybe have a track for the data platform. Broaden, broaden the appeal, expand the appeal. Because it's just such a fabulous event. But yeah, ah, it definitely has a. I don't really present a lot and talk a lot about business applications explicitly anymore. I talk about those technologies in the context of other things like the data platform and AI, et cetera. So my sessions were on AI strategy and architecture. So I think that you definitely go to dynamic Minds at least once if you are in the Microsoft orbit at all. Right. But for business applications folks, it has special appeal I think.

Speaker B: And I have to say last year I signed up for the conference. Obviously I couldn't go, but I watched on demand the sessions, your sessions and your colleagues sessions and they were excellent. So I would advise, yeah, it was well worth a uh, few hundred dollars it cost me to get the content. So thank you, Andrew. That's a cheap way to get the content. Just saying.

Speaker A: Thanks, Josh.

Speaker B: Giving you actually giving away.

Speaker A: I think it's Learn Cloudlight, Learn Cloudlight House or something like that. But anyway, actually it was funny this year. So last year in 2025 we had as an organization the center for Trustworthy AI and uh, our company Cloud Lighthouse, we had, I would say more of a presence. We just kind of went all out to have as many people there as possible. This year we had a uh, presence. There were, there were four of us but it was not quite as. We didn't have the level of support that we did the previous year just because there's so much going on. So to all these sessions I was running in with the tripod and the little camera and the microphone and getting people mic'd up. So it was, it was definitely managing partner and camera monkey, uh, from Cloud Lighthouse. But I learned a lot about audiovisual technology in a crash course at Dynamics Minds this year.

Speaker B: And one thing I was really interested last that from last year and whether that's evolved this year because I know a lot of the attendees are uh, partners but also customers, a wide range of perspectives about how AI projects are delivered differently to traditional software delivery or IT project. So do you want to maybe talk a little bit about that, your view there?

Speaker A: Yeah. So this is, this has been I think very interesting and I think it's been really challenging. Uh, last time we were together on the last episode we spoke a little bit. I think we touched on some of the challenges maybe that Microsoft partners are going to have in, are having now and are going to have in the years ahead. This is not new, but I think that it impacts maybe uh, partners that have a predisposition to business applications or specialty in business applications maybe more than others. So for the longest time, since it time immemorial, which was 70s, maybe 70s, 80s, but for the longest time what you've had is a company wants to do something, they want to implement a technology. We hear a lot about this word, implement. Right. And what they would do is they would go out and they might maybe they knew what technology they wanted to buy. Say they recognized they needed a new ERP or CRM or whatever it was, and they would go out to tender, they would acquire the technology itself. And then they might, maybe during, through the same tender or a separate tender, they would acquire the services to go implement it. And a partner would pitch, especially in the case of erp, for organizations of a certain size would pitch these pretty massive implementation projects, right? So maybe if you were a little bit of an organization on the smaller side, you could get it done in six months, a year. If you were a larger organization, these ERP implementation projects would go on for years. And some of the, I'm thinking, when I think erp, I'm thinking functions like operations and finance and supply chain. But excuse me, these patterns of implementation were repeated in other technologies as well. So HR systems, sales and marketing systems, CRM we came to call it eventually. And these projects would just go on and on. And that was, I worked on them. In 2007 there was a big merger of multiple rental car companies. And I remember being sort of socked away in a warehouse somewhere working on the, working on the SAP implementation for this new, newly minted mega rental car company. And that was the, that was both the model of technical implementation, but it was also the financial model for these partners. Right? You have a big team, you go out, you make millions and millions of whatever your currency on these projects and then you finish it and you move on to the next. But because the projects take so long, you have a lot of time to sell your book of business, your sort of forward facing book of business. And that's changed so significantly. I think it's started to change, change with the advent of cloud, public cloud technologies like Azure and aws. It then changed a bit more when these legacy technologies like ERP migrated to cloud technology. It changed a bit more with the introduction of what we call the data platform. And now with AI, we're just in this world where projects should be generally understood to be extraordinarily fast, to go very quickly and to be much, much smaller. So the way that I encapsulate this is that the old way of doing it were these kind of short and fat engagements where you might make millions of dollars or millions of pounds, millions of euro over a one to three year period and then you were done and you were going to go away for 10 years. Right now you're going to make maybe m, you're going to charge your customer much less, but you're trying to build a long Term relationship with that customer.

Speaker B: So sorry, from. So let's flip it around to the customer side of the table.

Speaker A: Sure.

Speaker B: What should they be thinking about? So I mean they still need to do all the requirements gathering and all the functional bits and all that sort of stuff. So is it because the technical configuration is less. Because we're doing a three I what's the. Yeah, I'm just trying to understand what that means from a client point of view.

Speaker A: Yeah, that's really worthwhile. Okay, so first of all, I think that a client if you're an end customer, right. If you're not the organization implementing the software, you're the organization hiring someone to implement the technology for you. First of all, I think that you should be generally suspicious of AI and data type projects that are pitched to you as these massive multi year millions and millions of whatever effort that, that you should be generally suspicious of. And I think that you should challenge the companies that you go out to hire to demonstrate value to you in much more incremental ways. So I'm seeing software packages, I'm seeing applications, I'm seeing data packages of data technology. We could talk about the details of this, right? But I'm seeing these technologies rolled out very quickly, not in full. This is not like we're just going to flip a switch and cut over from old ERP to new erp. But you don't need to wait anymore to start realizing the value of some of this stuff. And part of that is because we can use AI assisted development technologies to implement some of that is because the technologies that we are now implementing come in, I think, much smaller chunks. And part of that is because, listen, if you're. Most financial services institutions do not rely like I'm thinking, banks, insurance, etc. They don't rely on ERP in the way say manufacturing or retail does. So part of this is industry specific, but every industry is going to have their core technology. So in financial services it might be a core banking technology, and manufacturing it might be supply chain. But what we're seeing now is that those are getting implemented more so in a kind of a vanilla fashion. We're not spending years and years customizing. We're putting in the clean core erp. So this is the phrase we use is keep the core clean, put in that ERP and then build around it. Because ERP is no longer really the center. Really it's about. ERP is a container for business process. You're trying to get data into your data platform so you can use it throughout your ecosystem.

Speaker B: So having that ecosystem oriented architecture foundation is important. And once you have that in place, these things become quicker. Right? Is that I'll let you sip your coffee. Yeah. Is that, is that what you're saying?

Speaker A: I think that, yeah, that's definitely what I'm saying. And so, but I'm not what, I'm not saying, I'm not saying that the level of effort is lower. Right. I'm not saying that. So let's take some of the nuance here, let's take some of the gotchas. So one gotcha is that the old way of doing things was work really hard over a period of years and then at the end of the road you're going to get the thing. The new way of doing things is work hard, maybe still over a period of years, but you're going to get value a few weeks in maybe, or a few, you know, a month in or whatever. So one of the things that partners that consultancies can do to succeed is to get really good at building sincere, genuine, productive relationships with their customers over time. And one of the things that customers can do on that, on their side of the equation is to invest in that relationship, right? To invest to choose partners that they feel they can build a relationship with over time. So I think that there's work to be done on both sides. The other part that I see here that I think is customers are going to be mighty surprised about, right? And we, maybe we should talk about the cost of AI But I think that customers have this notion that oh, the partner can do it quickly now because they can do it with AI. Well, AI costs money to use, right? So customers that are trying to squeeze their partners on price, I think are making a big mistake because they're basically saying, listen, I don't want you to take as long, I don't want you to use as many people so you don't bill me as many hours or as many days. But I want you to use the AI. But I forgot or I didn't realize that AI costs money too, right?

Speaker B: Use the free AI. Use the free AI.

Speaker A: Use the free thing where my data

Speaker B: is going to go out into the web and someone else, but use it, charge us less.

Speaker A: Exactly, exactly. Responsibility on both sides.

Speaker B: I'm m glad you said that, Andrew, not me. But and I also reflect if when Tom was a customer, he would have been a lot less supportive of this, of this chat than he is now. He's on the other side of the desk now you're going, now you go. So I have another question. Sorry.

Speaker C: I'm.

Speaker B: I'm very passionate about this stuff. So talking about implementing the core and then having a flexible edge. So that was up until probably 80 months ago where we would say, okay, do a vanilla D365 CRM implementation, but customize around that using satellite power. PowerApps now is PowerApps dead, right? Where is Power Apps? What's going on? What's happened to PowerApps? Because I'll say, from what I can see and from what Microsoft is saying, it's now just a background tool that you can vibe code in front of. Anyway, my view. Uh, sorry, Andrew, over to, uh, you.

Speaker A: Yeah, so, funny story about this, actually, right? Over the years at Dynamics Minds, the conference we were talking about at the Open, over the years there has become this joke, right? About every year at Dynamics Minds, someone gets up on stage and proclaims something to be dead. So, uh, we've heard all sorts of business applications is dead. Dynamics is dead. Power Platform is dead. Actually, I think that Charles LaManna, I think I was not at this conference, but as I understand it, Charles lamanna, last year in Vegas at the Power Platform conference, kind of leaned into this meme and had this big slide that said like, Power Platform is dead. And then he dropped down and said, as we know it. Right?

Speaker B: Yes, I do see it. I do see that.

Speaker A: So this has been. And of course people whose livelihoods depend on Power Platform were outraged and terrified at this proclamation. But anyway, listen, is Power Platform dead? No, Power Platform is not dead, but Power Platform is going to evolve. Power Platform is going to evolve in ways that we've, I think, just seen the beginning of. And what I don't know, what I don't know is if I think that there's a world in which the components of Power Platform, the things that you do with Power Platform are, uh, taken and assigned to other buckets of technology. Right. So I think that, and I don't have inside information on this, I would say I would be very surprised if there have not been conversations within Microsoft as to whether or not Power Platform as a banner, as a top level banner, makes sense. We can talk about whether it does or doesn't. But here's what does still make sense inside a Power Platform. Dataverse makes a ton of sense, Right? If I woke up one day and heard that Dataverse had now been put inside of fabric, that would not surprise me. I'm not saying that's going to happen again. I have no inside information on that. But those are mostly those are to a large extent marketing decisions. But Dataverse is still a very important technology. I think that power apps is still a very important technology, though I would expect power apps to, to change significantly over the next several years. I think that PowerApps, there's a world, right, in which Microsoft decides to take PowerApps in the direction of being essentially kind of the user friendly user experience, like the user experience that sits over top of AI. Rather than we're going to, as my friend Keith Watling used to say, we're going to clicky, clicky, draggy, droppy and build a Canvas app. If you were, if you're just starting off in the world of technology or Microsoft or power platform, I would not spend a lot of time learning Canvas apps right now. I don't, I don't think that's the future at all.

Speaker B: No. Okay, so yeah, you have your sort of continuum of. Was it incremental? Differential.

Speaker A: Yeah.

Speaker B: How you describe sort of different AI workloads and power platform fitting into. So for our audience, Jonah, maybe for those who have spent a lot of time reading your stuff, like me, maybe explain to them what we're talking about.

Speaker A: Okay. And actually we're just this week, so we're recording this on the Monday 8th June, and just this week we are finishing. It's all written, all the math is done, all the writing is done, everything is done. We're just kind of putting the finishing touches on the next release of the AI strategy framework. And so by the time this podcast comes out, it's possible that will be live as well, and folks can go get that and use that. But essentially the framework breaks the whole of an organization's AI, AI strategy, AI journey, what have you, what you want to call it, down into five pillars. Those pillars are strategy and vision, ecosystem architecture, AI workloads, responsible AI and scaling AI. And then within that there are dimensions. Each pillar has five dimensions. So what you get is you get 25 things. 25, we call them dimensions, but you could think of them as topics, considerations, packages of kind of the. Where responsibility, authority and accountability all kind of come together. Someone needs to be responsible for, have the authority to control and be accountable for the success of each dimension. And the dimensions are things like programmatic rigor. Are you doing putting the work into making sure that your AI initiatives actually come to fruition rather than just get started and go nowhere. Data distribution. Are you building and architecting the technical components needed to distribute data across your ecosystem? Embedded AI. Embedded AI, we used to call it incremental AI. We're now calling it embedded AI. That's Embedded AI is your general purpose AI tools of choice. So that could be Copilot, that could be Claude, that could be Copilot for sales. Right. Like whatever that is. And I could go, I could go on and on, but basically it's 25 dimensions. And what we're doing, what we've done now in this next release that you'll be able to see within days, is we've now built a mathematical weighting model that sits behind each of those dimensions. Right. So the idea is that different dimensions require different levels of effort to come to fruition and they make a different impact on the organization. So in other words, not every dimension is equal to every other dimension. So we now put a model behind that to help organizations fine tune their investment decisions and understand where they're going to get, as Americans would say, the biggest bang for your buck, which sort of lands flat as a phrase in Europe.

Speaker B: But anyway, we're good here, we're good here, we're good here.

Speaker A: We're good in Australia.

Speaker B: Yeah. So let's go back to the question. So you have your embedded, which is your AI tools, things that are embedded in your existing tools, like M365 copilot, then incremental is. Go on. I can say it because I know. Okay, yeah.

Speaker A: Embedded, Embedded. I'm sorry, Embedded AI. It's useful to think of your AI scenarios, your AI workloads on a spectrum. Embedded AI, that's the stuff you can just go by and turn on. And um, your effort with embedded AI, this is think Copilot or Claude or ChatGPT embedded AI. The effort is going to be around adoption, it's going to be around building digital fluency amongst your workforce so that your workforce, your colleagues are able to make the best use of it. About rethinking how folks work and what business processes and organizations look like. Most of the technical work has already been done for you. You're buying AI that exists personal productivity land, right?

Speaker C: So this is how do I make an individual work better with AI though

Speaker A: I think that I'm really seeing the lines blur here, right? Like at this point, these general purpose AI tools. Copilot, Claude. I keep saying Copilot and Claude, because those are my favorite. ChatGPT would be one of them, Copilot for sales, et cetera. Those are becoming more and more capable. So for a long time we kind of. Not for a long time, for about a year we said, listen, you can buy Copilot or You can go use Copilot Studio to extend. And this is where you get into extensible AI, which is using a tool to extend the capability of AI. And we sort of viewed those as separate activities. Now, ask myself questions, right? Like, well, if I can write a, uh, if I can write a markdown file that provides guidance to Copilot or Claude and essentially gives that general purpose AI, that embedded AI a new skill, am I extending it or am I just telling it what to do? So the lines definitely blur, but I think the way to do it with

Speaker B: all the MCP servers and all that. Yeah. Okay.

Speaker A: Yeah.

Speaker B: So does that kill Power Platform even further?

Speaker A: He wants to know.

Speaker B: Like, I went to two or three Power Platform conferences in Vegas and the last one I went to, they spent the whole conference telling us about how there wasn't going to be any more work because it was going to be able to all done through a prompt. And everyone was cheering. I was thinking, why are you all cheering Ryan Cunningham when he's just telling you how all your jobs are going to go?

Speaker A: Yeah, this is a. This is. You've hit. Hit the nail on the head for a very. The phenomenon is most acute if you go to a power platform event. Like, one of the things that. One of the dangers that I see in conferences that are all about power platform right now or all about dynamics is it's like a bunch of people have gotten together to mutually convince themselves that they're good, that the skills of the old way of doing things are totally cool. You don't have anything to worry about. And usually I show up on stage and kind of in so many words, I'm like, yeah, we have things. This group has things to learn. Right? So, no, Power Platform is not dead.

Speaker B: But you haven't said why. You said it's not. But you haven't said how or why.

Speaker A: Okay, so. So let me give you some concrete examples and in a phrase, Power Platform is not dead, but folks who. The old way of doing Power Platform is dead.

Speaker B: Folks who made a living out of

Speaker C: building power apps happened since Access Microsoft Access had apps. You could build apps in mugs of legs.

Speaker B: Yeah, yeah.

Speaker A: Just weirdly hangs on like, why is it still a thing?

Speaker B: Yeah.

Speaker A: Okay, so. So why don't.

Speaker C: Josh is.

Speaker A: Yeah, Josh is going to be mad at me if I don't. If I do. Okay, some examples. Okay. When I hear people talk about vibe coding, one of the first things I ask is, where is the data? Where is your data? And almost nobody can tell me where their data is. When they have Vibe coded something. Right. I think, uh, where's my data? I didn't think about that. Right. So here is a way that Power platform is not dead. Dataverse is where your data should be in a lot of Vibe coding type scenarios. So I think that what Microsoft sees here is a world in which a lot of this data lives in Dataverse and you Vibe code on top of it. So diverse. Not dead. PowerApps, PowerApps becomes. Or maybe there's a world in which sort of the PowerApps and I don't know, I want to say this is not insider information. Maybe there's a world in which the PowerApps name sort of quietly fades away and is replaced by technology that allows folks to build apps to Vibe code apps on top of secure sources of data. Like already.

Speaker B: It's already there, Andrew.

Speaker A: Already there. It's already there. But I think that there's a lot of it's already there in a very early version, in a very early stage. Right. So it's not where it's going to land. Right. The Conditions today on 8th June 2026 are not where this.

Speaker B: So I'm saying as an owner of a, of an, of a, of an IT company that used to have a power platform practice, I've moved, pivoted off to after a fabric like D365AI, like that whole. We don't talk about having a power platform practice at all anymore. Wow.

Speaker A: Yeah, I think that's fair. I think Power Pages. So Power Pages has always sort of been this, in a little bit of a way like this orphaned technology because it was the last of the power products, um, to really come into its own. Power Pages very interesting. Right. In that it's moving away. It has over the last 18 months or so moved away from being a clicky, clicky draggy drop be build the thing right here like kind of a Microsoft version of Squarespace. It has moved into the realm of almost being an orchestrator for you've got public facing websites, public facing web portals that you Vibe coded or you've built in React or whatever it is and you have Dataverse and then Power Pages I think is entering a realm where it's almost more of an orchestrator. Right. It allows a Vibe coded thing to sit publicly on top of a secure data source. Dataverse and Power Pages is the orchestrator in the center.

Speaker B: You know, uh, funny you said that most of our uh, recent power apps work has been power Page building portals and integrating it with an internal power app. That's been where more of the work is. So I think that resonates. And this new AI strategy, the last one was 73 pages. Is it going to be 73 again? I'm just thinking someone, uh, a friend. Am I asking for a friend? Andrew? I'm asking for a friend.

Speaker A: Asking for a friend. Okay, so I have good news and bad news. The, the, uh, the bad news is, despite the fact that we set out, we actually cut a lot out of, uh, this edition. It's longer than the last one. I think we haven't. I haven't seen the final laid out with the graphics and everything, but it's longer than the last one. So that's the bad news. The good news, though, is that there's a lot of new stuff in it. And the other piece of good news is that we're working right now on a piece of technology, we call it Centru AI. And what Centru AI is, it's basically going to be a tool that uses both AI and graphical user interface to take everything in the framework and make it accessible. Right. So I don't want to give too much away, but basically, rather than read 90 pages of stuff and then evaluate where you stand and where you have risk around risk and opportunity around AI, let's use AI to understand the 90 pages and then help an organization put it into practice better.

Speaker B: So there is. Assuming you'll share that stuff with us and we can share it with our. Yes, I'd love to, absolutely. And the one, the one I thought Talking about your 25, 5 sort of things, the things that I find customers have the most difficulty working out is scaling AI. Ah, is AIOps. How are we going to operate this thing moving forward? What's it going to look like? How are we going to manage. When we have a couple of hundred agents, how are we going to manage it? So, uh, from. Maybe. Do you mind talking about.

Speaker A: Listen, so first of all, AIOps is something that I think we've been aware of for several years. It was one of those things that we kind of knew, okay, we need to do this. There's been this phase of doing AIOps, essentially. Basically, AIOps is DevOps for AI things, right? There's that phase. Uh, yeah, yeah. And I mean, Tom, did you see. Have you seen any of that in your travels?

Speaker C: Yeah, I mean, I think the AIOps is fascinating because people are saying, well, we've played with it, it's a pilot, and now it's in production. So we now really care about what it does on a day to day basis and we care about when things change. So when via code is introduced, version 1.0.1.3 I want to know what it changed. I want to see the evidence that um, someone's actually tested this thing and I want to know what the impact is on all the people processing data behind the scenes. And for my. What I'm saying is that is missing. Like people have just forgotten all the old rulebook about change management and change releasing pieces into production when it comes to AI Just for Borden, right?

Speaker A: Yeah. And I think that you've gotten at uh. For me the first rule of AIOps is don't throw away the best practice and the things that you've been doing. Right. There's this organization that my team was called in kind of on an emergency basis. Essentially what they did was they deployed some agents into production and they spiked almost US$150,000 of Azure consumption in 36 hours because. And then they had no idea what the lineage of these things were. And it turns out they had just copied code to their desktop and then chucked it into production. And it's like guys, just because it's AI doesn't mean that we don't do this thing.

Speaker B: You see the One was it $500 million of Claude. Some company did because they didn't put it. It was in the press the other day. $500 million top. Yeah, you idiot.

Speaker A: I mean if that makes your choice where you're signing up for the paid plan of Claude and you're like should I go with the 75? I think I'm going to give this to you in pounds because we're a British company so my credit card is in pounds. Like should I go with the 75 pound one or should I go with the 150 pound one? That blows it out of the water.

Speaker C: But in the old days, right, we had controls around releasing change. Right? Yeah. Things out play and uh, people were accountable for those kind of stuff ups. But it seems. Yeah.

Speaker A: And so, so listen one first, don't throw away all of the good DevOps hygiene that you've built over a period of years. Don't just throw that away. Because this is AI. Yes. We need kind of an AI centric way of doing this and that is evolving pretty rapidly. But don't just throw it away. There's a partner, there's a partner that, that I've come to know. Well they're based in the uk. They're called Godell. Those guys are not. Those guys are actually building. They're very focused on building AI ops technology that others can take and buy and use. Right. So there are people who are, I would even say, starting to specialize in this area. And it mirrors what we saw in the 2010s, right, where you had some partners that were. They were DevOps partners. That is what they did. They might even work with a partner that does the actual app development and they focus on DevOps. So we'll see where that. We'll see where that goes. In really practical terms for CIOs, right? There is a technology built into Microsoft Foundry called Foundry Control Plane. Foundry Control Plane, I think is. It's not where it needs to be or will be at the end of the day, but I think it is a solid indicator that folks at Microsoft are thinking very seriously about how to go from building and governing and putting into production one agent to how to do, uh. The phrase I've heard, and I don't like this phrase is a fleet of agents. But that's. That was the phrase that I heard bandied about a few months ago.

Speaker C: Sorry. And what is your preferred collective noun of agents?

Speaker A: You know, I don't know, but maybe a flock of agents. Not a fleet of, like a fleet to me.

Speaker B: To me.

Speaker A: So I was at the Coast Guard, right? To be a fleet, it must be ships. There is not a fleet of agents. It's a flock of agents.

Speaker B: Okay, all right.

Speaker C: A sidebar comment. That. That's good.

Speaker B: That's all right.

Speaker A: That's all right. Josh is over us. Josh.

Speaker B: I was looking up. No, I was looking up. What did they call a flock? Plural of crows. It's called a murder of crows. So it's a murder of agents.

Speaker C: Oh, no, that's right.

Speaker B: That's even worse.

Speaker C: He's saying we've got a murder of agents. One year would make.

Speaker B: Yes, yeah. Just a, uh, couple of things. So Microsoft had their build conference guys the other day. I don't know if you. With your travels. I don't want to. I don't want to do a big deep dive because we do an ignite. A big ignite one. But there was a couple of things that I thought were really pivotal. That fabric IQ was ga. Andrew.

Speaker A: Yeah.

Speaker B: So, uh, you familiar with fabric iq?

Speaker A: I am. And so this is one. All of Microsoft iq. This is one that I am bullish on. Right. I am, um, I feel very favorably about Microsoft iq, and I think that it is. There's a world actually where Microsoft IQ becomes more compelling of a reason to live in the Microsoft AI ecosystem. Than actually the end user experience of Copilot. Right. The strongest thing that Microsoft has going for it is the fact that a lot of organizations either keep or manage their data estate, uh, inside of Microsoft. And that includes SharePoint, that includes your emails, that includes your business data that's sitting in Dynamics Dataverse that now is living inside of fabric because of cutting and the fact that we've moved, we have SQL capability in fabric on and on. So what Microsoft iq. And for the uninitiated that includes work iq. So work IQ is essentially, this is your business context, right? So work IQ are your emails, your SharePoint documents, your Teams, chats, et cetera. Fabric IQ. Think of that as kind of the state of your business, the state of your organization. So this is your quantifiable, this is data coming out of your erp, your CRM, your custom applications, landing in fabric. So that's fabric iq, that's like your more real time intelligence layer. And then you have foundry iq. Foundry iq. Think of that as the standing grounding knowledge that is provided to agents across Microsoft. Right. So that could be your internal policies, that could be your codified ways of working, et cetera, et cetera. And there's a new kid on the block, WebIQ, which is very new. I won't spend, let's not spend a lot of time on that. But listen, what Microsoft IQ is doing is it is bundling all of this intelligence up, uh, in a way that agents can consume it and there's a clear lineage where that data is governed and it doesn't change hands from say one technology to another. And I feel very optimistic about.

Speaker B: We do a lot of fabric now and yeah, what the non IT users are excited about is iq, right? Yeah, you can say because we're basically AI, it's looking at ones and zeros. It's not going to be as powerful, is it actually teaching it the language of your business and then that's going to make the agents a lot more powerful. And that's. So that was exciting because that, that went ga. And the other one that was, I thought was interesting is that Microsoft are getting into their own models. So they announced seven of their own models. LLMs. Yeah, yeah, yeah, I thought that was.

Speaker A: What do you think guys? What do you think? Um, I don't want to be the one.

Speaker B: They're very light. Yeah, yeah, yeah. 99 was ringing and asked where they, where your Apple phone is. I don't know. It feels, it feels too late. I don't know. I don't.

Speaker C: Tom, what do you think, well, the horse is bolted. But I mean, who can say, like if these models turn out to be very useful, people adopt them, people jump on, you know, they don't care about who's late, who's first. I mean, look at Claude versus chatgpt. Is that. Who would have thought it would have come out of nowhere? But it has, I don't know, because

Speaker B: they're specialized and they're trained on more narrow data and they're. Yeah. Andrew, did you have a view on that?

Speaker A: I mean, when you said that they're late, I just had this notion that, yeah, all of the MY models, they're going to be, they're going to be delivered on a DVD in the mail, a CD in the mail. You're going to get it and it's going to, it's going to invite. You can get it for the low introductory price of 9.99 for the first six months if you install. Now, listen, so the my models, here's the way that I, uh, First of all, I'm a wait and see on Microsoft's models. I think that they are late. I think that it's interesting because in a way they almost seem peripheral to Microsoft's overall strategy and Microsoft's overall play around AI. But there is a world in which if they are very good and Microsoft developed something very compelling where we could look back on this and think, this is like when Apple said they were going to move away from intel, uh, intel chips and build their own silicon. And this is. That has been. It made everyone nervous and kind of people thought in the Apple world why wildly successful move for them. So, so we'll see. I'm happy they're doing it, but we'll see.

Speaker B: Tom, we were chatting the other day and you were saying you were seeing some business users finding some challenges with AI adoption. Did you want to talk about that?

Speaker C: Well, I mean, we talked about it briefly with you, uh, talked about that sort of personal productivity versus business productivity trade off. Right. So many people I'm speaking to go, we've got to train everyone on copilot to be AI, I think. Yeah, okay. That's a path you've got to go down. But that is like training people on Word and Excel is training them how to be personally more productive using AI. Great. But you are, uh, missing the big picture at the moment. And this is where, going back to your point, Andrew, about having a broader view of your AI, using that AI strategy at a board level, not just at an IT level is so important, right? But I mean, also so new. So it's going to take a long time for this to disseminate and get out there and people to start to realize the AI is not copilot on the side. It actually needs to be embedded in your business.

Speaker B: And if everyone's just doing that, it's not differentiated really, is it?

Speaker C: No, it's like everyone having Excel a big deal. That's, that's.

Speaker A: Yeah, I think that, I think that. So there's some interesting data. Anna, uh, my wife slash business partner slash colleague, and Donna Sarkar did, had a presentation at Dynamics Minds on a topic that, that they came up with and that we're now developing further at the center for Trustworthy AI, call it the Human Context Protocol hcp. And it's essentially a question of how do colleagues, how do humans use and integrate AI into their work, into their leadership, into their decision making? And one of the statistics that Anna and Donna shared during their session is something like, there's research now that indicates that there's a 57% salary premium for individuals who demonstrate who are really good at integrating and weaving AI into their work. And that's not a question of do you sit there and sometimes during the day you chat with copilot. That's a question of are you wholesale? Not in one big moment. Right. But over time, are you weaving AI into the fabric of your day, the fabric of your life? And are you meaningfully changing the pattern, the patterns and the habits that you have? That's. That goes beyond training. I hate when I hear that word, training nowadays, because I think it's just silly.

Speaker C: That's huge. So the 57%, in what differentiated between salaries that. What do you mean by that? 57%.

Speaker A: So, so, so I'd have to, I'd have to go back and. Because that's not my number. That was a, that was a number that, that Anna and Donna shared. But as I understand it, basically, and I do think that this accounts.

Speaker C: So.

Speaker A: So let's backtrack. We're hearing, we're now seeing a lot of evidence that workforces, that colleagues across the economy are resisting the use of AI. Right. So in some cases it's more just sort of kind of indifference, and in other cases it could be considered more obstruction. And what we're seeing is that the users, the individuals, the colleagues who say, take a very different approach and say, no, I'm going to use AI and really make it a major part of my workday and a major part of how I operate what they are seeing. Is in the form of increased productivity, which leads to promotions, which can lead to promotions, can lead to pay rises. In some cases it means that they're the colleague that stays when someone else is laid off. Right. So I think there's a lot of data that goes into that. As an early figure, I wouldn't put too much, much stock in it, but the fact that it's there and the fact that people are studying this now is meaningful.

Speaker C: Yes.

Speaker B: Yeah. Uh, well, definitely with hiring. I mean we're doing a lot of hiring at the moment. One of the things is can you, are you using AI? Do you have AI skills? And if you're differentiating between candidates, that's a key consideration at the moment for sure. Now we're getting to. I don't know how we get to an end of an hour, but I do want to. Before we go, Andrew, I want to talk about the Frontier Campus and the three sort of models that you came out with. I think it's worth, probably don't have time for a deep dive, but to talk about those three sort of why, where it came from and what the three sort of architectures are at a, at a high level and what customers should be doing and thinking about it.

Speaker A: Yep. Okay, so real quick, there's, there's a new ish. It came out in early April. There's a newish or no late March paper that we wrote at the center for Trustworthy AI. It's called the Frontier Campus. The Frontier Campus is a, uh, blueprint or a set of blueprints for a globally distributed, multi agent, multi cloud architecture. So you can scale it around the world, you can use it locally there at home. You don't need to be a big organization to use it. You can choose one of the blueprints and roll that out or you can use all of them. Inside of the Frontier Campus there are three architectural blueprints. And the architectural blueprints are meant to provide an organization with a starting point and actually maybe an end point to some degree of what multi agent architecture looks like and how to scale that across their organization. No matter the size the blueprints are. Now Blueprint A is an open architecture. It is not product specific. You could go out and you could stitch together Blueprint A using some Google technology, some Microsoft technology, some specialized technology, specialized tools. Blueprint B takes Blueprint A and says, listen, if we were going to build this almost entirely or entirely in the Microsoft stack, what would that look like? What are the technologies that you use to do that? So blueprint B is the Microsoft architecture and then blueprint circumstances is what we call the offline architecture. The reason that offline architecture was important is because one of the core principles of the Frontier Campus, and you can read all of this in the paper, was that it be available in dramatically different environments so that it be available to an office worker in Copenhagen or it be available to someone in the field or in a conflict zone or in a place where there is a lack of Internet connectivity or whatever the scenario might be. So Blueprint C is offline and what it does is it prescribes an architecture for how to build and deploy AI, how to build and deploy agents that are powered by small language models and everything operating on a single device. Microsoft Foundry Local is a big component of that offline thinking. There are some trade offs there. But Microsoft Foundry Local is a great technology and an emerging technology there. So what I would love to see is I would love to see partners taking that model, the Frontier campus. And I'm starting to see this now and scripting or creating IP that helps an organization roll this out quickly. Right? So say you're hosting, you're building Blueprint B, the Microsoft architecture. I would love to see partners scripting the deployment of the Azure app services components that are needed in order to host that Frontier campus. In order to host containerized agents there. Microsoft has built the scaffolding. Partners have an opportunity to build some of the plumbing that helps customers do this more easily or a customer can go do it all themselves. We're kind of in a, we're in an evolutionary phase I really like.

Speaker B: I mean it uses the ecosystem oriented architecture as its foundation and given most companies, as you say are invested in Microsoft, it makes sense to and to stitch a whole lot of different things together. Or even are you advocating in your first one for open source or more? You mean open as in more just whatever technology you like?

Speaker A: I think both. We could do a whole podcast on uh, concepts like open source and vendor lock in and some of these things that people talk about. You can probably tell I have opinions about it, but it could go either way. You could go with Google or you could go with open source technologies with.

Speaker B: I mean we're all locked into Microsoft anyway really. If people think they're not locked into Microsoft, I think they're potentially kidding themselves.

Speaker C: Try very hard not to, I'll tell you that much.

Speaker B: Yes, it's difficult. It's difficult. And also that for people who are interested in it, I'd uh, highly recommend looking for the Frontier Campus and you Also sort of build in those five pillars that you have from your AI strategy as well. So it's really. Yeah. The, uh, quality of the information you put out, Andrew, is extremely high, by the way.

Speaker C: So thank you.

Speaker B: I want to congratulate you on that. It's really good.

Speaker A: Thank you.

Speaker B: You make us, uh, owners of partners lives a lot easier. So what would we do without you, Andrew?

Speaker A: Hey, that is that, Listen, that, that's the goal. That's what we're trying to do. We're trying to put, we're trying to create, trying to put high quality information out there. That is not a bunch of fluff, not a bunch of hype that runs the gamut from strategic thinking to, okay, now what am I going to do with this? And the whole idea is for partners and customers to be able to use that, to be, to operate as independently and without dependencies as they can and to give my little team and me a shout when you're, when you're stuck. But that's, that's what we're trying to do. So I'm happy you're using it. That's, that thrills me, honestly.

Speaker B: Yeah, well, uh, the fact that I sat and read through and underlined your 73 page thing says a lot about how boring my life and also as we said earlier, how excited I was to see that SharePoint file archiving is now available in M365 Archive. Yay. Finally.

Speaker A: Finally. So we were talking about this, we were talking about this before you hit record. And my promised reaction is that. So I went through a phase where I was really into registering domain names that Afghani domain names. So I've long had this thing, usually often at a bar. You would be surprised by how many domain names I've registered in a bar. But one of them, I found this website where I could register Afghan, like from domains from Afghanistan. And the reason this was of interest is because it's af. And so for a while I owned the domain sharepointisdead af and I pointed it to Chris Huntingford's Twitter profile. And over the years I pointed it at some other Twitter profiles as well. So I'll let you know. To folks at Microsoft, congratulations, you are this month's recipient of SharePointIsDead AF, the official website.

Speaker C: I don't know if I want.

Speaker A: I don't own it anymore. It's gone. It's gone. I don't own it anymore. SharePoint has new life, by the way, SharePoint really does have new life.

Speaker B: I mean, SharePoint storage is just crazy.

Speaker C: Um, well, that is another huge topic.

Speaker B: Yeah.

Speaker C: That is their point in the new world. Right. It's completely changed.

Speaker A: Point is undead. It's undead. It used to be dead, but it's no longer dead. It is undead. It is this.

Speaker B: Yeah. You know, the next thing is I'll be talking about building workflows on top of SharePoint.

Speaker A: Right. I was going to call me and say, hey, I need to convert all of these legacy Power Automate flows over. Over to SharePoint very much as.

Speaker B: Uh. All right, Tom, have you got any last words?

Speaker C: Thanks. I'm looking forward to reading this. These five pillars when they come out. I'm, um, really m. Yeah. Really looking to engage.

Speaker B: They're out. It's just version three or four.

Speaker A: Version three. Version three.

Speaker B: Yeah. I'm looking to version three. If you could email it over, that'll be. That'll be great.

Speaker A: I will.

Speaker B: Please, please. And thank you so much, Andrew. Really appreciate you coming back and sharing your valuable time. And that's not, not. I'm not being facetious there, sharing your valuable time with us. We really appreciate you being on and being so open. It's been an absolute honor and, uh, privilege to have you on.

Speaker A: Thank you so much, guys. Thank you. I was actually thinking these guys would be really fun to have a regular podcast with. So anyway, it's a pleasure. Thank you so much.

Speaker B: Be careful. We'll definitely have you back. We can't do three in a row, but we'll definitely have you back again soon.

Speaker A: And I'll be back for 84. For 84. Yeah.

Speaker B: Yeah. Well. And Andrew, do you want to give yourself a plug before we go about where people should come and find all your stuff?

Speaker A: Yeah. So you can everyone feel free. Connect with me on LinkedIn. I'm Andrew D. Welch. My middle initial is D for David. You can also find me at my company at Cloudlight, uh, House and the center for Trustworthy AI that publishes all of. All of this at. @centerfortrustworthyai.org we have the American spelling and the everyone else spelling of center. So hit us up.

Speaker B: Thank you so much. If you want to reach out to Tom with his red, yellow, blue and doing sort of more strategy work out there, you can reach out to Tom on LinkedIn or myself. Everyone knows at Imperion it. So thanks so much, everyone. Uh, I hope you had a good king's, uh, holiday. We'll be, uh, seeing you soon. See you guys.

Speaker A: Sam,