Improving Your AI Skills for Fabric - Ep.522 - Power BI tips

In Episode 522 of Explicit Measures, Mike Carlo and Tommy Puglia unpack the latest Power BI and Microsoft Fabric topics from the show. You’ll get a quick read on the episode’s biggest ideas, why they matter, and where to dig deeper in the full conversation.

News & Announcements

• No linked announcements were available in the episode description for this post.

Main Discussion

This episode covers the major themes, opinions, and practical lessons Mike and Tommy surfaced during the conversation. The transcript below captures the full verbatim discussion if you want the exact phrasing and context.

• Mike and Tommy react to the episode’s biggest Power BI and Fabric developments and explain what stood out to them.
• They connect product announcements to day-to-day practitioner decisions instead of treating the news as abstract roadmap chatter.
• The conversation highlights where teams can move quickly, where they should slow down, and what tradeoffs deserve attention.
• They share candid perspective from real project work, which gives the discussion more practical value than a headline recap alone.
• The episode mixes tactical advice, opinionated takes, and a few forward-looking predictions about what listeners should watch next.

Looking Forward

If this episode’s topics affect your current Power BI or Fabric plans, use the transcript and linked resources to identify one concrete change you can test with your team this week.

Episode Transcript

0:23 Hello everyone and welcome back to the explicit measures podcast with Tommy and Mike. Tommy, good morning. How you doing? Oh man, it’s good to see your face again. It’s only been a week and I already missed you. [laughter] That’s funny. Tommy and I do talk a little bit when the show isn’t on, but we talk more probably on the show than off the show anymore now. Well yeah, we get sick of each other if I had to call you every Wednesday and Friday to not do the podcast. That’s true. Also, you’re really if I ever need Tommy, it’s really hard to get a hold of him when he’s when I’m not on the podcast. If it ain’t on a calendar, it ain’t happening.

0:54 ain’t on a calendar, it ain’t happening. So So Tommy’s picked up his phone before and he’s had like six missed calls and he’s like, “What the heck do you need?” Yeah, [laughter] I was like, “Are you alive?” Yeah. Yeah. “Are you alive? You You’ve totally gone dark, Tommy. Your Teams won’t work. Your cell phone doesn’t work. What’s going on?” He’s like, “Yeah, I’m I’m busy doing things.” though, you and I get along and I’m not saying that online just so people think we do, but there’s actually one of my favorite radio that I used to listen to or the shows was called Mike and the Mad Dog in New York and they hated each other for like 3 years, but they were just 4 hours

1:25 like 3 years, but they were just 4 hours a day. a day. Like so they had a documentary about this. So it would like they’re like, “Yeah, what is all the Yankees doing?” And literally when it went went into the commercial, both of them wouldn’t talk. They were just like there’s no computer, so they’re just like staring. Yeah, so Yeah, so I don’t know, man. So they had a show together and they absolutely hate each other off the air. 4 hours a day live radio in New York, yeah. yeah. Well maybe that was their problem. That was There was too many hours of of radio together. Maybe they was worn it out a bit. We’ve [snorts] been doing this for 5 years. We’ve been doing at least two

1:56 5 years. We’ve been doing at least two minimum 2 hours a week I’m talking to you. That’s true. It’s still fun. It’s still fun. good. Yeah, that being said If you want to talk to me, go to my bookings page, put it on the calendar. [laughter] That’s funny. I love it. Awesome. so today’s episode is going to be all about improving your AI skills for fab for fabric, right? So we have been starting to talk about a lot of the AI pieces. This is very pervasive. A lot of things are happening. Many, many announcements are coming out constantly actually now.

2:27 are coming out constantly actually now. I feel like the speed or the rate of agentic or AI announcements are ramping up here a little bit. So I must it must be that Microsoft is now using more AI to help build their code and their solutions. So anyways, that is our main topic for today. Tommy, bring us some news. What do we have going on? All right. Well, one of the news is there’s a new podcast on Power BI and AI tips for agentic things. I think we let’s keep pushing that, man. So you and Matthias have started a podcast, which

2:59 Matthias have started a podcast, which is awesome. You’re already what, two, three episodes in? I think we’re going to be doing five We’re on episode five. We’ve only been at it for two weeks. weeks. Yeah, we’ve been We’ve only been at it for 2 weeks so far, but we already got five episodes in. So we are starting if you are interested in more learning about agents. If you’re interested to understand the basics of agents, what they are, how they work, what is a context window, what is a harness, we’re going to start unpacking a lot of these like basic concepts. And yes, it’s kind like basic concepts. And yes, it’s a podcast, but it’s actually a bit of a podcast, but it’s actually a bit more of like a show. Honestly, it’s going to be more like a show than it is

3:29 going to be more like a show than it is just a podcast because on Tuesdays after this, you’ll see this in about 2 hours, a. m. Matthias and I will get back on online. Speaking of having too many talking conversations. Matthias and I will get back online and we’re going to just talk more about agentic experiences. Sorry, I got that wrong. Today’s Thursday. We already did it on Tuesday. You got to break today. Yeah, I got to break today. Friday Matthias and I will be back online talking about things. Our Friday sessions though are going to be less around like a podcast and more like

4:00 around like a podcast and more like screen share and demo. And so Matthias is taking us right through right now through the experience of MCP servers. And so one of the MCP servers you can go use is an MPM package. So the Power BI modeling MPC server is an MPM package you can actually install when you’re doing GitHub actions in GitHub. So you can actually go to GitHub, install the server. Matthias showed you that earlier. And then you can start talking to

4:30 talking to GitHub Copilot, the agent there, and say, “Look at my model. Review it.” And it uses its own tools. It talks to the agent. The agent then talks directly to your semantic model and you can securely connect directly to your Power BI stuff. So what we’re going to go through this Friday is, “Okay, once you’ve identified you can talk to the model with the MCP server, what if you want to make issues or changes or have a backlog of items you want to correct?” We can now talk directly to our agent and make a whole backlog of items. So we’re going to focus a lot on GitHub and

5:01 we’re going to focus a lot on GitHub and GitHub Copilot. That’s really the preferred method on on what we like to use for agents. We’re just going to explain to you like what we’re doing. What is our workflows? How are we working with these things currently? So anyways, No, I’m really excited for you. I do have two conflicting thoughts though about it because the first is I could not be happier for, I think it’s an awesome thing and needed. The other side of me and I’m going to do this as I drink coffee, call me when you get to 500. [laughter] Well, we’ll see we’ll see how long we last. podcast. That’s not an easy thing

5:31 last. podcast. That’s not an easy thing to do. to do. I told my wife that you you’re doing the other thing. She’s like, “When does he have time?” I’m like, “Dude, that guy I told you he’s [laughter] me on steroids. So yeah.” Well, I I find that my I’m really effective on camera and I’m really effective on communicating and sharing and then acknowledging things. So I’m trying to offload things on my plate to my team. I actually have a whole company behind me that helps run the business and the app development that we do. So I try and hire people that have been really good at other areas of my

6:01 areas of my, what I have to do day-to-day as a business owner so that I can focus on other things and and strategically I’m aligning myself more towards this this part of it, right? The education, the sharing, the knowledge. That’s an area that I really am passionate about and want to continue investing in. So that’s that’s what the reason why we’re going down this route. Is that what you tell junior developers? Who knows, maybe one day you can be a podcast host. I don’t know. I I don’t have a lot of interest from the developers on the team to like get on the show. They’re like I [laughter] I I throw all these ideas like they’re like “Cool.” Like yeah,

6:32 like they’re like “Cool.” Like yeah, that’s me. Tell us how it goes. I don’t know. know. When I was at I used to work at a Fortune 100 company for a number of years, about 10 years in my the beginning part of my career and I was a mechanical engineer while being there and I always got the comment of “You’re an energetic engineer. You’re an energetic engineer.” So like apparently my excitement and passion for these things comes out. So I’m going to try to leverage that to its fullest. fullest. what? I got the same thing. I would get “You’re really passionate.” Which I always took as a I thought that was a backhanded compliment where they’re going “You’re not really good at what

7:03 going “You’re not really good at what you do, but man, you make up for it with that energy.” But then you realize that people like the passion. I think so. I think there’s something but also it’s it’s also in in small doses, too, right? So like So like they can handle us in like small doses and then after a while you’re like, “Ah, I probably need to move on. Like we’re good. Thanks.” All right. So we do have two news articles. I let’s get into it. So the first one Mike, it’s our friend Power BI. We talk AI, but there’s the April 2026 feature

7:33 AI, but there’s the April 2026 feature update and Mike, I think there’s a good enough one here. We’ll do a bit of a round robin draft. What stuck out to you? What really stood out to you? What would be your first pick? I’ll go ahead and go first here and I’m going to give some love to the visual side of things. We can talk AI all day long, but Yes, we can. There’s three major updates here for the April desktop release. And the one I’m going to talk about here is this idea called the fixed size layout for cards, button slicers, and list slicers. Mhm.

8:03 button slicers, and list slicers. Mhm. Which simply now allows you to set a fixed size in the layout section of the format pane. And just for especially when you’re trying to lay out things to make sure they they are in the right size, it gives you more precise control over all the dimensions. As you resize the visual, the items maintain a specified height and width rather than scaling proportionally. Mike, I just did a training yesterday or the last 3 days downtown and it’s funny because day two is always DAX and it’s all day DAX semantic modeling.

8:33 and it’s all day DAX semantic modeling. It’s a lot for people. Mhm. But then when you shift to the report design, they go, “Oh, today’s going to be easier.” I’m like, “Yes, but there’s a lot of concepts that you have to know if they’re first time doing it. So the visual side can be almost more frustrating for people because there’s an accordion and accordion and accordion. Then and then there’s also why is the layout doing that?” Yes. And I really like this because I already see three use cases of reports existing reports I have for clients that I want to build this in.

9:07 There is Yeah. Yeah. There’s There’s There’s not a lot of updates I guess for this particular, but I think that one’s This is a really solid update. this This is one you you have to get the visualization The more we can push into the visuals side of things, the better it’ll it’ll This is going to always just win better cuz that’s what we’re using when we do things here and build out stuff in inside desktop. I I just feel like to your point there, Tommy, there’s a part of me that whenever we’re talking a lot about the experience of building visuals or things in the report, there’s consistently a

9:38 in the report, there’s consistently a heavy I’m going to call it a click tax whenever you’re stylizing things. And this is just an area that it just has always been heavy in the program in desktop. desktop. In order for you to make something easy to use, there’s usually a lot of like menus and drop-downs and items. And so you’re I understand like one visual has hundreds of properties to adjust. It’s just a lot. And to know where everything is, it’s it’s tricky. Right. Especially if you’re first time looking at it. So that’s my first pick. Like I’m

10:09 at it. So that’s my first pick. Like I’m intrigued to hear where you’re going to go with this because I have one I think you’re going to do, but I I’m intrigued to see what you’re going to do. We’ll see we’ll see where you if I land on this one, right? There’s not a ton of features out, but there is I’m going to say there’s the modern visual defaults Mhm. and the new customizing theme improvements. This is one that’s It’s actually out in preview right now. Yes. Yes. One of the things that you could always adjust your screen size of the screen

10:31 adjust your screen size of the screen size of the canvas to be whatever you wanted, right? But now when you select 16 by 9, there’s actually additional canvas sizes and we’ve actually upped the size of the screen. So by default now, Yeah. I think I think the new default visual setting setting is still going to be set to 16 by 9 ratios, ratios, which I’m seeing some people online argue 16 by 9 isn’t really the right ratio you want because whenever you open the filter pane, you already reduce the size of the screen to some degree.

11:03 size of the screen to some degree. whenever you publish a report into a Power BI app, there’s already some left navigation that’s eaten up by the screen. So 16 by 9 makes sense. You kind screen. So 16 by 9 makes sense. You think, oh yeah, that’s the size of my of think, oh yeah, that’s the size of my normal laptop screen, but in reality 16 by 9 actually isn’t the right ratio cuz there’s always a little bit of like space Right. taken up on the side. So people are now picking like another ratio of like two to one, right? Two to one. Which is actually a different ratio. But regardless, the feature is I I just a tangent

11:33 feature is I I just a tangent here. The feature he’s talking about the new default page page size will be full HD page sizes, 1920 by 1080. And what this does is it actually makes the font size allows the font size to go smaller on a report page. So if you think about it, when you’re building on,, an HD screen, 1280 by 720, that was the default way of building things. You had this You put a text box on there and you could only make the text so small, right? This the text size would go down I think maybe

12:04 text size would go down I think maybe eight is the smallest it would let you go. go. So you couldn’t shrink it beneath that. Now when you increase the canvas size, you actually get more pixels on the screen even though you’re only able to go down to the text size of eight, your visuals will feel a little bit smaller. You’re going to have a little bit more space. You’ll be able to put a couple more things on the page. the tables that you put on there will be a bit more columns. It’ll actually just fit better, I think a little bit. Anyways, I think this is a pretty

12:35 In in my experience, when you have people building reports and what you’re you are building theme files, we do a lot of this. So we see a lot of what people make. I listen, I would give this a B. I Is it Is it a blue chip? Maybe not, but like to your but to your point, like It’s going to be pretty It’s going to It’s going to really change how reports look, I think. Yeah. And honestly, all every template that I’ve ever created that I still use, they all have three different page sizes on them that I that I’ll duplicate that page of base page because generally I know I I haven’t used that 16 by 9

13:05 know I I haven’t used that 16 by 9 default in who knows how long because and I think it’s small. You’re the rare one, Tommy, honestly. I think the 16 I think You’re You’re [clears throat] You’re in the small group of people I would say I would argue probably 90 to 95% of all people just use the default screen size and build there. That’s it. That we just build in that 16 by 9, you That we just build in that 16 by 9,, know, to be there. 1280 by 720. And I guess earlier on in my development for visual stuff on reports was like, there’s a hack here. You can just do custom and you can make the page size whatever you want it to be. And I find there’s

13:36 want it to be. And I find there’s actually different storing storytelling that you would use with different page sizes, I think. 100% 100% and also too, I would do the 16 by 9 ratio just at larger size. So we’re actually really doing what they’re now having built in because you are very limited with the the normal size, the one that’s been out for 10 years. So this is a really great feature. So I’m I’m very happy to see that. all right, so I have one that I got to give some love here. Yep. So the

14:07 got to give some love here. Yep. So the training I did was PL-300 exam and I had a great group. So if anyone’s listening there, thanks again. again. Awesome. Awesome. You You You do training too and when you have a group who gets it and who understands the use cases for it, it energizes you. Mhm. And I very much had a group like this. That being said, part of the exam training is still on AI visuals. Yeah. And as from last month, it’s going away. The majority of them are. However, what I’m very happy to see

14:37 However, what I’m very happy to see because one of my demos that I do has a narrative AI feature, I glossed over explaining it, but because it was a good group, one of the persons like, hey, can you go back to that other page? I saw text on it. What was that? I’m like, yeah, that’s narrative AI. They said all AI visuals are going away, but they didn’t say which ones. So there’s a narrative visual default type update. And I thought this was interesting, Mike, because because again, if you’ve not heard, the AI visuals that were built into

15:07 the AI visuals that were built into Power BI for the last seven years minimum, Mhm. key visuals, the decomposition tree, Sure. are going to be deprecated. They’re going to be removed from desktop in favor of Copilot. Mhm. One of them I would consider the AI bucket is the narrative visual, which will have that on a left auto explain, auto update,, auto analyze your model in the context of the page you’re in. Well, Well, they now updated that visual, which tells me that they’re going to keep both

15:37 tells me that they’re going to keep both features of this where you can either choose a narrative type now. I can either have Copilot explain for me or custom, which is what the, the normal, mode was. The default is going to be Copilot, but that custom is that what you’ve been used to. So I’m very happy to see this because one is frustrating doing training when things go away so quickly, but two, I like this feature. I even like the non-Copilot version of this. So just wanted to give a little little love to

16:07 wanted to give a little little love to that. Good. Interesting. I I I haven’t tested this feature out yet. So that’s one I’m going to have to go explicitly go out and test myself. All right, I’ll give it my last call here, Tommy, then we can go into the main show here and talk about our main topic here. I’m going to do an anti-pattern here on this one. All right, so this is one This is a feature that is out in preview that I don’t like. [laughter] I’m going to be very clear about this one. And And you’ll see why, Tommy, here in just a second. So this is one you’re immediately trading This is for another draft. I’m buying it to trade it off. Get rid of

16:37 buying it to trade it off. Get rid of this thing as quickly as I can. I cannot wait to hear this. Sure. So let’s let’s do It’s another one here around data modeling, okay? So this is data modeling. This is the DirectQuery lake calculated columns and tables in preview. preview. Now I realize DirectQuery has a feature gap between what you can do currently in Power BI semantic models with desktop, right? So if you use import mode, you can create a calculated column on the imported table. You can also create a,, virtual table or calculated

17:08 , virtual table or calculated table using DAX. Okay, fine. When you were using DirectQuery models, there was a limitation on there. And so there is They’re having two notes here. One is now you can have a calculated column, which we’re calling it unmaterialized. All right, so this is a calculated column that is not in the lakehouse, but is in the semantic model, okay? And it’s it’s on DirectQuery on one lake tables where you can process it. It will be available in the service next in the next few weeks. So it’s coming out.

17:38 next few weeks. So it’s coming out. Then there’s calculated tables referring to DirectQuery on one lake column. So these are two features that are coming out. First and foremost, I understand it needs to exist because there’s it’s feature parity with everything else. Do I ever want to use this feature? No, I do not. I don’t think you should ever have any calculated columns. It makes the model bloated. We It doesn’t compact or compress that column into you or compress that column into, columnar storage of information.

18:09 know, columnar storage of information. It’s just not a good idea. And every time we see models showing up with like calculated columns, I’m like, great. We’re going to have some work here to do. do. It It’ll be fine. Like if you had a small model, Tommy, like if we’re in a place where the model’s really small,, couple mega 100 megabytes or less, less, and you just need to add a quick column in and maybe there are use cases where I don’t control the data source. I don’t control the upstream tables, right? It For whatever reason, right? That happens. happens. Most often though, I would rather do the

18:39 Most often though, I would rather do the transformations in Power Query or go back to my lakehouse and notebooks and actually make the column there. I understand. For For quick testing, like if if I’m in developer mode, I may make a calculated column to see what the value is or test a couple things, but I will not release any models to anyone else other than myself with a calculated column in it. I just don’t think it’s a good practice. So anyways, that’s my anti-pattern there that I’m going to,, hang on. I I do want to call it out, but this is a feature that’s coming out that I’ll never use. But I do want to call it out. It’s coming. It’s interesting, but

19:10 out. It’s coming. It’s interesting, but I’m not going to use it. The There are two points to the story or two sides of the story that you’re talking about. Yeah, you do want that parity with desktop. That being said, part of the question is, why is it still in desktop? Why is calculated columns still a thing?, I listen, you can’t get It isn’t. It’s there. People are going to ask for it. People are going to want to People are going to want to treat it just like Excel. Like I get it, but it’s not. not. There’s just better ways of doing it. This will be slight slightly crude, but you you if you’ve ever lived in an apartment and it was one of those old apartments and every time you turn the

19:40 apartments and every time you turn the lights on, you’re like, I know there’s going to be roaches in here. That’s calculated columns. Like you go to an apartment, there’s something to expect and yeah, it’s just like, okay, I guess this is This is the lay of the land here and yeah, most clients deal with it. There’s There’s other hidden gems in this model. [laughter] When I see calculated columns, I’m like, “Oh, there’s going to be some more hidden gems in these measures.” We just don’t know it yet. [laughter] If you If you If you pull that measure back, you’re going to see a lot of cobwebs in there that are not not happy.

20:11 cobwebs in there that are not not happy. Exactly. That’s awesome. Well, I I like I’m glad you did an anti-pattern, too, because again I I Yeah, I think this is good. This is good. Other good solid updates, too. There’s some more reporting things around some Azure Maps. , you can now easily identify preview visuals in the visualizations pane that the preview visuals have now moved to their own section. They’re somewhere else in the visualization pane. So, there’s a lot of like little tiny things that have been kind little tiny things that have been refining themselves here in of refining themselves here in desktop. So, anyways, so good up good solid updates. looking forward to see what goes on

20:42 looking forward to see what goes on next. All right, Tommy, over to the main topic. So, we got a mail back today and the title what we’re going to be talking about is improving your AI skills for Fabric. Mike, I thought as a a great way to really showcase what you can do with AI Fabric at the same time. I have an experiment running right now that you don’t know about. You may have

21:01 that you don’t know about. You may have noticed I’ve been looking over my shoulder a few times. Well, right now Yes. I thought that was your wife like making sure that you were doing real work or something that She’s No, yeah, she’s just in the corner. I’m I’m just on a podcast on. Like it’s okay. Just just just just shh. Yeah. She’s just over there like she’s got her got my eye on you. Like like what is that in Monsters Inc.? I got my eye on you, Wazowski. I see you, Wazowski. Yeah. [laughter] Wazowski We don’t need a recording for that to happen, but what I’m actually doing right now because I want to really

21:31 doing right now because I want to really actually showcase the users how impactful and how much more time AI and MCPs can really open up your workflow. What I’m trying to say is, Mike, right now on my Surface my Surface, I’m actually running Claude with instructions. And I need to update this model. I gave it three scenarios. Update existing models, wireframes that I actually received that I wanted to build the measures, not build a report, but build the measures off of these wireframes, and I needed to update new

22:03 wireframes, and I needed to update new measures with more date context. Yes. So, I just want to show that throughout this podcast, I will be able to do a podcast and have what will take what usually would take 3 6 hours probably to do from scratch Yeah. Yep. I’m going to be able to do in the podcast. So, obviously the goal is going to be talking to you, but this whole time this has been going on. I gave it instructions. I’m going to update it once or twice throughout the podcast, but I think the the real point I want to put across here is the fact that

22:33 is the fact that utilizing AI is not just so you don’t have to do the work. It frees up your own time as long as you have the expertise and the background knowledge. So, what do you think of that? Let’s throw that on you. This is good. we’re Tommy, we’re we’re definitely moving into the world of directing an AI to do things as opposed to actually having to write the code ourselves. Right. There’s There’s definitely an analogy sitting here somewhere. And I was my engineer My engineer was

23:04 And I was my engineer My engineer was like like [snorts] [snorts] Literally I was talking about this with So, in my company, one of the things I like to encourage is like this is all new, right? I was I was actually talking with a really good friend of mine recently and we’re talking about how AI is rapidly transforming like how we build things, what what goes on. And I said, I this is our the internet is created moment, I think, for us., I’m People People Some I think in in the initial stages I this might have been a bit overstated. I’m like, I don’t know if it is it really that Is it really that revolutionary of what’s happening?

23:36 revolutionary of what’s happening? And I think now, Tommy, I’m more convinced than ever that we’re in a place where And I’m hearing people at the very smart levels of this saying voice and talking to your computer with just natural language is going to be the new way you’re going to interact with computers and machines. We’re going to start typing a little bit less. Let me tell you some real habits I’m actually changing, right, Tommy? I don’t I don’t talk back and forth to

24:06 I don’t I don’t talk back and forth to my agents that I have working on things yet, but I think things are coming. Microsoft is starting to announce really interesting data models around voice to text and speech to text and things like that. So, and it’s really natural. It It actually sounds really good. So, there’s a lot of really interesting things that Microsoft’s developing in foundry and spaces where you can use these things. So, if you if you roll up this whole idea there’s been a couple concepts. We’ve talked about this on the podcast in the past. The new programming language is English. English. It is is the language you speak,

24:37 It is is the language you speak, whatever the language you’re talking. This is it. We’ve hit the apex of computer programming at this point. Right, Tommy? You’re You are instructing or giving general instructions to your Claude. Claude. It is able to listen to the words you’re saying, understand the intent of them, and from those written instructions it’s now going to do things. And so yes, you’re still typing to them now, but more and more of my workloads are starting to move. I used to be here in my basement doing things here. I’m

25:07 my basement doing things here. I’m moving out to like it’s a nice day outside. outside. I’m going to go out and just talk some ideas to my agent and and read and talk on my phone out there., I’m almost I’m starting to use a lot more of the voice inputting options when I’m talking or texting or so I’m I’m using AI to assist my thinking, how I talk to it. I’m getting programs. I’m doing a better job of slowing down my thoughts and I’m learning a new skill of how do I

25:37 and I’m learning a new skill of how do I just speak to the computer and get it to say what I want. Mike, I cannot I could not agree with you more here. So, I actually there’s a keyboard shortcut to do Okay. do Okay. [laughter] Big moment. It’s today. April 23rd. 23rd, 2026. 2026, Tommy agrees with Mike emphatically. 100%. Not 100%. 100%. 100%. thing it almost except further. So, there’s a keyboard shortcut on your computer that will do text for speech. If you have a Windows H. Yeah, Windows

26:08 If you have a Windows H. Yeah, Windows H. I have a program to calculate the calculator button from PowerToys. So, that’s and I’m this is what I’ve been doing this morning what I’ve been doing. I was sending you pictures yesterday or the this week of just showing my agent conversation with Notion. All that was me walking in the city going seeing an update and just saying, “Actually, we need this. Make sure you update this.” Everything I I showed you, Mike, was text to speech. Yep. I was not typing any of that and I was telling my wife I’m like, “I have an assistant. This is an assistant.” Yeah.

26:40 assistant. This is an assistant.” Yeah. Yes. Just to show, too. Now, I I really want to preface this because we’ll get into the topic that you still need It is cool because you do still need that experience in the skill in the thing you’re talking about, right? Because it’s really it doesn’t matter if you can talk to your AI. So, I’m actually do this. Yeah. He’s got to say the right words. Like So, there’s this also this idea of what is a What do words mean? This is going to get a little bit philosophical here for to some degree, right, Tommy? Like when

27:10 some degree, right, Tommy? Like when we’re working on things together we’re we’re talking you and I are talking back and forth. Like there’s words that we understand collect connectively. Like we understand like we say Power BI, we say Fabric, we say There’s words that we understand the context of those things. Those words could have multiple meanings. Also, when we start getting into these very technical spaces, like we have to be able to have consistent language what is with what is generally known around these words because the agent is being trained

27:40 words because the agent is being trained on whatever it’s being trained on. Actually, we don’t really know. We don’t really know what the agent’s being trained on at this point, right? It’s just being trained on a lot of information on the internet, right? And then when we talk to it, we have to know like if we say like I’m going to build a website, I want it to be a front end, I want to have a language and words, the agent is able to have context to them because there’s a general knowledge around what those things mean, right? So, you to your point, we have to know what the keywords

28:12 point, we have to know what the keywords are, how the concepts fit together. I’m feeling like I’m doing a lot more negotiation on, system, process, organization of ideas. This is becoming a skill that I’m using more and more and more with agents. And last night I was just talking with my developer about things. He goes, I said, “What’s the best thing you’ve been finding out and you’ve been learning right now? Like what’s what’s the one thing you’re taking away?” He goes, “Plans. Lots and lots of plans. I’m spending way more time planning and getting much better results out of it.”

28:44 getting much better results out of it.” And so we’re going through before we give agents tasks, we’re breaking their work up into like issues and features. And, here’s here’s the six-step plan. Okay, in phase one here’s a lot of details. And that way the agent can like step into it, do a lot of things initially, and then go into the next thing. And and so much so that we’ll do a whole plan plan and then we’ll say “Okay, take another pass at the plan. Rethink Based on this plan, rethink it.

29:15 Rethink Based on this plan, rethink it. Be critical. What did we miss? What should we add? What can we take out?” And for me, one of the things that I’ve been doing a lot is is just the word simplify. simplify. So, agents are very eager to build lots of things. At all times, yes. Yes. All the time. It seems like I’ll ask it to like, “Hey, I need to build a website.” They’re like, “Oh, great. I built you a website and a contact page and all this other stuff.” I’m like, “Oh!” “Oh!” I’m like, “Whoa, whoa, agent. Slow down.” Like, “Okay, I just need some So, one, it was probably I was not being descriptive enough at the beginning, right? So, a lot of times I’ll give it some direction and it overbuilds or it’s

29:47 some direction and it overbuilds or it’s too eager to build something. And I’m finding a lot now before I was like maybe was blindly trusting a little bit too much before, but I’m becoming a bit more skeptical about the agent and I’m doing a bit more pushing back and saying saying okay okay thinking through the plan, simplify it now. Do do this come up with three ways we can simplify this more. And then I like having the simplify step in here so it actually does a little bit less and it’s a bit more successful in

30:17 less and it’s a bit more successful in delivering a result there. What do you think Tommy? Is that what you’re seeing too? So the reason I’m laughing here is because I’m going to do real time what I was going to tell the agent here and the part about the planning Yeah. breaking things out. I planning Yeah. breaking things out. that is such an important part here mean that is such an important part here because one thing I I say especially when it is a bigger project is that I need you to think about this part. I need you to focus on this. Think about it before you do it. Yes, correct. say those words. So what I’m actually going to do everything you just said I think is spot on and I think people get

30:47 think is spot on and I think people get the misconception too that we’re doing this to be lazy. It’s like oh you just don’t have to do the work. And then I cannot disagree with that more because It’s shifting now. Again, I’m allowed to be strategic. I’m allowed to get other things done where again I’m not focusing on the format. So I what I’m going to do just for the sake of today and this will be the last thing I need to do. I want to do it in real time. How does Tommy talk to AI? How does Tommy actually work with AI? Yeah. This is exactly it. Okay, I need you to do two things here.

31:19 Okay, I need you to do two things here. First I need you to look at the context for the DAX measures with the time intelligence. They They require dates on the context from the outside. I need them to work as existing with a card or without a date column in

31:33 with a card or without a date column in the visual. That includes anything 10 week rolling or anything that has to do with time intelligence. Period. The other part I need you to take a look at and I need you to focus on this is are any other measures for pseudo measures that we talked about that need to be created. Please take a look at this before you say okay or build anything. Done. Now I’m back to podcast. And so obviously you can see there there’s me understanding that

32:03 understanding that Yep., and I’m just saying I don’t know what it’s building. It’s like okay, you’re testing things out. I know what date context is or, the filter context is so you can talk to it. But Mike But Mike I I was telling my wife I’m like if I ever got another job and they said you can’t use are your own AI, I don’t know if I would take it. Like that’s where I’m at right now. Okay, Tommy. That’s it’s funny that you mentioned that because I had the exact same So I’ve been getting a lot of calls recently about people help asking hey we’re building fabric. We need some more we’re

32:34 building fabric. We need some more we’re we’re we’re moving into this new world, right? So getting a lot of good calls around recently and I had one of the clients come in and there was some really strong language in my master services agreement about agents and and agents and using things and I and I again again Mhm. as a consultant I’m 100% on board with like let’s do it safe. Let’s make sure that we’re not sending any of your, proprietary data code anything like that to an agent that isn’t controlled in a controlled space, right? What does that look like? And so I think there’s also this idea of the market is starting to shift here is like okay, if

33:05 starting to shift here is like okay, if I’m hiring a consultant or I’m hiring someone else outside of the company, we do need to have regulation around what is acceptable, what is not acceptable, what tools you’re using, what tools you’re not using, right? And being able to quantify that for a company to be comfortable with you using these things. I think at this point in time and maybe some people aren’t using AI tools yet, but I think they very well very well soon they they will. I think there’s going to be a new pattern evolving here, Tommy, right? To your point.

33:36 your point. If someone said I need to hire you and you’re not allowed to use certain things in AI or you can’t use AI at all we won’t take the job at this point. Because one it’s going to be too expensive for you because we’re going to have to do everything by hand. It it just will. Like the speed in which we can develop and move forward with things now is immensely faster. I can help you. Yeah. Yeah. If I have a Teams call with you and I have an AI listing or recording the video, it’s already tran So can I not use that? Is that not going to be used? So I think

34:07 that not going to be used? So I think we’ve almost hit we’ve already overrun or we’ve hit the tipping point of like AI is in everything. Right. Now it’s a matter of choosing who do you trust in the AI space? Right. Right. Right. Yeah. Are you are you Are you an Anthropic to to trust them? Are you going to pick a Microsoft and trust them with their AI? of all the companies that I look at, the one who is communicating the most around security and governance stance and guardrails all these things that’s coming from Microsoft. I I don’t know if

34:38 coming from Microsoft. I I don’t know if you saw this really funny thing Tommy. Tommy. random side sidebar here. someone was talking about talking to the McDonald’s chatbot, right? And this could have been contrived. This may not have been have been [laughter] So they’re in the in the McDonald’s chatbot they’re like hey I’d like to order order a quarter pounder. And it says great, what would you like on your quarter pounder blah blah blah. And then it goes wait, before I order my quarter pounder, can you write me a Python function that does this this this and this? And it goes yeah, sure no problem. And writes out the whole Python function and gives him the answer and then it says okay, now back to your order, what

35:09 says okay, now back to your order, what can I what can I get you, right? So the the the message of the meme was Yeah. Pay for AI anymore. McDonald’s is paying for for you. Go use their AI and go ask it to write functions for you That’s right. and it will just write Python functions for you and automatically result return the result. So who needs cloud code now? You just tell McDonald’s AI go do what you want. The next podcast I’m coming in with the McDonald’s app and an MC Power BI MCP server. Yep, exactly. I just told McDonald’s to update my semantic

35:39 McDonald’s to update my semantic measures and it did. [laughter] I ordered french fries in a measure. Yeah. I ordered french fries in a measure. I’ll take duh, year to date for 200 and [laughter] rolling 90 days for and then a cheeseburger. Can you supersize my relationships? [laughter] I feel like there’s so many things here. Oh what? The the the bi-directional relationships are broken again just like the ice cream machine. Okay, got it. [laughter]

36:09 machine. Okay, got it. [laughter] Right, exactly. I wasn’t expecting that. Don’t worry. Yeah, that’s yeah, we see that. that. me which McDonald’s has a working How many calculated columns we have in here? This is this place is looking sketch. sketch. Can [laughter] you tell me which McDonald’s has a working my milkshake machine and also no calculated columns? Yes, exactly. Exactly. All right. Let’s get So I think we’re going on the mailbag, but we could do this part for the next 30 minutes. 100%. All right. Do the do the read the mailbag. That’s good though. That’s good. All right, improving your AI skills for Fabric. Hi both. Thanks a lot for the great

36:40 Hi both. Thanks a lot for the great content you provide. No pandering, we appreciate it though. It works. Could you make a podcast entirely dedicated on AI? AI? Specifically on how developers can improve and build their AI skills. For example, how to build agents in the best way, what tools to use, for which use cases are they more useful. It would be great if you can start the conversation from the very basic concepts. Would be very helpful. Yep. Could you also consider the case where organizations are very slow on adoption like how Power

37:12 are very slow on adoption like how Power BI Studio and Fabric are still not available. Yep. Thanks a lot no name. We would love to get the name here. Mike I honestly see this as three episodes almost. There’s a lot There’s a lot to unpack here. Yeah. Part of this is if you want to know about some of these basic things too,, Matthias and I are going through some of these basic things as well. We are going to unpack and explain them and and start talking about these things as well. So I’m going to rattle out something. Ha- Have you seen the website Tommy awesome co-pilot? Yes. Yes, you showed it last

37:42 co-pilot? Yes. Yes, you showed it last on Tuesday. Okay. This is really interesting to me. So awesome co- So when you’re getting started with agents or agentic things the the So in the subject Tommy you and I were just talking about, right? The way we talk to it, the way we interact with it, right? you have to somewhat take your domain knowledge that you already have and leverage that to help get good results from the AI, right? That that’s That is a feature of you you have to get understand this one.

38:13 you you have to get understand this one. There’s a lot of new skills I’m learning, right? I’m learning as you interact with the AI what it can do. You give it a prompt, it didn’t do what you want. You have to come back and reprompt it. I’m spending more time on planning and instructions and and and building systems befo- So here’s what I want to build and I describe at a high level what I want to build and say think about it. Reason about this. come up with a plan before I actually tell it to go do stuff cuz a lot of times in the past I

38:43 stuff cuz a lot of times in the past I would just go to the agent say I need a thing, do this thing. And you I would I would immediately first prompt was like go create. the other thing so that’s maybe one of my techniques that I would pull away and a tool. Let me go I’ll pass it back to you, Tommy. Yeah, so I think we need to even take a step back here because all the agentic things all the MCPs, the things that we talk about, none of that to me matters if you don’t start with skill number one here and that’s the idea of

39:13 number one here and that’s the idea of prompting. Context and prompting are by far the foundation for all of this. Whether you build your own agent, whether you’re using an agent, whether you’re using MCP server, Mike, you’re not going to get [clears throat] very far if you don’t have honestly a skill that I would put on a resume at this point is prompting and context awareness or context engineering. The previous term was prompt engineering. It’s now changed to context engineering. Whatever. There’s tech pros who will talk to you all bit about that. I don’t care about them. But it is true that you

39:46 care about them. But it is true that you do need a foundation and a format on how you actually chat, type, or command an AI AI chatbot or agent to do a specific task. Yes. and I think to your point, there are a lot of There’s a ton of methods. There’s a ton of approaches here. But even a basic one like build me a Python script, right? Yep. Yep. There’s still elements in there rather than build a Python script for a notebook, right? Where there is a specific outcome that you need, context

40:17 specific outcome that you need, context it needs for in a sense I always add why I’m building it. Like I’m building this for Fabric. It needs to work with PySpark. The outcome will be this. That That’s its own paragraph. Yes. And it’s part also of that formatting, too. It’s the next paragraph. I need to accomplish X, Y, and Z. Can you help start build it for me? I think you made a really important point here. Part of that skill is not just a giant prompt that used to be the old way of doing it. Mhm. But because And And if you start with ChatGPT in 2023, you would write this

40:49 ChatGPT in 2023, you would write this giant prompt to do everything. But because agents now have the ability to reason,, you see it all on all the tools think, reason. Well, there’s truth to that. So, breaking out your project into different prompts to start with Let’s start with the planning. Let’s start with the idea here. One of the things that I always do This is more of a actual hard tip, but I always say at the end of a beginning of a bigger project, ask me any questions before we begin. Mhm. That’s the last sentence I have. Yes.

41:20 sentence I have. Yes. it will actually go through. But to me, none of your ability to work with AI matters until you are proficient in the ability to provide context into a prompt. What do you think about that? Is that where you would start? Is there a different approach you’d take? I think these are good places to start, Tommy. Like going from the basics if we had to pull this really high-level, like really back to basics, right? right? I’m going to I’ll go like I I agree with

41:51 I’m going to I’ll go like I I agree with all the things you’re talking about, Tommy. Prompt engineering, all these things. But But what is a prompt? What is a large language model? Mhm. Good point. What is it Like so if I if I even need to step like let’s let’s go even even further basic. My mental model of this is I take this

42:06 My mental model of this is I take this concept of the model and the the piece of software that I use around the model, right? So, if we just talk to the model directly, the model is text in, text out. That’s all it is. Right. Yeah. It It’s just a It’s just a phrase of text that goes in, and this is how we And the You may hear this word tokens all the time. Tokens are so I even wrote a song about it. Give me more tokens because it’s so It’s a great [clears throat] song. And so, so, the idea here is every time you’re chatting with this AI, you’re you’re

42:37 chatting with this AI, you’re you’re exchanging tokens. You’re You’re sending the text, it’s returning text. There’s this thing called the framework, and it has now been recently branded the harness, right? So, you have to understand two major concepts when you’re talking about agentic systems. The harness, which is the software like Claude Code or VS Code or fabric. com, right? So, every the the thing that you type your text into, right? Or even a CLI, right? The command line interface, right? All of these things are harnesses around the large

43:09 things are harnesses around the large language model. The large language model is the things that you hear everyone talking about. It’s Claude Code or Opus 4. 6 or it’s ChatGPT 5. 2 or 5. 4. I think 5. 5 is now getting released. Like So, the model the model Claude now. So Yeah, 4. 7 in Claude, which is actually they’ve been They did a lot of sneaky things in 4. 7. I’m We’ll get to that one maybe a little bit later. [laughter] Another day. Another day. Yeah. Another day. Another day. But the model is the the the basically the exchange of text back and forth, right? And the models are getting smarter. They’re bigger.

43:40 are getting smarter. They’re bigger. They run on big machines. And so, that’s the model is something you can swap out. The harness is what you use to talk to the to the agent. And different harnesses make it easier or harder to do different things, right? VS Code’s harness makes it really easy to manipulate multiple files, makes it really easy to track your changes. It makes it really easy to revert your changes on things. If I go to chat. openai. com, I have just a text window. And it was actually I would send

44:10 window. And it was actually I would send text in, and then I would copy the response out and put it somewhere else, right? It doesn’t It couldn’t interact with my other code spaces. So, you need to one understand harness, and you need to understand model, okay? With that concept, the combination of those two elements you can think of it now as like an employee. employee. And so, someone explained it to me that if Let’s just say you’re hiring a contractor for your project. Mhm. Mhm. You every day you would show up to a project to work

44:40 you would show up to a project to work on something with someone. This contracted company is so big, they would just send you a brand new employee every single day. So, today I’d work with Tommy, and Tommy I would say, “Tommy, I’m going to explain to you my project. Here’s what we’re doing. Here’s what’s happening. Let me get you up to speed as to what we’re working on.” Great. Tommy’s like, “Great. I now understand what we’re doing today.” And then I can go talk to Tommy, “Okay, great. I now need you to build some semantic models, all these things, blah blah blah blah.” Right. Then tomorrow I show up. They don’t send me Tommy. They send me Alan or somebody else or Susan or

45:11 Alan or somebody else or Susan or whoever. And now I have to go back through the same thing. Here’s our project. Here’s what we’re doing. Here’s how we’re doing things. There’s other So, in this context window, right? The agent only knows what text you can send it. That’s the model. The model only understands It reasons about the text you send it. So, a lot of what we do now is we have to make these things called skills. We have to use these things called MC MCP servers. And all of those are basically helping the model have memory or give it instructions to do

45:42 memory or give it instructions to do certain tasks and things. So, that’s why this I treat this these agents and and systems like employees the the or other co-workers or cuz you have to explain what you’re going to do. I wouldn’t just show up and hire someone off the bat and be like, Yeah. “Build a semantic model.” Like I I know I I Yeah, right. What do you mean? Like So, I I’m going to I’m going to slightly push back or rather I’m going to ask you to elaborate a little more here because to me I I Not that I don’t agree with you what

46:13 I I Not that I don’t agree with you what you’re saying, but to me, what you’re talking about I would I would put it still as of today in the intermediate bucket. I would not put that on my I’m learning AI. I want to get better at that. I would still say, “Okay, now you’ve taken a step at this point.” I Because here’s the thing. Like the agent agent, I think that’s fair. Yeah. And I’m not saying that’s a bad thing. I’m just saying I don’t think we’re at this point right now where one MCPs are incredibly easy without thinking to use. Like the Power BI one to install, right?

46:44 Like the Power BI one to install, right? agree I would agree with the MCP conversation part. But that stuff’s going to get I think obfuscated away at some point. But the idea is like you need to understand that you need to understand you’re talking to a system, right? You’re talking to Yeah. That That’s more of my point. I’m not really caring about like I’m not going to expect a new user to concern about building skills for using [clears throat] NCP servers. But like those are things you’ll have to your point. Yes, I would agree. Those things are definitely intermediate level notes. But the mental model here is Yes. I know what you’re saying.

47:14 Yes. I know what you’re saying. I’m talking to a computer that has no context of what I just said to it yesterday. I can always think of it as it’s a brand new contractor showing up. No other information. I have to just talk to you for the first time, right? So, a lot So, a lot of what we do is giving it information so that way when I come back, it has memory and it has tools and it has these things at its disposal so it could be effective. I think to me what you’re saying is know thy environment. Know the platform that

47:44 thy environment. Know the platform that you’re playing in and know what tools are available. For example, if you’re using Claude, well, Claude has a feature called projects. To your point about context, I don’t want to have to go back every day and talk to it as if it’s a new employee on something I’m already done, right? That’s frustrating, and you’re That is going to lead to conflicts, errors, things that are not desired. So, if you’re using a certain tool like when I use Notion, right? Notion works well in a certain format,

48:14 Notion works well in a certain format, and it likes things in a certain way. And I can carry it an agent, but it works well with database relations and whatever the case Whatever the case may be. Whatever tool that you’re using. Understand the environment that you’re playing in. Understand what it likes as it input because I think that’s the really the building block that I would say is understand prompting, the fundamentals of prompting, the fundamental fundamentals of providing context. Then understand what your environment, your harness, is. And I

48:44 environment, your harness, is. And I think that’s really to me If you have that, you’re ready to go intermediate. And I think does does But Mike, I’m going to ask you the question, does this apply to Fabric? Because I feel Mhm. as I say this that we’re talking very much about AI general, right? But I don’t know for an Fabric developer or Power BI developer, is everything we’re saying applying or is it a different perspective? Yeah. Yeah, I I I’m I’m going to err on this on two

49:15 I’m I’m going to err on this on two maybe two directions here. First direction, Fabric has its own harness. It’s everywhere you see the little Copilot button and the chat window that appears, appears, that is a harness that Microsoft is building and providing to us as we use Fabric, right? So,, you don’t have to go get a model. You don’t have to go get a harness, right? They’re just built into the product, right? So,, I I will argue though, the harness that the Fabric experience gives us is a bit basic for my tastes, right?

49:47 us is a bit basic for my tastes, right? If you’re If you’re a professional developer developer and you’re one who understands deep semantic modeling and building elaborate reports and you’ve been doing this for a while, the experience of just the harness and and this is one of the things why I like to use these terms, Tommy, because people say frameworks and things and all these different terms, right? It’s the same thing when I use filter context context in a in a report.

50:17 in a in a report. If you look at a report page, you need to be able to identify what is the filter context on every visual or where measures are applied and not applied. So, I have a bar chart and I can clearly see that there are dates across the x-axis. Okay, the dates are dimensions. Okay? The filter context on a single bar in that chart is now a number, which is now a measure, right? Being able to identify in the same way what is the filter context of

50:49 same way what is the filter context of this data point that lives on this visual, I think is an important skill because that helps frame out when you start understanding filter context, that helps the whole DAX world lights up for you. In the same way, when you start talking about agents, you need to be able to identify whatever experience you may be given. given. Websites, Websites, Fabric, Fabric, programs you download, you need to be able to distinguish between what is the model and what is the harness. And and that’s the language you need to understand because I would argue to you,

51:19 understand because I would argue to you, Tommy, the harness in Fabric is very simple right now and it’s not It doesn’t have all the capabilities we have in other harnesses. And so, I think one of the reasons why I’ve been disillusioned by using the Copilot experience in Fabric is because the harness is just weak. It doesn’t have the features or the interactions or the editing ability that I want harnesses to have. Honestly, Mike, you to me almost for the Fabric developer, we’re shot where if you want to start,

51:50 shot where if you want to start, you need the MCP even if it’s the local one. I I almost feel like I should make that argument here where to your point, right? There is not a lot of features unless you have the Copilot license, but even then it’s such a different harness. It’s such a different experience where Yes. If you are a Fabric developer and want to start directly using AI with your models with with Fabric, you you need an MCP. I I don’t know. Is that too much of a hot take, but I almost feel that But you’ve already jumped Tommy. You start there.

52:20 You start there. Yes, but you’ve already jumped like to your point earlier, right? You’re saying things become really effective for you at the MCP level. Well, you’re already a jump from beginner mode to intermediate mode. And and so, I think what Fabric’s trying to do is it’s trying to be this onboarding

52:33 do is it’s trying to be this onboarding ramp for like beginners all the way into advanced users, right? So,, when I look at the harness of Copilot in Fabric, it’s not really designed for pro developers at this point. I’m I’m leaving that harness to go to like VS Code. People are using Claude Code Desktop. Like so, the harnesses that are really effective for me are CLI, Claude Code, and GitHub or VS Code harnesses. Actually, I only exclusively basically exclusively use VS Code as the only harness that I really use. So,

53:05 only harness that I really use. So, again, being able to put your fingers on what these terms mean helps you understand like what you can do and what you can’t do. Sure. Of course, I’ve used some harnesses before or I’ve used some Copilots to like help me write a little function, but then it’s a lot of like copy-pasting things. It’s not really actionable. For me, I think at the Fabric conference though, right? So, Fab Con in March, really moved the needle for me, right? I really felt like I saw Microsoft directly shifting their stance on what is an agent and really improving the harness. And I think we’re going to get

53:36 harness. And I think we’re going to get a lot of new features that are going to make Copilot feel a lot more capable and and easier to use because that ramp-up story of the very basic user to intermediate user will be easier to onboard, I think in the future. Honestly, I Yeah. I’ll pause there. You go ahead. No, I I think you you mentioned Copilot right now in the Fabric experience and as of today, I would argue that going from Copilot in Fabric to MCPs is not a stepping stone. They’re to me, they’re a different set

54:06 They’re to me, they’re a different set of stairs if you want to keep going around that analogy. That is not the continuous improvement of your own skill because because let’s face it. As of today, again, the day that we agreed 100% on something, Copilot is has a lot to be desired and it is very much for a beginner. It is not for that in experience. And if I would only have the skill or only have experience using Copilot in Fabric and then someone told me, “Okay, you’re good at this now. You can get to the MCP.” I don’t know what I’m doing. That

54:37 MCP.” I don’t know what I’m doing. That is such a different experience and a different method, approach. Yes. So, they they really are different skills at this point. Hopefully, they’re going to have that continuous one where I can, graduate from ChatGPT to Claude Code, which to me makes more sense. But right now for I I I do feel for Fabric developers who are just getting started because you don’t have a continuous experience in Fabric using AI. Like I have to have the MCP,? I have to understand

55:09 the MCP,? I have to understand that authentication or from using a local model and the tools available for me right now are not fully integrated the way that a lot of other AI environments are. So, we’re we’re you have that backstep here, but as we as we move on with that, I don’t know if you’d agree with that with right now the current Copilot experience. I think where we see where it’s going, it’s going to get there, but as of today, like let me ask you this. I think this is a good way to ask the question.

55:41 is a good way to ask the question. Someone wants to get started with AI and they’re a Fabric developer. Where are you first guiding them? Are you guiding them to Copilot in a notebook or are you guiding them to Claude Code and skills and MCP? And this is Let’s imagine again, I’ll give you the scope. This person Yeah, what’s the full-fledged Power BI semantic model developer, full-fledged Fabric developer. They do this every day. But let’s say they’re just new to AI. So, where Where are you guiding them first? Where are you saying, “Here’s where you’re going to start working?” Great.

56:12 you’re going to start working?” Great. Yep. So, Yep. So, heavy developer heavy pro dev in the Fabric semantic modeling space. For me, I’m immediately pushing you towards learn how to use VS Code. Learn how to use to your point, Tommy, the Fabric MCP modeling server. Those are the two biggest wins you’re going to have. Start with those things. And that was actually the main session when we did our 8-hour training with TS and I in Atlanta, the whole day was spent around using the Power BI MCP server, talking to models, moving things around. It was

56:42 to models, moving things around. It was an 8-hour day of just using that tool to help you build semantic models. These were professional Power BI developers using that MCP server with an agent to talk to your model, make changes, do things, right? And it was less around like ask me a question and get data out. There was zero of that. It was only about making measures, moving measures, columns, columns, look at my model and tell me what’s better or what’s what’s worse. What can we improve here? So, that is where I would spend your time. So, again, you

57:14 would spend your time. So, again, you would spend your time. So, again,, how do you know, how do you to the questions here real quick, I’m going to answer some of them, right? What use cases are most useful, right? Allowing agents and harnesses to talk to your existing existing built things. Notebooks. I’m actually I’m using agents right now to build Fabric notebooks. I’ve given my agent access to be able to use the APIs to go download a notebook, look at the information, make changes, and then upload the notebook again. So, you can you can literally download items, work on them, and then push them back up. So, I’m using my agents to do this. this. So, that the places where that’s

57:45 So, that the places where that’s happening, anywhere you’re writing large amounts of like software and code, that’s where I’m applying agents. The question here around,, how do you build agents the best way? I’m not necessarily at this If you’re a beginner user, I’m not interested in helping you build I’m not saying your skills should be building custom agents right out of the gate, right? That’s where I would say go to Awesome GitHub Copilot or Awesome Go Copilot. Go use that. Go find some custom agents there.

58:15 that. Go find some custom agents there. There’s a lot of great things. Actually, if you go to Awesome Copilot and you search for the word Fabric, there’s actually a number of Fabric items, items, Fabric Lakehouse items, Fabric architecture co-autopilot, and you then you search for Power BI, there’s a Power BI data modeling expert. There’s a Power BI DAX expert mode. There’s a Power BI performance expert mode. Visualization expert mode. Custom visual developer expert. Power BI data modeling best practices. There’s a whole

58:45 modeling best practices. There’s a whole bunch of like Power BI related skills and custom agents that enhance this. And so, when you get to building your own agents, these are the things you’ll use. My last word of advice here is you’re going to find way more value in advanced harnesses. Get a VS Code is going to be a staple. Like so, that would be one I would highly recommend you learn how to use an agent inside VS Code because that’s that’s the method. And

59:16 because that’s that’s the method. And then when you want to go super pro level, you go to the CLI. You don’t even use VS Code anymore. You just ditch VS Code. You just go to CLI and you just talk to your agents via CLI. In a terminal. In a terminal. It’s It’s incredible and the experience is really good. good. All of my developers now exclusively use CLI. They don’t even use VS code for talking to agents anymore. They they love the the the terminal command line interface now. And that’s But that’s Think about it. That we when Tommy when we distill things down to the most basic level,

59:47 things down to the most basic level, we’re just talking to the agent. The only thing we really need is a terminal to talk to it. All the other fluffy things, buttons, and clickies, and all the We don’t really need any of that. Buttons and clickies. We don’t. [snorts], that’s that’s We’re We’re going down to the most rudimentary fundamental thing of this, which is flat files, folders, and talking with language. Yeah. That That’s where we’re at right now. now. 100% and Mike, I actually already put in our in our backlog for topics building agents for Fabric and orgs that are slow on AI adoption because I think these are

60:17 on AI adoption because I think these are other two great topics. I’ll close with this because I know we’re at time, but I’m just going to reiterate what you said here. There is a pathway to learning AI and there is a different pathway for learning AI here around Fabric too. That being said, there is still a foundation that you need to make sure regardless whatever harness, whatever whatever environment, whatever tool or project that you’re doing. And to me, it it really does start with that ability to prompt. Again, the approach There are I don’t

60:47 Again, the approach There are I don’t want to say as much skills as much as techniques. It’s probably a better way to say it of techniques of depending on what you’re trying to accomplish, but this was a great question. Loved it today, so Good discussion. There’ll be many more conversations around the intersection between agents and things and back to power. You got your model done already? Everything I was hoping to get done within the podcast. Completed. Looks good, so yeah. Now I got to go test it. So now I got to go test it. I even asked it to do a sanity check, so that’s another thing I was doing. amazing. Awesome. Absolutely love it. So

61:17 amazing. Awesome. Absolutely love it. So with that being said, thank you all so much for listening to the podcast. Hopefully this was a good discussion around like agentic things, how does this work, what things can we pull on, how do we start pulling that stuff forward into our workflows with Fabric and Power BI. So we’re we’re going to continue to explore this. This is This is We have We have factory reset the whole world’s knowledge on like how we build things now with all these agents. We’re all learning together. We’re all trying to figure out how this is going to move forward. What I will say is the future is agentic. There’s going to be this is going to shift what you do. And

61:49 this is going to shift what you do. And if you thought you were managing a lot before, 10x the amount of things you’re going to need to manage because now you’re going to have to have You’re going to be wielding agents to help you manage manage five or 10x more things. You thought we had a lot of semantic models before Tommy in our environments. We’re going to really seriously ramp that up with now agents being a part of this as well and being able to build things programmatically. So really good stuff there. And then with that being said, we really appreciate your listening. Tommy, where else can you find the podcast? find us on Apple, Spotify, or wherever you get your podcast. Make sure to subscribe and leave a rating. It really

62:20 subscribe and leave a rating. It really helps us out a ton. Do you have a question, idea, or topic that you want us to talk about in a future episode like today? Head over to powerbi. tips/podcast. And finally, join us live like today on Tuesdays and Thursdays at a. m. Central on all powerbi. tips social media channels. channels. Thank you all so much and we’ll see you next time. Yeah. Yeah. Explicit Measures pump [music] it up PR high. Tommy and Mike lighting up the sky. Dance to the data laughs in the mix. Fabric and AI get your fix. [music]

62:51 mix. Fabric and AI get your fix. [music] Explicit Measures Drop the beat now. Podcast [music] kings feel the crown. Explicit Measures [music] Drop the beat loud.

Thank You

Want to catch us live? Join every Tuesday and Thursday at 7:30 AM Central on YouTube and LinkedIn.

Got a question? Head to powerbi.tips/empodcast and submit your topic ideas.

Listen on Spotify, Apple Podcasts, or wherever you get your podcasts.