Magnus Gether Sørensen's Tech Journey and the Role of Generative AI

Mark Smith: 0:19
Welcome to the MVP show. My intention is that you listen to the stories of these MVP guests and are inspired to become an MVP and bring value to the world through your skills. If you have not checked it out already, I do a YouTube series called how to Become an MVP. The link is in the show notes. With that, let's get on with the show. Today's guest is from Denmark. He works at Delegate as a Managing Specialist. He was the first award as MVP in 2022. He has got the nickname of XRM tooling wizard for creating developer tools. His goal is to bring best practice from both the academic and the real world into the world of business applications. You can find links to his bio and social media, et cetera, in the show notes for this episode. Welcome to the show, Magnus.

Magnus Gether Sørensen: 1:16
Thank you, thank you.

Mark Smith: 1:20
I'm pleased to have you on Like. I remember observing you at MVP Summit. I call you the flashiest dressed MVP in the game. You always look smart, put together on point.

Magnus Gether Sørensen: 1:32
I always feel comfortable in a suit. You know I come from a consultancy background and it's sort of part of the uniform and of course if you have off the rack suits it's not as comfortable. But I have tailored suits so it's comfortable in any weather at all.

Mark Smith: 1:52
So yeah, very very classy, Very very classy Tell me about food, family and fun.

Magnus Gether Sørensen: 2:00
Oh, food, I love food, just any food at all. My favorite food is just buffet. You know, if you ask anyone, they'll tell you how much I slaughter food. I've had a lot of issues with that. When I was pretty inactive I gained some weight. But during my paternity leave I walked a lot with my daughter and I think I lost 10 kilos or something like that. Yeah, yeah, just walking. I walked 600 kilometers in. You're American, what is a kilometer? But 600 kilometers in like two months or something.

Mark Smith: 2:35
Kilometers works. I'm not American, I'm Kiwi, so we are in kilometers. Perfect, that's perfect, nice nice.

Magnus Gether Sørensen: 2:44
Wow, that's a lot of walking. And yeah, I have my wife. We met in high school, so stuck with her ever since. Still love each other for some reason. We're insane like that and our daughter just turned one year just a week ago.

Mark Smith: 3:03
Very cool, very cool, I love.

Magnus Gether Sørensen: 3:08
Denmark yeah, I love it as well.

Mark Smith: 3:13
I don't want to leave. So my wife and I were in Denmark and we had an Airbnb and we did a design thinking exercise on the wall to set out where we wanted to be in the future and we put up all our dreams, aspirations, desires. We'd been traveling the world for some time at this point. We recently went and looked at the photos that we'd taken in that Airbnb. Every one of our dreams have come true. Every one of the things we you know we didn't own any house, property, land, we didn't have children All the things that we desired in our careers, and everything. Everything came to pass that we put in our like our five, 10 year vision, that we put up in that Airbnb in Denmark, in Copenhagen.

Magnus Gether Sørensen: 3:57
Crazy, crazy. So just the atmosphere, it's the atmosphere man, it's a great city.

Mark Smith: 4:02
I've got two nephews that are Danish. My brother married multiple Danish women over his life. He seemed to have a taste for them, and so, yeah, I've got two Danish nephews. They'd be in their late 20s now. One's in the military there and the other one's into, uh, audio production. Um, but, yeah, yeah, yeah, yeah, so I do. I do have a, uh, a connection to denmark. Tell me about how you got into tech uh.

Magnus Gether Sørensen: 4:34
So I originally thought, you know, when I was in what was that 9th grade? Or something that I was going to do a lot with money and finance and so on. But then there was this high school teacher in the first year of high school that showed me I could do websites and build them on my own, and I was gaming a lot back then 12-14 hours a day in world of warcraft so having a job that you can do on your computer sounded perfect for me. So began building websites for small businesses. Uh, began building apps and so on and uh, yeah, things just rolled from there. Yeah, things just rolled from there. Became a dynamics developer during my university studies and began building tools from there. Yeah, just stuck with it ever since.

Mark Smith: 5:28
So when you say dynamics, are you on the F&O or CE side?

Magnus Gether Sørensen: 5:31
CE side. Yeah, funny thing about that, though. I've never touched a first-party product in my life.

Mark Smith: 5:37
Wow, so you just do model-driven apps, so you use a clean dataverse. Perfect thing about that though I've never touched the first party product in my life. Wow, so just you do, you just do model driven apps so you use clean dataverse perfect, perfect. Um, it's uh, I love that. I love that.

Magnus Gether Sørensen: 5:46
Tell me about xrm tooling wizard yeah, the idea was just, you know, I saw everyone who had some fancy name and then I thought, the people that I saw who built tools, they built you know one or two, but at that point as a student I I had built one and I was maintaining three others. So I mean I I thought I could grab that and then later on I just dropped the tooling part of it because I mean I knew a lot about the platform anyway. So I mean I loved Harry Potter and XM Wizard seemed fine.

Mark Smith: 6:21
So I love it.

Magnus Gether Sørensen: 6:22
I love it. I also didn't use LinkedIn for anything at that point, so my LinkedIn tag is also XM Wizard. I don't know when that's going to bite my ass at some point in 40 years or something.

Mark Smith: 6:33
But I think it's very cool. I mean you can change it at any time and so you know when you look at what you do in the community, kind of what are you known for.

Magnus Gether Sørensen: 6:46
What do people come to you for? Deep technical knowledge? I would say so. So my, the tool that I've built is just a copy of a dataverse basically. So it like fake XMEC. It knows how to run business logic from Dataverse locally so developers can test stuff without contacting a real environment, and in that I have had to experiment a lot with what Dataverse can do and all the quirks and so on. So of course that gains you some knowledge. I have built an interpreter for the old classical workflows so on, so of course that gains you some knowledge, like I've built an interpreter for the old classical workflows so they can run locally. That was a lot of fun, you know, pulling out the definitions of the workflow and looking at how it works and so on.

Mark Smith: 7:30
So when you did that, did you kind of because that was built on Windows Workflow Foundation, right. So do you have a local instance of that running? Are you using that or did you create a full new version?

Magnus Gether Sørensen: 7:44
Just create my own custom interpreter in C Sharp. So taking that SAML definition, compiling it over to some classes In academia, you would call it an abstract syntax tree and then you run an interpreter on that and then, yeah, then you have an interpreter. It took two months to build. You know as my part-time stuff doing as a study.

Mark Smith: 8:08
That's epic. So what are your thoughts between Power Automate and what was provided in that workflow tool that was part of the original Dynamics product.

Magnus Gether Sørensen: 8:24
The most sad part is that Power Automate is asynchronous. So if you want to Google my name, you can probably find some super old article where, when Power Automate came out, I said that it would suck, it wouldn't scale and so on. And the whole point is that if all of your business logic is running asynchronously, your data quality is going to go bad, and that was the great part about the classical workflows was that they can run synchronously.

Mark Smith: 8:51
You know they only could run synchronously. I thought towards the end. I don't remember the OG product having the synchronous functionality, like back in 20 crm4 days and things like that.

Magnus Gether Sørensen: 9:04
Now it could be wrong, but I thought, but at least the 2011, they had it so yeah, yeah power debate is pretty old by that standard, so they haven't followed suit so, so.

Mark Smith: 9:16
So let's unpack this because it was interesting. Ryan cunningham coined this term in the fullness of time that it would achieve that, and Steven Siciliano, who then was the Power Automate owner for some time, coined that some years ago when it first came out, because our biggest complaint was when are you going to give us parity with what we had with Windows Workflow Foundation in Power Automate? Was, when are you going to give us parity with what we had with windows workflow foundation in power automate? And of course, we're now. What? Uh? 2017, 2016? That's a long time ago, right where? And it still hasn't come about. Why do you think it's one not being a desire from the product team to get the synchronous functionality in there? Is it just too big a challenge for them to solve in the architecture they've created with Power Automate?

Magnus Gether Sørensen: 10:10
I mean we've also created we. I say I have a colleague who created an interpreter and I gave him comments on it for CloudFlows. So there's a pretty big difference in how they're written. They are much more abstract, so it's easier for them to add more connections and so on, but it also means that it's harder for you to run an interpreter that doesn't spin up stuff. So, inherently, power Automate interpreting is slower than classic workflows. Currently, power Automate interpreting is slower than classic workflows. So I think a much more fast and easy transition would be the low-code plugins or whatever they might be called in the future. Right?

Mark Smith: 10:55
Okay, so is that the future way to do things you're thinking?

Magnus Gether Sørensen: 10:58
Yeah, I think so.

Mark Smith: 10:59
Low-code plugins addresses the scenario and the problem, though, I suppose, with the low-code plugins, which you didn't have with Workload Foundation, was a GUI that allowed a quote citizen developer to build them.

Magnus Gether Sørensen: 11:18
But I think with natural language, helping writing formulas and so on, I think having a GUI is more of a hindrance than just having something helping out writing formulas. I think the actual media that we interact with will change in the next five to 10 years anyway. I think everything will be more abstract, right, but the Power Fx will have the same strength in that it can use all the connections, but it's inherently more lightweight. So if you think that you can still have named formulas and so on so you can break down your big complex logic, but you can still do the same stuff that you can call the environment, so you can call custom APIs or actions that you can patch and retrieve and so on, the real big gap right now is that the triggers are not as sophisticated as plugins. If there was full parity for these local plugins as plugins, that they can have access to the same plugin execution context, then I don't see a reason why a lot of the simple stuff wouldn't be written in that Interesting, and why do you think that is not there?

Magnus Gether Sørensen: 13:03
I just think it hasn't been a focus. I mean, you must also have realized that as long as as well as everyone who's listening that AI has been a great focus. I think everything that was on the board two years ago has been pushed. But looking at the release wave that we've seen now, it seems like they're picking up steam again and actually fixing stuff on the platform, but there's been a one and a half year gap.

Mark Smith: 13:31
So I made a prediction last week that Power Automate will disappear into the background in the next five years. Developer or configurer. Because I believe that, as speech becomes ubiquitous in how we interact with AI, that although AI will build out the automations for us, there will be no need for us to go in and build out an automation as such as that we would, you know, maybe hit a record button on our screen and show kind of what's already been proposed, or literally just talk to it and say I need an automation that does this and it might, you know, schematic that up on screen for you and you might yeah, I need this to happen at this point and basically verbally go through and then behind the scenes, it'll go okay, I need these connectors to do that, and then, if they're not, don't exist, it'll build them and, and you know, you'll provide access to data sets etc. Um, it'll have a full formula engine in it.

Mark Smith: 14:39
Blah, blah, blah. Do you see that? That's, you know, the, the demand. For you know, in the last five, ten years I've seen a lot of people put you know their power automate guru on their profiles and stuff and really built a brand around that. Do you see it all kind of just disappearing into the background. As in that, ai will be our interface.

Magnus Gether Sørensen: 15:01
It depends on how slow we as consultancies and so on are changing, but I definitely think that the future is that direction. The future will be that systems describe how data should be understood in their domain, and so that would be kind of closed source. So let's say you have a warehouse system In your system. You know what an item is in a warehouse, but how an item gets in there. You might have a single interface for adding items. But from the rest of the business perspective, you don't really care.

Magnus Gether Sørensen: 15:42
Your solution that you've built still on Dataverse, still on so on, you have all your business logic that runs, hopefully, synchronously, so your data remains intact, is is full of your domain knowledge, and then you have these points in. And that's where your point is right that right now we would put your Power Automate guru in there and have them describe exactly how these different systems should connect. I definitely think that layer will be blurred out. As long as your warehouse system can describe how to interact with it well enough, such that an AI can interact with it, there's no reason why you need the logic.

Mark Smith: 16:26
Have you looked at semantic kernel?

Magnus Gether Sørensen: 16:29
No, I don't think so.

Mark Smith: 16:32

So Microsoft announced it in July this year and it's pretty mind-blowing because I watched last week a demonstration of a shopping cart scenario where somebody says you know, the shopping cart metaphor, right, based on kind of a supermarket checkout, is cumbersome and particularly if you're elderly, right as a necessary, understand that, and so I noticed this example. So, using voice interaction, it said you know, somebody goes to the websites that has a catalog of products and just said I need some boots. And they showed that in semantic kernel. Well, what's a boot? Right? So, using generative AI, then it can say a boot is this type of thing.

Mark Smith: 17:14
Oh, we, and you hand that to the API, the product API, and says okay, do you have a product that has a dimension called boot? Yeah, I do. Well, can you display those on screen? Great, there they are. I want a black boot? Well, it has no concept of black, yellow, green, but we can go, hang on a second. What is black? And your generative AI goes well, black is a color, is a color. And so, okay, do we have a dimension called black?

Mark Smith: 17:42
Yeah, here's a, you know, and all of a sudden, the way you're going to interact becomes pretty compelling, right? Well, I'll have two of those and, by the way, just use my credit card to to complete the, the purchase, rather than you know. And you might have a validator saying put your cv, cvc or whatever it is, you know code in what. Might have a validator saying put your CVC or whatever it is, you know code in what is it and it just give it to you. It might give it to you auditory and all of a sudden the way we can create experiences for people massively changes by using this kind of. So check out Semantic Kernel. Anyhow, I'm a big fan because it's kernel. Anyhow, I'm a big fan because it's very human, readable code and what it gets, what's generated. And I just see a lot of runway in the area of the power, platform dynamics, that type of thing of where this is going to be incredibly powerful tool to create new experiences yeah, and of course it's.

Magnus Gether Sørensen: 18:36
it's easy to get hyped right. My big worry is that I can see the vision. I can see that there will no longer be a Canvas app, a model-driven app. You just have a blank screen and it auto-generates whatever UI you want. But I am much more interested in how close are we now? How close is Code Interpreter in ChatGPT to be able to generate a React app for me that can interact with an API? It's not there yet, but I would love for these examples to be what we study, because it's super easy to talk about five to ten years. Who the fuck is going to check up?

Mark Smith: 19:18
Yeah, exactly right.

Magnus Gether Sørensen: 19:19
It's much easier to figure out. What can you do now? You can point Copilot to your Dataverse database and it can answer some stuff. It can probably not answer stuff if you have a complex table structure. So it's not there yet. It can probably get there, but there are a lot of scenarios, like user guidance, where you can point it to some documentation you have and so on. That might be a good idea. How do I do stuff? Well, if you have written pretty good documentation, why not just give it to Copilot and then it can figure it out?

Magnus Gether Sørensen: 19:57
So a lot of these. I need to read a lot of documents in order to understand something. It can do that now, and I'm guessing you must have read a lot of the examples of different businesses that are using generative AI now in success A lot of them. We've also built a bunch of these systems.

Magnus Gether Sørensen: 20:24
They have a lot of cases and usually there are a lot of documents tied to it, and it makes it a lot easier for them to read a summary from generative AI and then they can dig into the documents if they need to, rather than they have to figure it out, and a lot of these cases also have pretty short SLAs, which allows them to actually read through stuff and maybe win more cases and so on. Maybe, yeah, solve them faster or at least you know, do them thoroughly enough, follow company policy and so on, a lot of these different companies. They aren't following their own compliance internally because there's not enough time, right? So that is what we can do now, and I'm much more interested in what is the next six months in the year, because I don't think anyone knows any more than that anyway.

Mark Smith: 21:14
Yeah. So with that in mind, final question for you is what are you learning, what are you studying? What is your focus? That is only six months out.

Magnus Gether Sørensen: 21:28
Right now it is a lot on how we can do so. It's not AI focused at all. Actually, it's how we can get to the actual treat your environments as cattle and not kittens story. So Microsoft has been pushing this narrative a lot, that they want you to spin up environments whenever you create a new feature and then merge it in and then deploy it. But the story hasn't really been there.

Magnus Gether Sørensen: 21:56
But I do believe that in the next six months to a year we will get to that point and I can do my part from community tooling that a lot of community tooling right now needs an environment to run. But if the stuff that is stored inside of your version control is good enough, in a good enough format, then we can also point to that instead. Stuff that is stored inside of your version control is good enough in a good enough format, then we can also point to that instead. And if the act of spinning up new environments becomes faster, if we can configure them correctly, then it becomes much faster and safer to be a lot of developers on the same team. So right now I'm on a project where I think we are 15, 17 developers on the same environment, right, and we are stomping on each other's toes. We have reduced the stomping a lot because we can do local testing and so on, but we're still creating these fields on the same entities. That will just cause problems. So that is my focus I want environments to be cattle.

Mark Smith: 23:12
How do you solve that? Because what you just described there, I can remember in CRM four days that being a problem. In this case it would be three different Microsoft partners working with the same customer and you would get this overwriting happen of solutions or customizations built in the system, being the last person in wins type thing, and it would create problems. How far down the solved path are we?

Magnus Gether Sørensen: 23:43
wins type thing and it would create problems. How far down the solved path are we? We will be far down soon. So the reason I can't say specifics is that it's a private preview running right now, but it should have been announced last week, I think so. So it's out soon. It's the new uh git solution structure, right, it's. It's in the release wave, so we can say that much um.

Magnus Gether Sørensen: 24:08
But that story will become much better. And previously it's only been a demand from a developer point of view, because we know better I mean, get off my high horse and so on. But we know how software development can be done in parallel because we have version control and stuff like that. And if that story gets better from a platform point of view and it becomes easy for people who just sit inside Make, then the storytelling gets better at least. And it's also something I will talk about in the new year at different conferences, because it is such a pain that you can't be many developers on the same project, right. And yeah, just the last part that I don't expect Microsoft to fix is just the community tooling part. If we have a good enough story that matches what Microsoft is doing, then we'll also keep investing in it. That's how it's always been.

Mark Smith: 25:13
Hey, thanks for listening. I'm your host business application MVP Mark Smith, otherwise known as the NZ365 guy. If you like the show and want to be a supporter, check out buymeacoffeecom forward slash NZ365 guy. Thanks again and see you next time. Thank you.