Crawl, Walk, Run with AI in PBI – Ep.553

Meagan Longoria wrote a staged maturity model for bringing agentic development into Power BI work, from chatbot questions through to skills and MCP servers. It’s the organized version of a conversation this podcast has been having piecemeal for months. Both hosts recommend it, and Mike takes issue with exactly one stage.

News & Announcements

• Crawl, walk, run with agentic development of Power BI assets — Meagan Longoria’s five-stage maturity model, and the basis for the episode. Both hosts note the framing they respect most: if you have ethical objections to AI, follow your convictions — her position is that it’s inevitable in this industry, so she wants to use it carefully while advocating for mitigation of the societal and environmental damage.

• Chicago Fabric user group — August 20 — Tommy will demo his Fabric Task Flow Studio fork: running it locally, generating dummy data through the environment, and the tips for getting real use out of it. He notes Alex Powers’ underlying project already understands variable libraries — tell it you want dev, test, and production and it wires the associations correctly.

Main Discussion

Topic: The maturity path from chatbot to skills, and where to actually start

• Mike’s disagreement is about crawling. The article starts people with chatbot questions — how do I write a rolling 12-month measure, how do I build small multiples. His view: if you already use Power BI and have talked to a chatbot, that stage is behind you. Install the Power BI Semantic Model MCP server and start by asking it what it can do for you.

• Questions without model context can mislead. Tommy’s version of the same point: DAX advice divorced from your actual model and its relationships can be worse than no advice.

• The harness matters more than the model. Mike’s estimate: 70–80% of output quality comes down to how good the harness is. A strong model in a weak harness produces weak results.

• Skills are the walk-stage requirement. Whatever editor or chatbot you’re in, Tommy thinks you hit a wall without getting a handle on skills. The trigger is noticing you’re saying the same thing repeatedly — that’s the moment to make it reusable.

• Tools are scripts with defined inputs and outputs. Mike’s clean definition. The agent shouldn’t write code to create a table; it should call a create table script with parameters.

• Use git, because the agent will break something. His practical addendum to the article: have file versioning and commit history, because he’s had to revert more than once.

• The IDE is disappearing. VS Code, Cursor, and Claude all put the agent front and center now, with files pushed to the periphery — the same move Replit made. Mike wouldn’t send a newcomer straight into a file tree with a chat panel on the side.

Why so much Claude

Both hosts address the question directly: their customers use it, so they have to know it. Beyond that, Anthropic developed skills and open-sourced the format, and the integration with local files was the tipping point. Mike’s framing is that Anthropic was the first to publicly show the AI could build itself — you still direct it, but the loop closes.

On em dashes

Mike’s affectionate ribbing of the article: if you’re going to write about AI, don’t use em dashes, because readers now treat them as a tell. He trusts Meagan and knows she reviews her work — the broader point is his own growing skepticism, and that misspelled words have become the reassuring sign a human was involved. He closes warmly: she was a genuine skeptic early on, and watching her come around and teach others how she got started is the valuable part.

Looking Forward

If you’re at the crawl stage, ask your IT department for the Power BI Semantic Model MCP server rather than another chatbot tab — then let the agent tell you what it can do.

Episode Transcript

0:29 Hello and welcome back to the Explicit Measures podcast with Tommy and Mike. Tommy, welcome back to the show. Hello again. again. Hello, hello. What’s going on in your world? Dude, not too much. Just a lot of busy things. We have our next Chicago Fabric and Power BI user group coming up on August 20th. It’s coming up pretty soon and we’re doing the Fabric task flow and the fork I did the Fabric task flow studio. How to get started running the application on your machine. some of the tips and tricks on using it and how you can actually really do

0:59 it and how you can actually really do your full development in just this UI in this great application to actually get your artifacts into Fabric. And then actually even add your data and continue the conversation all using Fabric task flow studio. And so Tommy, this is a an evolution on top of what Alex Powers built which was Yep. Yep. So to Alex built like a very like task flow assistant basically that you just kind assistant basically that you just communicate to it with the CLI. Hey,

1:29 of communicate to it with the CLI. Hey, I’m here. I’m going to, I want to build this kind , I want to build this infrastructure. And it says, “Oh, let of infrastructure. And it says, “Oh, let me reason about what you ask for.” And then it’ll come up with like a recommendation. It’ll give you all the scripts. It’ll build everything for you. it basically it’ll make an entire environment for you. Tommy, have one of the things I want to ask about this one is is do you have or have you built not only the infrastructure with it, but also the infrastructure and sample data with it as well. So we’re going to show that during the user group, actually creating dummy data throughout it and going through a

2:00 throughout it and going through a full development. Oh, yeah. Oh, yeah. It’s not just going to show the UI. We’re going to actually show if you are trying to add data once you deploy the artifacts, you can even get started with I’m actually using this API here or I have data over here. Let’s get started with the current things that you deployed and how easy it is to work with the agent. Yes. that’s one of the really important pieces of this is Yeah, one of these important pieces of this is, you these important pieces of this is,, not only can you get a again for know, not only can you get a again for demos, for testing out things, it’s really useful there in that way to make

2:32 really useful there in that way to make sure things are just able to be to be built quickly. And one of the things I was astonished by this project that Alex has built, it already understands how to use variable libraries. And so, if you tell it I’m going to make dev, test, and production environments, it wires everything together with all the appropriate associations between a variable library of hey, this is a semantic model that’s pulling data from a lake house, but we realized that the lake house will have a different good name in these future environments, and

3:02 name in these future environments, and it parametrizes all that. It makes it all really easy for you to build a pipeline that actually has multiple environments, which is this is amazing. Anyways, Tommy, love this idea, love the studio, love the wrapper that you put around it. It’s super useful. I’m very excited to see this come to fruition. All right. That being said, Tommy, let us lead us into our main topic today. We’re not going to go into anymore on news or announcements. We’re going to make this a bit of a brief episode. This is recorded. What’s our topic for today?

3:32 topic for today? So, this is a We’re actually using an inspired by a great article by Megan Longoria on datasaavy. me, her website, her blog. And this is I think a culmination, really organized way of way of things we’ve talked about, but it’s called crawl, walk, and run with a gentic development of Power BI And she’s really going through a really five-stage model and maturity model for you bringing in anything agentic into the Power BI development. Moving from

4:02 the Power BI development. Moving from your simple chats all the way to your autonomous workflows using skills, MCB servers, and customer instruction files. And I love what Megan said at the very beginning here, Mike, because I think this is really important. so, I’m just going to read here just her beginning of statement. If you’ve been watching AI roll through the data community data community and thinking, “This seems useful, but I have no idea where to start.” this post is for you. If you have ethical objections to using AI,

4:32 have ethical objections to using AI, follow your convictions. This post isn’t here to convince you otherwise. My position is that AI is inev- inevitable in this industry, so I want to understand how to use it and carefully and effectively while advocating for mitigation to the societal and environmental damage it causes. So, I think there’s a big point here just starting off, Mike, where you and I are so deep in the waters. You and I are like I more or less the Michael Phelps of just being knowing how to swim in these waters, but I think a

5:03 to swim in these waters, but I think a lot of people are still again, not because of their own decision, but because of the organizational side of it where organizations are finally now wanting to introduce AI into their organizations for the development. Well, where do you actually start? And I think the big point really the big point I want to just take from her this is Mike, no matter what we are in an industry where AI is going to be essential to what we do. I think there’s no going around that. I also think we’re one of the early

5:33 I also think we’re one of the early the early like if we think about the waves of technology or things that are going to get hit with where AI is going to be applied., we’re already in a very heavy technology-centric area anyways. We’re doing data manipulation. We’re already using computers a lot. We’re already most of Tommy when we were consulting even before AI, we just spent tons of time on computers, period. We were just always here. We’re looking at things, working on stuff. So, I think our industry in general with data and AI natural fit for them to go together. So,

6:03 natural fit for them to go together. So, I also I’m looking at this going, well yeah, this makes sense. We if we don’t and this is we used to say this on the podcast before Tommy and I think I I think we agree. If we don’t move forward and at least learn how and where to use it, it, your job’s done. Yeah. Yeah. It’s if if you don’t learn how to do it, someone else will learn how to do it and they’re going to take your job because

6:25 they’re going to take your job because they actually know how to wield and do things with AI. Already Tommy, I’ve seen AI do incredible things. It can do stuff much faster than I can. It builds things faster than I can. But I can I still harness it. It’s It’s I use it as my It’s one of my tools now in my arsenal. one of the things I’m going to I’m going to rib Megan here a little bit on that first paragraph that you read. Tommy in that first paragraph, the the entire thing is awesome, well written. Yeah. Yeah. However, there’s a very glaring tell in

6:55 However, there’s a very glaring tell in the beginning of that paragraph. There is an em dash between follow your visions dash this post is here is this post isn’t here to convince you otherwise. I’m I’m I’m going to Yeah, wait to hear I can’t wait to hear this. Go ahead. Nobody puts dashes in unless you’re an agent or an AI. So, I’m thinking that is an AI. an AI. But this is a this is a different kind But this is a this is a different dash. Dashes that you use are the of dash. Dashes that you use are the short ones. This is a really long one. I don’t know. I

7:26 I don’t know. I Which which indicates to me that this is an agent related. Now, again, I’m totally fine with agents writing and helping you write the articles and I know so know Megan very well. I know she’s not just blanketly like throwing down her ideas and letting an agent write an entire article for her. So, I know she’s reviewing it and she’s going through it. So, I trust Megan cuz I know her as a person. She’s doing an amazing job, but job, but if you’re going to write an article about AI, don’t put any EM dashes in it because even if you’re even if it was intentionally written there by you, the

7:56 intentionally written there by you, the user, user, other people acknowledge that an EM dash comes from an AI agent. And therefore, we live in now where that’s the where we at where that’s the way we’re thinking? Tommy, it’s it’s weird now. If I see a spelling mistake on a website now, I now feel like it’s written by a person. So, you don’t like my articles, yeah. So, peace., all of Tommy’s articles are written by people. Right? You There’s a spelling mistakes all over the place, right? But

8:26 place, right? But but to your but like Tommy, AIs don’t spell words wrong. They don’t. Oh, yeah. Oh, yeah. There’s like these like little glimmers of like this is still human related because right now my assumption is everything on the internet is made by an AI, period., , we are skeptics at heart, I know. So, I’m becoming more skeptical right now. But anyways, You already were we were we already were skeptics. skeptics. This doesn’t this doesn’t negate the the value that Megan’s put in here. I just like look at it and I I laugh I chuckle a little bit at thinking, “Oh, it looks

8:56 a little bit at thinking, “Oh, it looks like that agent wrote the the intro of your or you had an agent review your intro of your podcast or your your post because there’s a dash in there.” Now, that dash could have totally been there normally, but I think in general people are tainted by the idea that dashes mean agents write things. Anyways, let’s move on. on. All right. So, yeah, but I think the biggest thing here let’s let’s just jump right in here where I think for a lot of people is very intimidating and we’ve talked about this. So, I really like the approach that Megan has here when it comes to comes to first off, let’s just get again, right

9:27 first off, let’s just get again, right off the bat, no if you are in the data space, it’s going to be very hard to do your job or to I think move forward in in the future without really taking hold of the of tooling available from an AI point of And I think that’s an important distinction, right? Not necessarily that AI is part of your job because again, a lot of buzzwords here, but the tooling available is going to be an essential part of our workflow. But I’m not going to start someone off with an MCP and

9:57 to start someone off with an MCP and Megan starts off right here in terms of this again this idea of crawling, walking, and running with AI. So starting small. And obviously we would talk about the your organizational AI policies. She starts off with talking about asking questions in AI chat interface. And this is actually something I went back in 2024. I had an article about diagramming DAX for generative AI. And it was really much about taking like screenshots of my model, asking it questions about hey I want to build

10:28 questions about hey I want to build these measures and I want to also say,, how I want things to interact with each other. Just provided a visual help in it. And it worked really well where I’m using a single chatbot. I think for a lot of people this is a pretty good place to start. start. But I’m going to just throw it to you real quick because I’m getting more to the point where the chat point of view is becoming more more and more of a I can skip it even if I’m just getting started because it’s a tedious step,

10:59 started because it’s a tedious step, right? Because you have to provide a lot of information. Okay. Okay. I understand. I think this is a lessons learned. Like you don’t in order for you to get to the next stage, you’re going to start this way. Like I think about like my family and in my family how people interact with like AI differently than I do at work. In my work AI is is very much catered to like my workflow and how I’m using it to edit files. But my family family calls they they talk to Claude or they talk to chat GPT and

11:29 or they talk to chat GPT and they’re like they just call it chat. Chat said chat said this and so it’s just a question and an answer and I also I think now our minds are starting to shift as well. When you go to Google and you type in any search question, the first thing that returns is some AI-generated answer. Always now. So So Google for a while there Google was slow to embrace AI in search and they would not do it. And they were like, “We’re going to cut into all of our advertising, marketing, and stuff like that.” And so what I feel

12:00 and stuff like that.” And so what I feel like I’ve seen now is the first page of any search results from Google now is like nothing but ads. If you If you look down that page, it’s like a whole section of ads, two or three really good answers, and then a whole section of more ads. See? See? So that you hopefully click on something. You’re proving my point. at the top of this page is just AI. I think you’re confirming my point here because I think that crawling stage just asking questions in a chat interface, to me that’s making the assumption that this is also the first time you’re using

12:30 this is also the first time you’re using AI. Because most people nowadays, the majority of people, I think especially in our space, have at least chatted with a chatbot, whether it be a ChatGPT or to your point, can’t get away from it from Google. So for me there’s more steps involved. No, I don’t think so. I think I think that is the first step. I think that that The fact that our The fact that our

12:51 that The fact that our The fact that our audience are our groups of people that use Power BI, the fact that you’re probably already using this, great. That just means you’ve already been crawling. And maybe you don’t You’re crawling and you didn’t even know it. But I still think this is the first step. I still I still agree that this is this is the right first step. It just so happens that a large portion of our audience is already doing it. And I think more and more of the general public will be doing this one of very soon if not already. Here’s why I think the chatting by itself in in its own separate harness is

13:22 itself in in its own separate harness is misleading, too. So she mentions here that And again, I would agree with this if you’re the first time using AI really, but how do I write a DAX measure that calculates a rolling 12-month sales? How do I build small multiple charts in Power BI? Create a theme file that uses hex colors. Those are all fine, but I’m realizing more and more, Mike, without the context of what model what you’re trying to do in the model and what the model is, these questions can be more misleading than they can be helpful. Yeah, if you want to learn, “Hey, what

13:53 Yeah, if you want to learn, “Hey, what are some best practice what people should I follow on,, for Power BI?” It’ll It’ll say, “Not explicit measures podcast basically,.” Yeah, whatever you do, don’t listen. But I think it’s really hard because you get general answers, right? You’re just going to get very high-level, here’s a rank X and then you’re like, “Okay, so how do I actually do it?” It can be helpful. Fine, but again, that’s also like, “Okay, how do I cook,, chicken piccata?” That’s fine. There’s a general way to do that, but let me propose something to

14:24 that, but let me propose something to you here. And I’m going to go right almost into her run a bit here where if I’m someone starting off, I already use Power BI and I already know how to use a chatbot, right? Let’s make that basic assumption. Because I really think we can. I don’t think that’s a far stretch. If my organization installs for me the Power BI Semantic Model MCP server, I can get away with doing the crawling just like Megan’s talking about here, but in a lot more effective way. Because

14:54 but in a lot more effective way. Because you’ve mentioned this. If you have an MCP server installed on your machine, you can actually start with what can you do? What’s possible with this MCP server? server? I think the general questions, the general output is, I would even call that even before crawl. To me, crawl is much more with the workflow you’re trying to do. Because again, let’s assume here you’re already using Power BI. You have used a chatbot before. It’s more steps to really understand how

15:26 It’s more steps to really understand how these things work together if I just have MCPs installed, which again my IT department can do, and then you start with, “Hey, I noticed that you have an MCP server Power BI. I have no idea what to do. Tell me some of the things that you can help me with out with.” And now now you’re beginning to actually be in motion,, in terms of to take that analogy of crawling, walking, running. You’re You’re actually getting somewhere, right? I think crawling with this the chatbot, you’re just crawling

15:56 this the chatbot, you’re just crawling around in a circle. The point of this is to move forward and be,, more effective in the work you do. So, let me throw that to you where if my organization already has the MCP installed, I don’t have to install it, to me that seems more like a good place to crawl or at least to walk. Yeah, I I agree with that one, Tommy. I think that it’s a good place to start, but again, I just it this is all going to be like a learning progression for everyone.

16:26 a learning progression for everyone. other side note, Tommy, I had to I had to divert you here just slightly. I went over to my Claude instance and said, “What are the some of the best podcasts Power BI and Fabric podcasts?” Number one was Explicit Measures. All right. That’s cool. One of the most one of the most is the one most people in the community are pointing to first, right? And then it says Yeah, then the next one it says BI Focal, and then P3 Raw Data by P3 Adaptive Rob Collie.

16:58 P3 Raw Data by P3 Adaptive Rob Collie. Anyways, it so that was funny that it listed us first. It It must know me. It must know I run an Explicit Measures podcast. And, Claude probably understands that it I make it edit all of our short videos anyways. It knows it better say it first or else it’s going to get the Yeah, exactly. Or I will use you. That’s why you don’t say it first. What is that? Sleeping Beauty? Mirror, mirror on the wall, basically? Yes, yes, exactly. Maybe it’s smarter than I think. Anyways, I just thought that was funny. All right, go back back to the back to the crawl, walk, run. so, I guess we exhausted, I

17:29 so, I guess we exhausted, I think maybe just the crawling part, Tommy. I think we might slightly like disagree a little bit around what crawling looks like. but, what what does a walk mode look like for you here? And I I I think I align fairly well with again, if I think about my progression of how I used AI agents. agents. This feels very natural to me. All right, I would have Power BI Desktop open on one window. I would ask an agent or something else, even Tommy, me, I was pushing very heavily on build models in the service.

17:59 models in the service. The service is everything we want to do for building Power BI models and things. And so, And so, I spent a lot of time in powerbi. com opening up semantic models and writing measures and building things directly inside powerbi. com. I love it. I think it’s a great experience. But I didn’t want to pay for Copilot inside Fabric. It’s too expensive. So, I used the free Copilot, which was go to my browser window, open up the Copilot, and talk to it there. So, I would use that, or I would go over to OpenAI and

18:30 that, or I would go over to OpenAI and and go to ChatGPT. I would talk to it there. I would ask it to build things for me. Build me a measure. Build me this DAX., here’s a pattern. Write me the SQL. That was my first entry is like I had an actual website open or I had something I was working on, and I was using supplementary questions and I have to give it a lot more context. Then, I think my next logical step was I really started getting into GitHub Copilot. I turned that on. I started getting really into it, more so than Claude Code, cuz Claude Code has been more of a recent

19:01 Claude Code has been more of a recent investment for me as far as understanding and time, but only because they’re like an extreme leader in the space, and you have to know what they’re doing. I’m even finding out, Tommy, my clients are heavily using Claude in various forms

19:15 heavily using Claude in various forms across the organizations. Certain departments, specific users. And so because of that, since our organization, since my customers are using Claude, I decided I also have to know how to use Claude and be able to aptly talk to them or directly to them around what Claude is doing. So that So that’s one of the reasons why I’m Yeah. Yeah. heavily investing in both GitHub Copilot and in Claude because those are the two I think primary tools that organizations are using. I I think it’s a quick caveat there too or just a note on why Claude and

19:46 or just a note on why Claude and I think a lot of people who listen go, “Why are they always talking about Claude and not ChatGPT?” Right? That is the most popular one or has been. has been. To me I I look at this is not just so much that what it produces, but Anthropic, the company who has built Claude, has been on the forefront of for skills and then the integration with your machine and MCPs much more than ChatGPT has. That was the direction I think the Anthropic team decided to go was, “We want this to be part of the workflow.” Then when they had Claude co-work, when they had

20:17 co-work, when they had Claude code, and then just that integration with your actual files. And also also They were the ones who developed skills and then they made it open source. For me, that was a tipping point because it was getting to a point with just the chatbot and even in VS code we’re like, “This is fine, but it’s still this general output.” And the fact that skills and that Claude can talk to everything so seamlessly, it’s hard to now go back. So I just want to put that out caveat there why do we talk about Claude so much?

20:48 do we talk about Claude so much? I think you’re right about that Tommy, but also but also I think Anthropic was one of the first companies and you can read articles about this one. Most of the net new features that come out from Claude code and Anthropic, those are almost 100% built entirely by AI itself. Yeah. Yeah. So, I think Anthropic was the first company that was able to leverage make the AI good enough to build itself.

21:18 make the AI good enough to build itself. Now, you still direct it. There’s still things that are happening there, but I think Anthropic was the first company that really publicly showed that that it could do this. this. We’re we’re we’re actively talking Now, Mhm. Mhm. Chat GPT may be doing this already. May have been doing it first, but they weren’t as publicly open about how much information that Mhm. Mhm. the the Chat GPT system was using. So, for that respect, I look at this going, “Okay, well, that’s interesting. we we now have, Claude doing

21:51 we we now have, Claude doing better things.” So, anyways, this is where I think we intersect things and I would I would I’d also argue, Tommy, while I do use Claude, I don’t really like their code editing stuff. stuff. They have a code editor. There’s one in Claude. I don’t love it as much as I love using Claude inside VS Code. VS Code Code is like my favorite harness to use. And there’s a number of articles that also describe if you’re not the harness, you’re the AI, AI, the large language model. And

22:22 the large language model. And I maybe would even argue maybe more of your results from an AI even get higher now, but now, but if you don’t have a good Mhm. Mhm. harness, harness, the AI will not produce good output, period. Like 70, maybe up to 80% of the output of your AI is reliant on how good the harness is, period. 100% 100%. I think that’s also why we, talk sweet or we give flowers to Claude and and also VS Code. My preference is the cursor. Well, let’s

22:52 preference is the cursor. Well, let’s actually talk about VS Code a bit here because for Megan, her walk is your code editor. And I’m I’m actually going to disagree here. I’m going to change up her run and her walk because for me I’m again if we’re assuming someone’s just using a chatbot and just getting started with AI, odds are they’re probably not having not been using an IDE before or really the pip files. Maybe not necessarily where they’re going. going. And I think when you introduce VS Code,

23:22 And I think when you introduce VS Code, pip, you introduce tim bill and now you’re getting to the AI, well AI, well you’re giving a lot of extra,, or things that people need to know about the tooling. Her run is skills and the MCP server. Which to me is a better workflow. If I’m starting with a chatbot, well, I can use an MCP server in a chatbot, right? I can use Claude and say, “Look at the Power BI model. What can you do with since you can you can connect to it?” Okay, now what can do

23:52 connect to it?” Okay, now what can do since you have MCP? Skills are super easy to get started with in Claude in the Claude desktop application. And I think to me when you go to VS Code, that is a full development life cycle. That’s a full development view, right? That’s your ability to look at the files in a very distinct way. Claude Code still gives you that same user interface. So for me, Mike, I wouldn’t recommend someone goes from asking questions about DAX and diagramming and what’s the best way to use rank X. Say, “Okay, now we’re

24:23 way to use rank X. Say, “Okay, now we’re going to get VS Code. Oh, by the way, now you have to get to something called pip. Welcome to tim bill. And now you’re going to start chatting, with the chatbot because I think that really breaks up your your motion. Mhm. Mhm. So what’s your thoughts there? I don’t know if I have a lot of thoughts on this one, honestly. Like I, I just I’ve just fallen in in maybe my maybe my the the the pitfall or the the rut I have fallen into is I’ve just started using it and just

24:54 started using it and just started figuring out where’s, what works for my workflow. And then one of the things that I think is interesting here is with these AIs is you can make a workflow that is custom to just how you think and how you work. It still gets a lot of things done, right?, how how do you like to How does Tommy like to build reports? Tommy, you do a a lot more work around planning and thinking and like working with customers and getting a lot more requirements up. Mike may not take that same approach. I I may have focused a bit more on like, okay, well, help me understand at least the relationships in your data. Let me

25:24 the relationships in your data. Let me get some stubbed out version of this. And I might come to come to the user with more of my opinion on how the report is going to work. So, I want the agent to a guide me build guide get me to a place where I’m building more prototypes first, right? So,

25:39 prototypes first, right? So, each user the one of the things that I think I’ve internalized here is when you use an AI, AIs are here to help you build hyper-personalized experiences. And what by that is aside from like you have to pick a code editor, right? Aside from simple things. You have to all the skills are can be customized to you. All the workflows you build can be custom to you. What makes sense, right? So, I want another example here, Tommy.

26:09 So, I want another example here, Tommy. that’s been really useful for me is I use Opus Clip or used to use Opus Clip to build to build short videos for all the social medias. No longer. Don’t use it. Don’t need it. I have built my own AI generated generated I have I have built my own AI thing now. And now any shorts that I make, I just tell the AI to Here’s the long-form video. You go get the transcript. You do all the word writing. You build all the the scripting. you you come up with little for every single

26:40 you come up with little for every single short, come up with a scene and edit. And so, it cuts down all these really interesting things for me. It makes it really easy for me to like produce what I want to produce. So, This is where I think things are going to get really interesting now is because you can build extremely customizable things. And so, how you get to that customized process, I think might be slightly different for everyone. And Mike, I I think this is a good point, too, because do you use VS Code

27:11 point, too, because do you use VS Code when you when you built that application or that project? That is my preferred tool to use. I I think I get around VS Code the best. One thing I’ve been trying to use recently, if you go into VS Code, there is a second second There’s this thing called the agent window. Have you used that one yet? Is by our cursor. Yeah, it’s been 100%. So, it’s just a pure agentic window without looking at all the files and all the normal things you do in an IDE. IDE. The files are still there. You can still

27:42 The files are still there. You can still see them. You can still look at the like it’s not hiding them away from you, but it’s like the emphasis is not on the focus isn’t like the code and the files. The focus is talk to an AI and have it work through things with you. So, the shift is So, right now when you look at VS Code today, it’s code first, right? Chat window, terminal, and then see see files. That’s the main focus. And then on the side, the right hand panel is, “Oh, by the way, there’s a chat thing on the side.” Right. Right. VS Code like every one of these major

28:12 VS Code like every one of these major other code editing solutions, Claude and Cursor, Cursor, they put the agent front and center. Mhm. Mhm. The chat window is the middle of the screen. And your MCPs and your tools and the the conversations you’re having with it are now on the periphery, on the side. And then, the files are now just on the side. And this is what I feel like also is happening at Replit. Replit is doing a very similar thing. Yeah. Yeah. It’s front and center. Talk to the AI. Have it build things for you. And then,

28:42 Have it build things for you. And then, oh by the way, there’s a bunch of files behind the scene. But Right. Right. less and less, Tommy, am I actually editing or going into the files? I’m spending less time there. I do want to see what it does. I am reading through the code and seeing the diffs that it’s doing. And maybe approving or not approving based on what I think it’s doing correctly, but more often than not now I’m not going in and directly editing files. files. And this is an important point, Pete. I think this would go into walking to run where where I would I would feel weary giving

29:13 I would I would feel weary giving someone going from the chat right to an IDE where they have to look at the files and they can see the chat on the right hand side. Like it’s a It can be intimidating for your first time, right? I wouldn’t even call it an IDE anymore. I think that’s even changing. It is, but not today. Like I think yeah, we’re going to get to That’s a good point, Mike, because as of today You’re calling it that right now. Like an IDE I would think is Visual Studio. Like Visual Studio is an IDE is an integrated development environment. VS

29:44 integrated development environment. VS Code Code Yes, it is, Mike. But I would I would even No, I would say VS Code is a is not it is a lighter weight development environment. And even now with the agents mode, right? So, we’re moving the UI is moving further and further away from what I would consider a traditional IDE. You say IDE to me, I think VS I think Visual Studio. Bloated, lots of packages,. NET, really heavy, big big system, right? That’s

30:15 heavy, big big system, right? That’s what I think of when you say IDE. I don’t consider VS Code an IDE because it does so many other different things. VS Code is a much better lightweight version of an IDE. Yeah, well, that was the intention. And maybe that’s how it was designed, but I I prefer using that It’s simple, it works on the web, it works on my local computer, it works on Mac and PC. It’s not a thousand extensions I need to Like I hate using VS Code. Or sorry, Visual Studio. Visual Studio is I can’t stand it. I will only use

30:45 stand it. I will only use Visual Studio Code. And we’re going to we’re going to agree to disagree on whether VS Code is an IDE, but I think the important point is it’s a somewhat of an development environment, right? The point you use VS Tommy, you’re wrong. I’m literally asking this to on Google. I said I literally typed in IDE meaning, it said integrated development environment. Yeah. Yeah. The question down below, people are asking is VS Code an IDE? And what’s this? It says

31:16 And what’s this? It says answer is Here’s the answer. Technically no. no. Visual Studio Code is a source code editor, not a fully integrated development environment IDE out of the box. It’s developed by Microsoft and it’s designed to be a lightweight and fast text editor that’s focused only on core core file editing capabilities. However, because of its massive ecosystem and its extension, you could say it’s a highly capable text editor acting like an IDE.

31:48 capable text editor acting like an IDE. So, it’s saying both to It’s the view of this is how I perceive it. it. that’s what I’m saying. It’s basically basically It’s basically an I IDE because it supports debugging, computer interpreter building, project build management. You can

32:05 can Fine. Fine. Not without all the extensions. Fine. Well, guess what? That’s part of it. And that’s fine. I’ll move on. I don’t give a I don’t I don’t think you’re going to win this one, Tommy. Whatever. It’s, it’s not a code editor and it’s not an IDE. It’s a It’s a text editor. It’s It’s That’s what it is. I would agree with that. that. to move on from here. So, why would you use a text editor? , that’s all we Microsoft called it an IDE.

32:36 Microsoft called it an IDE. No, it’s not. You can do interpreting building. You have The difference with Visual Studio is it all comes integrated out of the box. With VS Code you have the extensions. And then you have the extensions, does the same damn thing. They think Oh my goodness, you’re getting me heated up, Mike. Let’s move on. Sorry, Tommy, you’re wrong here on this one. one. Listen, no, we’re moving on. We’re just going to move on and agree to disagree here, still. All right, so I think We’re going to disagree that you’re wrong and we’ll move on. Okay, next topic.

33:06 topic. I think that as we get to near the end here, because again we’re going to keep it brief is I think for me I would look at the walking stage and really as soon as you whatever you want to call the walking stage in whatever environment you want to do that, be it a text editor or in a chatbot, I really think at this stage, Mike, it’s really required that you also get a har- you required that you also get a har-, get a handle on skills. I don’t know, get a handle on skills. I don’t think you’re going to do well walking or going to run into you’re going to walk into a wall

33:37 into a wall really without actually being able to be comfortable with skills by the walk stage. stage. And and Megan really clearly calls out that skills and MCP servers are the run stage. That’s that’s really when you’re like really finding success with AI agents. And I would agree, Tommy. I think that I think I agree with that one as well., when you’re letting an AI work with your code collaboratively, that’s the way to go. Like that’s the next step for you to learn. I don’t think a lot of people understand how this works. I don’t think a lot of people understand how to like use AI with their text editor and and

34:08 use AI with their text editor and and edit things there directly. But what But once you once you you spent you need to spend some time there Yeah. Yeah. just to get familiar with it. And then you move on to the next step, which is, okay, great. You find that I’m saying the same things over and over again, and that’s where skills pick up, which is, okay, we built something together, AI, learn this, right? I need a reusable process that I can give to multiple

34:38 process that I can give to multiple agent sessions over and over again. That’s what skills are doing. Skills are memories and process documented. That’s how I see that. And then the MCP side of things is a whole bunch of extra tools you give capabilities to your AI, right? So, it’s almost like,, your AI has its hands tied behind its back until you get an MCP server and then the MCP server allows the agent to have understanding of different tools and how they work. The MCP server for Power BI modeling tells you how to work with a

35:08 modeling tells you how to work with a semantic model. It knows how to write Tim Dale. It knows how to correctly edit the semantic model using the XML endpoint. All the tools there are good. And tools, I think, are scripts with clear defined inputs and outputs. That’s what a tool is. That’s a good way to Because then the agent can say, “Oh, I need to make a table.” Instead of the agent going in and trying to write the code to make the table, there’s already a script available that says, “Create table.” And you give it parameters and it does it. I think that’s what tools are doing for the agent. So,

35:38 are doing for the agent. So, anyways, I’d agree with this article wholeheartedly. Megan did a a knockout job here. I think this is really, really good. she gives some very practical examples. she does touch on one area here that I think is understated in the article. article. Source control. if you are ever using an agent to modify files on your computer, my biggest word of guidance and advice would be is you need to at least have some Git or Git commit

36:08 have some Git or Git commit history or integration of a file versioning because the agent will screw things up. I’ve had it happen to me a number of times and I’ve had to revert back to what it previously was. So, anyways, that’s that’s my maybe word of advice in addition to this article. Tommy, any final thoughts? No, honestly, Mike, I think it’s a great article and especially for people getting started. Definitely take your steps. Don’t just try to run into this or you’re going to run right into a wall. wall. Yeah, it’ll be I I think it’ll be a a little bit too confusing. And you don’t maybe you don’t need to spend a lot of time walking and running or walking and

36:39 time walking and running or walking and crawling. crawling. Right. Right. But just spend a little bit of time there just at least you get your head around like what that step is cuz that is that is a good progression. And one of the things I think also is very interesting here is Megan and I have had a number of conversations around how I got started with AI and I’m finding it’s very interesting that she’s writing articles about how she’s getting started with AI and a lot of these were conversations we were having and I was like, “You got to check this out. You got to use this. You got to check these things.” And so she was I I think in the initial phases of this she was having some really big aha moments as we’ve been talking about how pro AI I was. And

37:10 been talking about how pro AI I was. And I think at the beginning she was quite skeptical about AI, not really bought into the values of it. And so to see her now coming full circle and finding value from it and and now teaching people in her way how she got started with AI, I think it’s extremely useful. useful. Love it, dude. I absolutely love it. All right, man. Well, hey, great conversation today. All right, Tommy. Where else can you find the podcast? You can find us on Apple, Spotify, wherever you get podcast. Make sure you subscribe and leave a rating. It helps us out a ton. You have a question, idea, or topic that you want us to talk about

37:40 or topic that you want us to talk about in a future episode, head over our the powerbi. tips/podcast, leave your name and a great question. And finally, join us live every Tuesday and Thursday a. m. Central and whatever. So on all Power BI social media channels. Man. Oh, boy. Oh, boy. Trouble flying. Yeah, landing that plane. plane. Geez, Tommy. We’ve had some trouble here recently. Can’t can’t quite land that one very much every every so often. Thank you very much and we’ll see you next time.

38:10 Next twisted measures pump it up be a hype. Tommy and Mike lighting up the sky. Dance to the data laughs in the mix. Fabricate a narrative get your feels. Next twisted measures, drop the beat now. Rock is king, feel the crowd. Next twisted measures Explicit measures. Try

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.