AI Agents in 2026: Govern Without Slowing Innovation

00:00:01 Mark Smith
Welcome to the Power Platform Show. Thanks for joining me today. I hope today's guest inspires and educates you on the possibilities of the Microsoft Power Platform. Now, let's get on with the show. Welcome back to the Power Platform Show. Today's guest is joining me from the United Kingdom. where he's getting ready for Christmas. All the links for what we talk about today will be in the show notes, as always. Thanks for listening. Let's get started. Craig, welcome to the show.

00:00:41 Craig White
Thanks. Thanks for having us again. It's really good to be on.

00:00:44 Mark Smith
Love having you on. There's been a lot of change and upheaval in your life. Tell me what's top of mind for you when we talk about food, family, and fun right now.

00:00:53 Craig White
Christmas, that is going to be lots of food. Definitely lots of family time and definitely lots of fun in our house, especially with my daughter. I think this might be the last year where Father Christmas is real. So sorry if I disappointed anyone else out there who thought he still exists. But yeah, this might be the last year we get to enjoy it. So we're going to make the most of the festive period and have a lot of good time and drink way too much beer and eat lots of mince pies. So it's definitely the plan for this coming weeks.

00:01:25 Mark Smith
I love it. I love it. It's funny about Father Christmas is then I've brought my kids up and I've got a five-year-old and a three-year-old that Father Christmas is not real because I didn't want Father Christmas taking credit for all the gifts we give them at Christmas. And so last night we had what they call a tree ceremony at a school. And you know, she's only been going to school for three weeks, starting school. And They have a father Christmas and she's like, he's not real.

00:01:57 Craig White
What have you done? That's brilliant. But you've got like, these are old school traditions now, but we've got new ones. Like, do you have to do anything with an elf every night?

00:02:05 Mark Smith
No, we don't do the elf on the shelf stuff. I mean, we came across it heavily when we lived in the UK, but no, What we do is we go and we forage for a Christmas tree, right? That means go over a neighbor's farmland, find a tree, get one for the neighbors and stuff at the same time. And so that's a whole day trip, well, evening trip experience. I get to cut it down and set it up. Quite different when we were living in London, right, where you just, go to the corner store and there is a beautiful wrapped Christmas tree that you come and put in your... And then after Christmas, I was always surprised in London then a couple of weeks later, they would run a big cleanup service of Christmas trees and you'd take them all to your local park and they'd process them off. Yeah.

00:02:53 Craig White
They still do even like the, I mean, I'm quite remote in the UK, but even we have that service down here where like, yeah, if you leave your real Christmas trees out, they'll come up and they'll dispose of them in a nice, friendly way, which is quite cool, right? I think that's a pretty neat thing to do. I've always wanted to get a real tree, but just like the mess they create, it's pretty epic, right? So at least my fake one, like there's no mess to clean up.

00:03:15 Mark Smith
Well, the thing is, I keep bees now. And when you kind of want to settle a beehive down, you got to smoke them. And pine needles are some of the best type of smoke. It's like a bulk, it's like a cold, white, dense smoke that it creates, which is perfect for for sedating the bees.

00:03:38 Craig White
Nice.

00:03:39 Mark Smith
So it'll have a second life.

00:03:41 Craig White
There you go. Like it. That's good. Yeah.

00:03:45 Mark Smith
So this is the last episode, as I said before, that I'm recording for the year in 2025. 2026 is the next cab off the rack. What are your thoughts about this point in the year and when you look ahead to 2026? What are you thinking?

00:04:01 Craig White
I'm thinking agents and AI is all I'm thinking in terms of technology and work, right? I think 2024 was, is it just a buzzword? Is it real? 2025, I think people are like, yeah, okay, this AI thing is pretty well is here to stay. 2026 feels like when people actually start doubling down and taking it seriously. That's kind of where I'm at. And personally, I think I'm going to follow that trajectory, right? Just you tend to follow where the noise and the tech is. And I think sometimes we can kind of get absorbed in all of the hearsay and whatever else. But yeah, it's an exciting time to be in tech, not just from a career perspective, but at home, like sitting and showing my daughter different things she can do with agents and co-pilots. She loves horses and she likes creating PowerPoint slides because that's what daddy does because that's what I do for a living. I do lots of PowerPoint. And she's like, I want to create a PowerPoint slide based on horses. So I'm teaching her a natural language, knowing that that's going to be a skill that she's probably not even got in school yet, that she's going to be using from here on in her own life. Like AI is such a powerful thing that even in our personal lives can be introduced into households. And they have been for years, right? Anyway, most of us have got Alexas and whatever else, but just to see my daughter kind of embracing AI and agents and we built an agent the ceiling in Foundry. looking at horse stuff. just for mucking around, my learning, but also for her to be exposed to like, this is the future. This is what you're going to be growing into. These are your expectations when you get to a workplace. If you haven't got a copilot license, if you can't build agents, you need to go somewhere else. And giving her those expectations. And when she gets out into the professional world, this stuff is real and this is what she needs to be looking at. I think next year it's going to be doubling down a lot of that career wise, personal wise, it's like it's here, let's embrace it, let's have fun with it, right?

00:05:57 Mark Smith
Yeah. Interesting. What have you been doing for the last five years? If you had to summarize what your work life was for the last five years, what was it?

00:06:05 Craig White
Mainly governance, power platform governance, helping organizations to govern their power platform estate so they can go and build cool stuff, basically. In A nutshell, it's probably what I've been doing. And that's now expanding into AI straight agent governance. I think that's the kind of the big shift from what I'm doing and what I'm seeing is, okay, people now need to govern these things with agents. It's a nice extension from apps and flows, right? Because in the Microsoft world, we've got this thing called Copilot Studio, Agent Builder, lots of other cool stuff. Okay, people want to do the fun bits, but how do we do it securely? How do we stop it leaking data, so on and so forth? So Yeah, my last five years is just sort of nicely evolving into what I suspect to be the next five years and helping people with their AI and agent governance, not just the power platform governance, right?

00:06:53 Mark Smith
And so when you're engaged with customers, it was mainly then around governance. What were you doing internally for the organizations you work for?

00:07:02 Craig White 
Probably not a lot for where I work at the moment, I'm Microsoft partner. So most of my time is spent with customers from other organizations to then get their stuff sorted out and up and running. But before I was in partner land, I was that one person that, because I built an app and a flow once, hey, here's a whole platform and a service to go and govern. So when I was working end user, I was that one sole focus who was expected to be a developer and a product owner and an administrator and whatever else, because no one really understood the Power Platform at the time. You could argue, I'm sure people still do, because it's, I still think that top end of the business are made-up of, and this is generalizing massively, right? But 20 years ago, people that were IT service engineers working with on-prem are now the majority of who take up the senior position. So they're used to all this on-prem stuff, exchange admins, database administrators. So when those roles went to the cloud, they understood those things. But the Power Platform is just this sort of thing that came along cloud only. No one really knows what to do with it and hasn't really invested in it. So yeah, I think lots of people have been in my position, right? You've built a few apps and flows and agents here go and govern it as well, which is a whole different beast. And that's what I try and help people with because it's quite important, I think. Like we said, what does next year looks like? Well, agents are going to be big, right? So we need to make sure people can build them safely and not do dumb stuff, basically.

00:08:31 Mark Smith
So what's the evolution you've seen from a governance perspective? And where do you see governance really going in regards to both the Power Platform and the impact governance will have across the AI landscape that an organization might have?

00:08:48 Craig White
I see the change and the shift in two ways. One, from an administrative perspective, You can't just get away with, if you're looking for agents and governing agents and AI, you can't just put in a toolkit and configure a couple of policies anymore. And that's not discrediting any Power Platform administrators who have been working hard in the last few years. I know there's lots of them out there. But my point is, you've now got a lot more you need to know and understand if you want to govern agents in your Microsoft landscape. You've got to know a bit of Entra. You've got to know a bit of Purview. You've got to know a bit of Microsoft 365 admin. You've got to know Power Platform admin. And all of these tools and things do different things for different reasons, whether it's managing identities or different settings or different types of agents, different types of APIs. The shift now for administrators is massive and you have to learn quick because the penalties for data being leaked is bigger than ever, right? But from an organizational perspective, you can't afford to stand still and hide behind that sort of, oh, no, governance, I can't do this, or no, I'm not going to use agents, I'm not going to use a power platform, I'm not going to use AI. I don't think you can get away with that. People can get away of not doing the power platform. If you don't want to build apps and flows, that's cool. I don't think organizations can ignore agents. So not only have we got to think of the wider governance perspective from an administrative point of view, but organizations need to start really thinking about how can we stop governance from slowing innovation and get on board with this stuff because it's not going to go away. And the quicker people start getting on board of agents, the quicker they'll understand, oh, look at all the cool stuff it can do. So I think there's two big shifts there from a illustrative and from a business point of view of, you said earlier on, what does 2026 look like? I think businesses have to get really involved in agencies. They have to start thinking about, we can't not purchase additional licenses anymore. We can't ignore this big thing called AI because we're going to lose competitive advantage. We're going to use staff. We're going to lose all sorts of stuff. I think that's a big shift, I think. And maybe that's a few misconceptions that need to be broken down, right? Governance doesn't have to slow innovation. You can still have some cool stuff, well managed, but go and innovate and do the cool bits at the same time, right? And maybe that's where the shift needs to come.

00:11:04 Mark Smith
So yeah, so what are those misconceptions that you're seeing, you know, with the clients you work with?

00:11:10 Craig White
I think it's just, maybe not misconceptions, it's more just being slow on the uptake. I think we work with a lot of industries in the UK that probably have quite a lot of technical debt. So they've got massive roadmaps that aren't just AI or low-code or whatever. They've got a lot of network issues, telephony issues, they need to invest in the service desk or whatever else. So with this whole AI thing coming along, I think it's like, well, we can't fit that into this five-year plan that we've had. So we're going to go do all this other stuff first, and then we'll worry about AI into that. I think it's maybe less of a misconception and more of like, no, you need to do it now, or you need to weave it into your existing roadmap of how AI can augment it. How can you use agents to do some of the stuff in your new world, your new infrastructure? I think it's more of a case of people just need to get going quicker rather than just sort of dragging their heels a little bit. I think maybe that's, I still get the sense that people sort of like, they think it's going to go away. They think agents is just this fad that's going to disappear. I don't think it will. Maybe that's a misconception in itself, right? This is not a phase. This is, to my mind, this is like as big as when the internet happened. It's here and it's only going to get bigger. So the quicker people get used to that, I think the quicker their businesses will start to grow and flourish.

00:12:29 Mark Smith
I know there's a lot of folks that have worked in the Power Platform space for many years that are wondering what the future holds when it comes to AI, and particularly in context of the Power Platform. And I feel like, the Power Platform, as we know, it's been around 10 years now. Power Apps came out 10 years ago, and then the Power Platform, you know, and it turned into such. So, And I think that anybody that's worked in the space has absolutely set up for the AI world and what we're going to do in AI. Because if you look at some of the parallels, we had center of excellence as a framework to operate within and really scale apps, digital transformation inside an organization. And now I see this move to AI-based center of excellence. Will we see things like Power Platform administrators becoming AI administrators. Well, we see this kind of transition of role, and I'm wondering what your thoughts are around folks that are typically being, let's say, Power Platform architects, so application architects, or they've been in the functional consultant type role, building out low-code solutions. When they look at their careers, in your mind, what should they be focusing and investing their time in now? to make themselves relevant as we see, right, the world transition much more this way. How do you see those architects, administrators, developers, keeping their skill level up? What should they be doing? What are your thoughts?

00:14:18 Craig White
It's a hugely important question because, we're all, I've heard it or might have even subconsciously thought it at some time, AI is going to take my job. And I'm a firm believer that AI will not take your job. AI will just change your job. And how can it improve the way we do things we think? And I just kind of like break it down to three things. Like from my own perception, this is what I've done in the last sort of 12 months or so is look at 3 areas, right? Number one, skills. I've got a lot of skills to work in the new AI world, building agents, building, like a lot of Copilot Studio is Power FX. Well, I know that because I've done it for the last 10 years in Canvas Apps. Under the hood, it's Power Automate to an extent. Okay, well, I know some good stuff around there. So I've got transferable skills, but the skills I need to add are things like prompt engineering, how to orchestrate agents to kind of hand off and do certain tasks. How do I compartmentalize Agent design compared to it might be completely different architecture to what I'm used to. So how do I design A agent to agent system or a multi-agent orchestration? How do I become an agent boss? These are kind of skills I need to learn and understand. Property engineering I say is probably one of them. And obviously I've been working with Mr. Huntingford for quite a while. I know he's huge on red teaming. And that is a skill I think people need to learn and understand to help coach others. If we're building agents for people, how do we properly test them? It's not just a test plan and ADO anymore. You've got to expand how you test agents because it's completely different. There's skills, and I think those are some skills. I think there needs to be a mindset shift in how to use AI and when to use it, because I don't need to do all the manual lifting anymore. I've got so many AI tools that can write code for me, that can wireframe for me, do architectural diagrams. So my mindset needs to shift from, I'm going to open up this blank canvas and do stuff manually to, I'm going to get AI to get me to a certain point and then validate it. And I can use my experience to validate the outputs because I know if co-pilots can tell me something wrong because I know that bit of code won't work or that idea doesn't make sense because of my previous experience. So mindset shift, I think there's another one. And finally, I think it's platform knowledge. You just need to understand that there are now more moving parts to you can't just get away with building an app anymore. Like I said, a bit like administrators, you've got to think of, you know, you can now register an agent as an ID and Entra. Okay, so I wanted to know a bit of Entra. There are responsible AI controls that sit in purview. Okay, I've got to understand a bit of DSPM for AI, for example, that sits somewhere differently. I need to learn that and understand it. Some agent controls live in Microsoft 365 Admin Center. Even as a solution architect, I need to know if something's not working, it might be because of setting A, B, and C. And I've got to expand beyond my realms of the Power Platform now to look at other parts of the stack. And also like when to use, maybe this is partly a skill statement, partly a platform thing, when to not think about studio and start thinking about Foundry. And that's kind of the next layer of architecture for me is like what to use and when, because there's so many models, there's so many different tools. How many models in Foundry now? There's thousands, isn't there? I wouldn't even know where to start. I know GPT 4.0 because that's the one I use the most, but there's tons of stuff. And I think that whole shift is huge for, because there's lots of transferable skills. I think this is a new age where we need to start thinking about how do we use all these things for our advantage. And yeah, prompting, I think, is a big one. I don't see enough about that. We do quite a lot of that with customers at the moment. I teach them how to prompt. So I think it's quite good. There's lots of different methods. That's something I've been trying to hone, just for my biggest thing this year, is trying to hone how I talk to AI. And I think that's a big skill shift for a lot of people, I think, is, okay, I've got it, but how do I use it and how do I get the best out of it, right?

00:18:15 Mark Smith
I've always been very broad in the role definition in a Power Platform team across my career. In other words, there's really only been 3 core roles. They are architect, developer, and functional. right? And then, if you're doing a delivery team, I would add to that, let's say somebody, a project manager, I would add maybe a UX designer, a change manager to kind of that team delivery mix. And then you might even have somebody specializing on the DevOps side of things just to round it out. How soon till we have the architect agent? How soon do we have the functional consultant agent? How soon do we have the developer agent? In other words, these are people trained like you and I through our careers have been on the Power Platforms explicitly. They know everything because they can read all the content that Microsoft ever produced around a feature and every blog post and everything related. You know, let's say we've given it a good healthy dose of life experience around failed projects, successful projects, what a good look like, etc.

00:19:18 Craig White
Yeah, I might have had a few of those in my past, yeah.

00:19:21 Mark Smith
How soon do we go, you know what? Why don't we use our functional agent and give it a crack at first run at this project? And even like, project managers is the role of project manager. Could that become, could you have a PM agent that actually runs the entire project?

00:19:41 Craig White
I don't know we're that close to that yet. Certainly from a, let's say, let's get back to the original part of that, the architects and the consultants. This day and age, right, you could sit on a refinement call of a customer, get the transcript, whack it into an AI tool of choice to summarize, build your prompts you can put into Power Platform plans, build your wireframe, right? Wicked. If that application that AI builds, is that enterprise ready for you to go and deploy right away? Hell no. It's a POC at the absolute best. For the unique requirements, you've still got to do a lot of manual extending because what we use AI-wise in the Power Platform is nowhere close to building those greater solutions that we have deployed over the last five to 10 years, if not more. So it's a starting point, but I don't think it's like I said earlier on, I don't think it's going to replace those roles. It's just going to augment and speed things up. It'll get me to a point where I haven't done everything manually so I can get to do the extension stuff a lot quicker because I'm not manually re-watching calls, taking notes, building POCs from scratch. AI will get me to a certain point. But I think I need to validate those outputs based on my experience. And then I still have to extend. So I think we're somewhere away from it being able to replicate some of the stuff we've built over the last 10 years. From a PM point of view, it may be, like maybe some of the administrative hang up. Like I know you've working for a partner, there's a lot of PM time spent putting our weekly hours into a timesheet system of like, this is how many hours you're doing this project each week. I'd love to be able to kind of like have an agent that would sort of like take the scope of work, stick it into DevOps, score it, assess it, estimate it, and then go, here's a potential plan of work for an architect and two consultants over three months. But you still need a PM to be in the loop and reason over it. What about annual leave? What about holidays? What about bank holidays? No, I can't do that here. I've got to move these things. You still need people to assess and just saving time. I think, I don't think it's going to replace roles. And I've probably, there's probably quite a few PMs I work with. I listen to this one anyway, and I'll probably thank me if I go, yeah, we can just get a PM agent. But I think there's certainly elements of all of our jobs that could be improved with AI, but I think that time we're going to be spending validating the outputs for certainly for this, the foreseeable future. I can't see things being so mature that it just completely eradicates what we do at the moment. Personal opinion. What do you think? How do you see it happening?

00:22:16 Mark Smith
I think it's, because I'll give you a couple of feedback of what you've just said, right? What about holidays? What about annual leave, all that type of stuff? Well, that's all just data, right? Why couldn't we get the agent to be intelligent enough to go, you know what, I need to just validate the HR. In fact, I need to go to the person on the gig and say, listen, you're taking any time in the next six weeks. And that might be a routine that runs that the agent goes, you know what, I want to check on that. I'm dealing with a human, they're fallible. As an agent, I was going to monitor that piece, right? And in fact, what? I'm going to profile their sick leave for the last five years and identify that this season coming up, they're potentially going to get sick because they did last year, right? So all of a sudden that little data nuances start to have stuff in memory, if you like, that most humans don't keep in memory around in every individual. And so, you know, you could trade it on the latest and best techniques of what is working in any project methodology, do all the reporting on time, kick the *** of the person that hasn't filled in their time sheets, extract it from a system that they potentially can, find where, with the transparency and honesty to know where a project was going to fail in three weeks time, way ahead of it because protective activity, the customer's not attending meetings, they're not getting sign off, whatever. I wonder if and I'm talking about an agent that is not so much an agent as we see in the co-pilot's world of building, but something that has full control of a desktop or a computer and can do anything a human can do in that computer, right? So it could run power automate desktop, for example, and manipulate anything at a device level, websites, passwords, everything, and we feed it the font of knowledge of the best practices of our organization, and we have it constantly learning what good looks like, I see that potentially there is that agent world where even, I was building an app this week on the Power Platform, something I haven't done in years. And I would go down the path and then I would see, oh, this feature is going to be deprecated in the future. Don't use this way of doing a roll-up. You need to do it another way. And so I can see, with full screen control, an agent seeing that and going, shoot, let's just back check the reference to what happened here. Excellent. Straight away, I wanted to do it. Rewrite as an I was using an agent as an AI the whole time with what I was doing. It said, hang on a second, I'll use some power. Power effects code is the best practice now for doing it. Spat it out, everything worked, tested up fine. in no time at all. I'm just wondering if we have that level of type of interface and interactivity that it's always going to use the best practice technique. It's going to shy away from deprecated techniques, even though in a human, they're the way I've always done it and it's easy and I can clearly do it that way. They're not going to take those because they're like, hang on, I'm creating technical debt by doing that. So that technical debt might be another lever that we build into these. But then even then, let's say a live in production system, The telemetry data coming off that being fed to an agent could be incredibly powerful to keep it tuned to its optimal state, both from a security architecture point of view, an ALM point of view, making sure we're not introducing any issues into the system. I'm just wondering how soon until someone sits down and says, you know what, this is an agent that I could build. And it might, let's start with the entry level roles that we give to grads and stuff coming in, which might be to build the demos. But they're building demos that are so ******** that we go, okay, how do we get this to a production level grade and work our way up through those traditional pillars that we have in a business applications type practice? I'm just wondering how soon that is possible.

00:26:14 Craig White
But you still need the people to build those agents, constantly refine, monitor, govern. red team. Again, I don't think the roles disappear. They just change, right? Because you're taking those skills and repurposing them to say, well, actually, I can go and do all of that cool stuff autonomously with, and I could be an agent boss, but any boss still needs to manage and lead a team. So your role, like my role has changed, right? Recently, I've been promoted. So I'm I'm lucky enough to kind of look after a team. I don't do much of the hands-on anymore, but I empower other people, well I hope I do, to do the doing, which is wicked. It's the same as in the scenarios, like I don't long, I won't necessarily do the building stuff anymore, because natural language could do some of that. But my dynamic is I've still got to go and Maybe I've got to micromanage some agents. Maybe one's misbehaving. Maybe 1 hallucinates. Maybe one is so fixated on doing things the right way that it can't see a workaround that might give me some technical debt, but is the only solution to a business problem. And it hasn't got that in its language model. And I have to intervene as a human to go, yes, take this route that might not be best practice, but it's going to solve a problem and get a project over the line. I think our role will change in how we take, develop, manage, govern, and then interact with all of these agents. I'm excited for that future because it would be quite wicked just to sort of sit there and like, it was like the, is the guy at the front of the orchestra, like a conductor, right? It's almost like you've got like a sea of agents doing different things and you're just sitting there at the front with your stick, wiggling it around and like getting them to queue off and do different things. I'm quite excited by that. Although, yeah, I do want to do the building bit as well. I think architects and consultants will still have some fun. building out these concepts to save time that others can use, but just being aware that the fact that they still need to be close to it, test it, refine it, so your work is probably never done. When have you finished? When have you built something that is finished? I don't think I ever have. There's always things I could do to improve something, even when it's productionized. There's always a backlog in my mind of, oh, if any could do that, oh, I didn't get time to do that. Oh, this nice to have. I think it should be a should have. Do you know what I mean? There's always things you could do to improve stuff. I don't think we'll think any differently when it comes to aging. And maybe that's a good mindset that we can carry over, right?

00:28:36 Mark Smith
So you know how you're leading a team now. If your team came to you in January 2026 and said, listen, let's say just one individual came to you and said, what advice would you give me around the future of my role? Let's say they're a real deep power automate expert, deep skills in power automate. Do you see that in two years' time, we're going to need people that are deeply expert in power automate and just have that as their maybe primary school role in a team? Or do you see that being abstracted away, which I do, that agents will do the interaction and I won't have to worry about a clunky power automate UI anymore? I'll be able to speak into existence what I want and it can use that clunky, you know, UI behind the scenes. but I might never see it. What advice are you going to give to your team members in 2026 if I said, listen, these are the three skills you need to just dedicate and master on top of your existing Power Plat experience? What would you advise them?

00:29:43 Craig White
Again, I'm probably sounding like other people you might have spoken to recently, but we've certainly seen this education of, we can fixate on that little thing is building the agent. But how do you test it and how do you get your data right and secure in the 1st place? So if I was to kind of mention three things, I'd be like, one, understanding data security, whatever shape that takes, whether or not you're a Power Platform expert or someone who's still mucking around with VB and spreadsheets, data and security are the fundamentals in IT. No matter what role you do, if you don't understand those two things, in my opinion, everyone in IT should go and see SC 900. Just get an understanding of zero zero trust, principle least privilege, it will serve you well, no matter what domain you're doing. So that's one of my bit of advice. And I think even us experienced veterans can sometimes get a bit, I don't know, not complacent is the wrong word, but maybe we're just sort of ground down by dodgy practices over the years we might have learned and maybe just sort of regrounding ourselves into what does good look like for data security. That would be a starting point. Testing. I think we can all pump out some good stuff and we can speak into a tool that are built as an agent. All right, wicked, but what is your mindset for testing it? And again, talking about that red team, I think that's a huge skill. We worked with a customer recently, it's really good, very, very simple knowledge base, retrieve an agent over some HR documents. I won't say the company, but it's pretty funny, but also pretty cool. And he was like, okay, have we got any social events on? Yeah, cool. Can I meet people? Yeah, cool. Can I find myself a girlfriend? started asking this agent for relationship advice and could he find himself a girlfriend? Is he going to have a good night tonight? Just based off a SharePoint document library with some HR policies in it. That kind of mindset to test an agent and what kind of stuff is it bringing back? I think that is skill hugely #2. Forget the actual tech. Think of the mindsets in terms of data security, testing, whether that's like prompt injections and stuff. And maybe you can even build an agent that does your prompt injection test. I don't know, but understanding that sort of bigger concept. I think those are the two main ones. Maybe prompt engineering, I think, is still something that I would like to see people. I want to try and make it so people aren't just necessarily focused on tech. Like what are the skills that are going to serve them well for the next 10 years? It's going to be data security, prompt engineering, red teaming. You can apply that to any low code tool, any agent you see on the web, go to find a random agent that people have bought a car for like a quid or something like that, because they're just going randomly batter an agent on a website, just go and do stuff like that. Like that skill is available anywhere, regardless of the platform. Those are the three things that I would say double down on.

00:32:34 Mark Smith
I think it's a massive tip, the SC 900. as a really good grounding security. Yeah, I like that. I love that.

00:32:45 Craig White
I think more than ever, it sounds, I'm sure there's people listening to this going, I've been working on IT for 30 years. I don't need to do that. I know my security. I'm sure you do. I'm not disrespecting anyone by mentioning that as a bit of advice. But even myself, I think I was lucky enough that I knew 90% of that exam and the course because it's just what I I've been lucky in some respects that I've worked with some good people that taught me some good practices and lots of bad ones, but I still learnt some 10%, or as a bit of a refresher of like, Oh yeah, I forgot about that. I need to kind of, you know, remember, and then pass those messages on to people I work with, and... For those new grants and apprentices that you mentioned, maybe we can get an agent to replace them, but I can still teach them some of these wider concepts. So when they're building stuff, they've got security and whatever else at the heart of everything they do, right? Which I think sometimes we might forget about and just focus on the business problem 1st and building the cool stuff and not the security and the data behind it, which is arguable important, right?

00:33:40 Mark Smith 
I love it. As we wrap up, What's the one thing on top of what we've discussed today that's going to be your kind of focus in 2026? You personally?

00:33:53 Craig White
Me personally. I think just getting further into, me personally, I'd like to get more into Foundry. The reasons being, I think I'm quite lucky that I've had a few customer engagements over the last few years where I've managed to build some stuff in Copilot and Studio and what was Power Virtual Agents when it came out. I don't know how long it's been around. Was it 2019, 18, I want to say.

00:34:19 Mark Smith 
Nah, it's much, it's, no. Well, sorry, I built my first Power Virtual Agent the day it got announced. And I just got back to New Zealand, so that was November 2019.

00:34:34 Craig White
Yes. I remember building my first one of asking, I think I had a guy sitting next to my team and he was mucking about with stuff. So I set up an agent that called him some rude names if he put in some keywords or something like that. It's like, yes, I put an agent. Very inappropriate use case, but I played around with it. So I'm lucky that I've covered a fair little bit of that lower level agent stuff, if you like. My personal goal next year is to get more into what Foundry has to offer, understanding the different models, the different licensing, the different capabilities. For example, having more control over your responsible AI filters is a big thing I've learned recently in terms of you can control that in Foundry, you can't in Coponix Studio. So that's my personal goal, is just to get closer to the usual landscape side of things and what AI looks like in that part of the service. Because I think That's just a natural progression for me and something I'm excited to learn and actually already started. Why wait till 2026? Just get on with it right now, right?

00:35:34 Mark Smith
Hey, thanks for listening. I'm your host, Business Application MVP Mark Smith, otherwise known as the nz365guy. If there's a guest you'd like to see on the show, please message me on LinkedIn. If you want to be a supporter of the show, please check out buymeacoffee.com forward slash nz365guy. Stay safe out there and shoot for the stars.