Importance of Skills for the Fabric Developer - Ep.528

In this episode, Mike and Tommy dig into why building the right skills matters for Fabric developers and how that thinking changes what you learn first. They also discuss how to evaluate skill depth, avoid tool-chasing, and create a plan for steady growth.

News & Announcements

• Watch Episode 528 on YouTube — Join the live conversation and catch the full discussion, including the Q&A and examples referenced throughout the show.

Main Discussion

Topic: Importance of Skills for the Fabric Developer

• Skills beat tools: focus on fundamentals (modeling, SQL, data engineering patterns) before chasing whatever feature shipped this week.
• Depth matters: being “good” often means understanding tradeoffs (cost, performance, governance), not just getting something to run.
• Learn in layers: start with a small end-to-end scenario and expand your capability by iterating on real problems.
• Build a portfolio: small repeatable projects create compounding learning and make it easier to explain your value.
• Don’t skip the boring stuff: version control, documentation, naming, and testing are the difference between demos and production.
• Adopt new Fabric features intentionally: evaluate where they fit in an architecture instead of swapping your stack every month.

Looking Forward

Pick one practical scenario you can repeat weekly (ingest → model → report), and use it as your baseline to measure skill growth over time.

Episode Transcript

0:01 Be it high. Tommy and Mike lighting [music] up the sky. Dance to the day to laugh in the mix. Fabric and A. I get your feels. Explicit [music] measures. Drop the beat now. H feel the crowd. Explicit [music] measures. Hello everyone and good morning. Welcome back to the explicit measures podcast with Tommy and Mike. Tommy, good morning. morning. Good morning. Just so, if

0:31 Good morning. Just so, if there’s an album of the guy who wrote our podcast song, I would I would buy it. It is so catchy still. The the intro. Oh, yeah. Oh, yeah. I’m still making songs, man. Every day I’m like making more stuff. I’m checking your Spotify. Yeah. I just every every couple weeks I’m putting out like another handful of songs. I think like a week or two ago I released a couple more. It’s I’m just slowly ripping them out. DJ Carlo, man. Yeah. the I’ve been pushing my kids a little bit because I’m now making

1:01 little bit because I’m now making some songs about like things like around the house or family or things that we’re interacting with. So it’s fun. I enjoy it. It’s actually well two things. First one is my kids are now singing the songs that I’ve written. So we’re I’ll play them and they like they like, “Oh, I like this one.” And then like they’ll sit down and they’ll sing it. I’m like, “Okay, that’s fun.” Like we have a little like Carlo family,, thing that’s just us. It’s our our little music. Yeah. Yeah. Which is fun. It’s actually cool. Yeah.

1:32 It’s actually cool. Yeah. Are they all though like club songs or do you have a few more that are a little back for a fireplace? I’ve branched out slightly. I have a handful of like rap songs. I have a handful of like I’m not a big fan of country or twangy country, so a lot of them are more upbeat. I have some a I have a lot more of like classic rock. Okay. or like heavy guitar solo rock stuff. So I do a lot of like rock rock like rap music and then it’s been like some just like pop EDM dance mix

2:03 some just like pop EDM dance mix stuff. Anyways, so the first one is like the family is now singing things. But the second point here I want to make is like it’s so easy to make these things anymore. I think I’m a little bit more invested in listening to the music that I create than what other people create for whatever reason. Cuz there’s like a story a personal story behind why I’m saying it. Like one of the recent songs I released was I don’t want to work, I don’t want to play. And I was just I was just sitting here. I was like, “Oh man, I don’t want to work. I just want to play.” Like, “Oh, that

2:33 I just want to play.” Like, “Oh, that could be.” And then I just wrote a little song around it. So I I that’s another one that just I think just recently got released. And isn’t that already a song though? I don’t want to work. I want to play on the drums all day. Bang. Bang on my drum all day. Bang. Okay. Okay. All right. All right. Slightly different. Yeah. Not as good though,. Not as good as yours. Not as not as good as the one I what I just wrote. so anyways, that that’s been fun. and now we So the challenge now is we I have a handful of these

3:04 now is we I have a handful of these songs that I have not released. They’re not out there. Oh, and I will say this is fun because now Carlos Solutions is an artist in Spotify. I can say, “Hey Google, play the artist Carlos Solutions.” and it now plays. It just goes right to my music and plays it for me right off of the Apple the YouTube music now. So just I can just command it things and they just know my music and it just plays. I’m like this is amazing. So that because of that I now want to figure out how do all these personal songs that just like familyenter

3:35 songs that just like familyenter centric ones. How do I publish them? How do I make them get out there? Yeah. So now I’m looking at going, “Okay, how do I like now?” I’m challenging the kids with, “Okay, you need to come up with an artist name.” And they’re coming up with the stupidest names right now. They’re like,, let’s be like Bpop Rock Steady. I’m like, no, that’s not No, we’re not going to do that. Ninja Turtles. We’re not doing that., they’re just they’re just making up just random words,, skippy music. Like, no, we’re not doing that. No, you have to think of something that would be last. Like I what what they don’t understand is

4:05 don’t understand is whatever you put out now on the internet is there forever. Yeah. But I’m going to I’m going to actually side with your kids on this one with the album name. Some of my best album album album artist some of my best artists have the craziest names. For example, back in my teenage years, my pop years, Fear Before the March of Flames was a good one. I set my friends on fire. [laughter] Yeah. So,, you can you can go a lot of places. You can go a lot of

4:35 You can go a lot of, I had some weird names, but like Yeah, I get it. But they just you want them to be like timeless a little bit, though. You don’t want to turn into a Nickelback. Oh, [laughter] let’s not mention that. Yeah. Demonetized. Demon., seriously though, I’m glad you’re not doing the country thing because I don’t care if we lose listeners. Country is not music. It is. It is. It has instruments. It has instruments, but music is not. But hey, what are we talking about today? today? Let’s get off of this topic. Let’s go

5:05 Let’s get off of this topic. Let’s go into the So, let’s talk about skills. Important skills for the fabric developer. So, that’s going to be our main topic today. The importance of skills for the fabric developer will be our main topic for today. So, that’ll be our our main topic for everything., and then we’ll jump into some news. So, Tommy, before we get to the main topic, what news articles have you found for us? got two quick ones around the fabric space and then one I think is going to be a good segue for our topic today. The first one is I guess it deserved a blog article. It’s on the fabric blog and it’s an update to

5:35 on the fabric blog and it’s an update to the semantic model settings experience. So sounds fancier than it is but what I’m happy to see this and in a nutshell what’s happened here is the settings Oh wow. Oh wow. Yeah. So Yeah. So this is a lot of change. Yeah. So rather before if you wanted to look at your settings for a semantic model, you got buried refresh. Yeah, it was a little buried. It was a different page, a different UI. Well, now it works actually works more or less like the report settings in the service. Every other setting on every other item

6:06 Every other setting on every other item that’s in there. Yeah, exactly. It works like every other setting in in fabric. And it’s simply what it is. It’s a pain. P A NE on the right hand side which allows you to really be aligned with a modern fabric user interface. It opens instantly. You’re not navigating away. So again, it’s designed for that long list of settings. There’s a search box. So because Mike, let me ask you without looking at the page, how many settings in the service are there would you guesstimate for a

6:38 are there would you guesstimate for a semantic model alone? I would I’m I I don’t know the exact number of settings, but I’m going like by like the number of sections that I remember seeing. So like you go into the set, you click the ellipsus, you go to settings, and there would be like another page dedicated just for the settings of the semantic model, right? So I’m just kind semantic model, right? So I’m just going through and I’m think what of going through and I’m think what you’re saying. I’m thinking of literally the sections that are there. Fair. Yep. 15. 15. Over 30. Over 30. Oh my word. Yeah. So we have o every semantic model

7:08 Yeah. So we have o every semantic model has over 30 configurations or toggles or switches you can do. Makes sense. it’s just it’s just providing like in desktop you could change all these things right the model’s there. All this is doing is just exposing the settings that were existing the model just like a better UI. I think this is great. great. Now it is off by default. So you have to be open from the classic settings during preview but then it will become the default experience. So it’s off by default. Yeah they’re taking time. I don’t know why this is not one of those. We just turn it on.

7:40 turn it on. It just seems like Well, maybe it’s Is it one or the other? Like you get one’s experience or the other experience? No. So, you’re on the classic one and there’s going to be one of those banners that says, “Do you want to see the preview experience?” H H Yeah, it’s not an admin setting either. That seems odd. I just would have expected just to like Me, too. Me, too. turn on. I’m not this this wouldn’t this shouldn’t throw anyone off if this is now the default experience. So yeah, yeah, great find on that one, Tommy. That’s a

8:10 great find on that one, Tommy. That’s a good that’s a good article. I like this. I this is a great improvement. I think this is one of these creature comfort things that we need to get to. someone else was throwing out a picture. I believe someone was looking at one of the blogs and they were highlighting everything that went G for a lot of the visual side and there was like nine things that went GA or something like that. A crazy amount of things. I was like, “Wow, we’re finally getting things out the door. We’re finally getting things finished and completed and moved on. on. I think this summer we need to do another kill it or complete it segment.

8:40 another kill it or complete it segment. Oh, I think you’re right. I think it’s about time. So,, we got another quick article here and I’m intrigued to hear your thoughts. This is now about the DAX queries REST API. So, this is a new execute queries. This is execute DAX queries, right? Yeah, that’s actually a very very good point here. This is double tax queries. I I had to yesterday I was trying to understand. Yeah, I was yesterday literally the same thing. I was looking at the the other

9:10 thing. I was looking at the the other one. I’m like what’s this new thing? And I saw some of people talking about on social media. I thought this is cool. I’m like isn’t we already have one of these? Like what the heck? Why do we have another one now? Oh, this one’s different. Okay, so go ahead. Go ahead. Yeah, some of the differences here is the API returns rather than returning in JSON returns an arrow. So this preserves the native data types and improves performance. Removes fixed rows limits. You can submit multiple DAX evaluate statements in a single request. Also a little different available in PowerBI premium in

9:40 available in PowerBI premium in Microsoft fabric capacities only. So you must be able to process arrow streams. Whatever caller you’re doing must be able to process arrow. And you can do this in multiple languages. You can do this in Python, C, Java, JavaScript. So it’s pretty accessible. You can do it in a fabric notebook. You can do it in and again any caller or client that accepts arrow and can use Python or C. So again, not is it a game changer? No. But

10:10 changer? No. But Mike, do you see this changing anything that you’re doing or testing things out? Dude, 100%. Tommy, this is this is this is revolutionary, I think, on things. So wow, that’s a big word. Yeah, this is this is going to this is an enabler for a lot of other things. So let me be clear, Tommy. I’m already building like I have content nudge which I’ve been using to help like collect information about like you collect information about like social media channels like where we know social media channels like where we post how much how much impact that’s having right so we collect our own data around like our impact or or social

10:41 around like our impact or or social media space content nudge is designed to be this application this application lives on top of fabric so I’m making a website from there I’m vibe coding the website or I’m I’m agentically coding the website And then I’m calling back to fabric to do certain things. So one of the calls we come back to is we we talk to a sim a SQL database, right? There’s a SQL endpoint, we have a connection string. I’m able to like make those calls directly. The app knows how to directly call back to the endpoint. endpoint. Same thing for semantic models, right?

11:12 Same thing for semantic models, right? If you have reporting our we have a semantic model that directly pulls all the data right from the SQL database using a a mirroring. Basically, it’s a mirrored database. And then the semantic model gets all these direct lake tables. Awesome. Very neat. Very happy about all this stuff. But the challenge of this is now that I have all this stuff operational. Right now you have to make API calls using the existing execute API queries. It’s just exe execute queries and you get a rowbyrow detail back. Not

11:44 and you get a rowbyrow detail back. Not bad. It’s just not compressed for data in transit. Arrow I believe is a column store format. So the arrow format and that’s why they’re using I believe this format because there’s like libraries interpreters for different different applications and software. Now you can send in a query and get back a compressed table in columner storage

12:08 compressed table in columner storage which you can just read and then use in whatever language you want which I think this is genius because now the data you would have to transmit between your application and the Microsoft semantic models is now less weight right right this this is incredible running multiple DAX queries multiple tables coming back at the same time this is great you can get more tables longer wider information with less time because it takes less to package it up and bundle it over to it. So I think this is really good Tommy

12:38 it. So I think this is really good Tommy and to be honest I am very close to dropping reports in general and just and just building what I want to build with like my agentic solutions these days clients. I don’t know Tommy. It’s fun. It’s really fun. I really enjoy it. I get it’s fun. I get it’s fun. It’s just really hard to set up right now. Like it’s really difficult to get like like my application talking to the SQL server. You really have to know how to connection string things through.

13:09 connection string things through. And me personally, I chose the back end of. NET. of. NET. So because I chose a. NET backend and this was intentional. The reason I chose ANET backend was because I want to be able to build with the mod client library. Right. Right. Right. So, so that gives me full control to see everything I need to see from the semantic model. The AMD AO and D library. I just say, “Hey, agent, here’s my AM to OD library. This is how we’re going to authenticate. We’re going to talk directly to the back end.” So, if I chose net intentionally for that

13:41 chose net intentionally for that library, so I could just use it in my application. This opens up another whole realm, which is like now you just need an API call to almost do the same thing., you you lose I think maybe a little bit of control here directly modifying the semantic model., but at least you can query data and get it out very fast. So, I think there’s going to be a proliferation of new development patterns that are going to come from this. I’ll keep my eye on that. So, that’s that’s interesting you’d say because to me this looks at just a pure

14:11 because to me this looks at just a pure developer testing out thing, but you’re saying this in a completely different light. So, yes, yes, we’ll keep an eye on it. I I’m intrigued to see where that goes, especially that. But the point, Mike, is it does open up a lot more compared to the just the JSON output that was previous. And let me just be really explicitly clear around this thing, right? We have a client right now. I’m building the data part of the application, right? right? And we’re now talking with the client and building net new websites and applications and pieces for

14:42 websites and applications and pieces for them. And every time we build something new, instead of letting the backend just go wherever, right? it’s a MySQL database in some system or it’s wherever it lives on top of the app. We’re now building all of the backend directly in fabric like there it’s a fabric SQL database that runs all of the operational websites and now multiple websites are getting their own schemas and they’re getting their own area where they can read and write data directly from fabric. We’re using fabric as a backend and I love this because all the

15:14 backend and I love this because all the friction and extra copying there’s there’s always a whole step of like hey I have the database it just lives over here I know to run a pipeline I got to get there’s a whole layer of like data engineering work that has to happen to get the data just to fabric that step is gone I’m not wasting any compute there but this is what we’ve talked about though too really the the limitlessness of fabric where before with powerbi it was the end of the road right anything building or building for. So the semantic model was it that was the finish line. But now with fabric to your

15:45 finish line. But now with fabric to your point, you’re still doing PowerBI like things, but you have all these other avenues and outputs that it can do, which is really really to me the selling point., outside of what else we talked about, I’m going to go back to our we talked about this in the podcast a while ago, Tom, and I want to keep I want to re-emphasize this again, right? Think of the semantic model as the pinch point, as the lynch pin, as the centerstone of what you’re building, the semantic model. Now, we can debate whether or not it’s one semantic model

16:15 whether or not it’s one semantic model or many semantic models, but the idea the concept of semantics and related data and calculations. That’s the that’s the lynch pin of what we’re doing here. And on the one on the left hand side of this, you have all these data engineering tools which fabric provides. It’s pipelines, it’s lakeouses, it’s notebooks, you have KQL, real-time streaming, like all these other things on the other side of the world that you can use and like yeah, you can connect to them directly, but my mental model really is bring all that data down to

16:46 really is bring all that data down to the semantic model. So to me, it’s like a a funnel that all of these data sources and processes are then helping me curate, groom, protect,, clean whatever down to the semantic model level. And then once we get out of the semantic model, everything expands again. Now I’ve got reports. Now I’ve got pageionated. Now I’ve got explorations. Now I have Mike’s building apps on top of the semantic model. So this API like let me rough out the architecture. this API getting DAX

17:17 architecture. this API getting DAX queries back from a model, presenting them to a user, letting the user interact with the data in some way, some form, and then writing the data back to a fabric SQL database from the application is now one step easier and and more useful to people. So, I really think I think we’re moving into an era, Tommy, of the agents are accelerating the ability to make data apps easier. It’s always been there. It’s just easier now than it ever has been before and I

17:47 now than it ever has been before and I think this is a step in the right direction direction and it’s cyclical in a sense or because like for example I’m working on a new project and the purpose of that project in fabric is analytical and agentic right it’s to get their data into a format that can actually run for agentic solutions that they’re trying to do because previously it was it’s built I’m obviously I’m not going to tell what application it’s in but it’s in some more application data source y and again this was not possible for a PowerBI engineer back in the day. You

18:17 PowerBI engineer back in the day. You built everything for reporting and you did everything in Power Query, maybe you had data flow. So, we really are, I think, finally seeing what Fabric’s supposed to do, right? And I think we’ve talked about this and I think a lot of organizations are finally seeing, okay, fabric can do more. It’s more than just PowerBI 2. 0. And I think that was the misconception there. There was a moment in time, Tommy. Tommy. Okay, I’m I’m gonna I I’m recalling a memory I had many many months ago when

18:48 memory I had many many months ago when Fabric SQL came out for I think this was one of the announcements at like either build or one of the FabCons or something like that. It was Ignite Chicago. Ignite Chicago, right? One of the events they announced sequel in Fabric and I’m like yeah, I don’t know about that. I remember this. Yeah, I was very like but I wasn’t it wasn’t like a I don’t like it. It’s not going it’s not good a good fit. It was more of like a I’m not sure if this is going to land the way people expect it to land. It was I was just a I was just a bit hesitant to say to say I’m not sure I love it. I’m going to

19:18 I’m not sure I love it. I’m going to hold my I’m going to reserve my preference until the future. Okay, here’s the future. I’m in the future now. This is future Mike talking to old Mike. Mike, you were you were you were okay to hold on to your hesitation, but it worked out just fine. It was not worth you. It was not worth it for you to worry about because now I look at it going, “Yeah, this this feels robust.” Like the things that I can do now. This feels clean. This feels like we’ve got the capability. And to be honest, Tommy,

19:49 the capability. And to be honest, Tommy, it wasn’t it wasn’t the fact that Microsoft provided the APIs and the tooling. It was the fact that agents got so good that I feel confident building on top of it without any issue. That was that’s the I didn’t see that coming. I didn’t see I didn’t see the influence of agents in that comment I made a while ago. I just thought, “Oh, this is interesting. It’s going to be there.” Like maybe we’ll teach people how to learn how to work with it. But dude, now that agents are here, Mike Mike not scared about it at all. I don’t know if you realize what just

20:19 I don’t know if you realize what just happened. Future Mike couldn’t even tell past Mike that he was wrong. Mike couldn’t even tell Mike that he was wrong. wrong. I wasn’t wrong. I just was holding my I was just holding my res. I I said in the in the last episode I reserved my I was reserving my comment around like not not I I didn’t want to commit to it yet and I’m like now you can do it like it’s not like really advice. This is more like you you weren’t wrong in holding back but you don’t have to worry about it. It’s it’s going to be just fine. And I think it again the thought though here is it’s not the fact that

20:50 though here is it’s not the fact that SQL showed up in fabric. It’s the fact that agents are getting so much more capable that these really technical things that you just didn’t know how to run or build or do, you can just throw documentation at an agent and it now knows how to build these things. And and so that’s that’s what I’m finding. That’s the part I didn’t see coming. We we I think we we’re in the same conclusion now, but I think we’re always going to diverge on SQL runs the world and it has always run the world. But I think we’re in the same place now. But for the sake of time, Mike, and I think a good segue, I got a skill for you. And

21:22 a good segue, I got a skill for you. And maybe this becomes a part a new segment for us, too, called share a skill where we share. Yeah. So, this one’s a good one. And [music] we’ll use a little dream animation there. All right. Tommy’s going to share now to sit down in the corner with Tommy’s shares of skill. You need that music of like where they’re like talking about the old history book like welcome to TGF, you history book like welcome to TGF,, Turner classic everything movies know, Turner classic everything movies thing from 1952 thing. Tommy will now share a skill.

21:52 Tommy will now share a skill. This is called I like that. I like that. This is called Grill Me and I love this. This is This is I told you I said I’ve talked about this one before. This is from Matt Hawk. Yeah, Yeah, he’s amazing. So this one is interview. I’m gonna read what the skill says and then I want to hear your thoughts here and how we can apply it. And this is a perfect segue for our topic. This guy’s one note, this gentleman by the name of Matt PCO. He’s an engineer. Yeah. Amazing. Follow him. He’s so good.

22:22 Yeah. Amazing. Follow him. He’s so good. And his stuff that he puts out is gold. Go ahead, Tommy. Well, this is great. So, this is what the skill the instructions are. Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of a design tree, resolving dependencies between decisions one by one. For each question, provide your recommended answer. Ask the question one at a time. If a question can be answered by exploring the codebase, explore the codebase instead. Now this is obviously

22:52 codebase instead. Now this is obviously meant more from a developer point of view but I don’t think it takes a far too much imaginative thinking Mike to think about this from a fabric point of view or even designing a project around data. So I saw that this I saw this this morning actually and this looks phenomenal. I love every aspect of this. I wish I saw this yesterday actually when I was writing a statement of work. This is a solid skill. And not only is this one, it’s really short. So some skills you’ll look at have like, you

23:23 skills you’ll look at have like, you skills you’ll look at have like,, know, three 400 lines of code. It’s a lot of things in there. So this one is incredibly short. And Matt has directly said on his YouTube channel and and the social media that he puts out, he goes, “This one skill, even as brief as it is, has been revolutionary one in his workflow, but it’s gotten a lot of attention by the community as people are then absorbing this skill and using this. So, two things. I linked the entire skills repo from Matt inside the chat window. I’ll make sure I put it here in the description as well at the end of the show. And then also, I have

23:54 end of the show. And then also, I have the grill me skill the skill specifically where you can go read the exact skill that he wrote., and you can see that one as well. And the the part here that I think makes a lot of sense in here, it says until we get to a mutual shared understanding. Yeah. And then at some point like he said it’s gone 30 45 questions

24:14 said it’s gone 30 45 questions sometimes of just asking questions and if there’s anything that I find I I just saw another tool from a gentleman that I was connecting with. He has built a he calls it the crucible and it’s this idea of like having multiple agents pick apart an idea from different angles of things, right? You it’s really neat. You can use the same large language model and give it different instructions and it can pick apart your idea or your thought or your design three different ways. And this is

24:44 design three different ways. And this is one of these ideas of like really making sure that it understands what’s going on. And I think again if you look at what Matt PCO is doing in this pattern, pattern, he comes up with an idea. He has a solution, feature, whatever he wants to build. He presents the idea to the agent and then says, “Grill me about it.” It it goes through the grilling exercise and then the from the output of that he says, “Okay, based on what we’ve talked about here now, now go write up detailed instructions on how we would deliver this, go turn it into issues and then go chew on it.”

25:14 chew on it.” Well, so like for example, just putting together a project on architecture around fabric, right? Like if you already have the you can combine skills and I don’t think people realize this too. You can have the same conversation with using multiple skills. For example, use the agentic skills that Kurt Guler or Alex have built and use this grill me in that process right because this is great for a codebase but again a lot of people go well that’s great but how do I know that this has all the

25:44 how do I know that this has all the context of fabric or powerbi? Well, combine it with the skills that are already available around fabric. We are at a place now, Mike, where I you could not say this six months ago. We have plenty of skills for people who are fabric developers or fabric architects, whatever you want to call them, that you can now combine and provide ample context, knowledge, and wisdom to really build your workflow. This was not this was not something six months ago, but we’re at a point now

26:14 months ago, but we’re at a point now where it’s part of my workflow. I’m using skills that people have built. I have my own skills for notebooks and the architecture., but I think this is a perfect example of that conversationbased assistant, right? Because I don’t know about you, but I am maybe 10% of the time now when I’m talking with an agent typing because I am always using the microphone option that they have or using the text to speech on Windows

26:46 or using the text to speech on Windows I think Windows H and and Windows H I use I’m starting to use that more and more now more and more of my workload is shifting to speaking to this is this is the next revolution Tommy though like this is it now let me let Let me add one little bit of context here too, Tommy. When you’re doing text to speech through the Windows computer, I think Windows 8 is that’s a local model running probably on your machine doing some spec text speech to text and then it comes out. when you’re talking in the application, what you’re using like VS Code this is notion or claude. They both have

27:16 this is notion or claude. They both have a microphone feature and VS Code also has a microphone feature too where it’ll still do text to speech but it’s just in their own application regardless of which one I’m in Mike because I I it’s allowing me to think rather than type because I’m going well no actually what for section two literary vice I’m literally talking like I’m talking to a person. Yeah. Yeah. I’m going actually section two no scratch that. And I’m not let me delete that instructions here. Yep. So using things like grill me, using the fabric skills, using my own so building

27:48 fabric skills, using my own so building skills, Mike, I we’re at a point right now where if I have the backend knowledge, I have what I would pay for in the past without a doubt. I’m I’m going to I want to Yes, I agree with you. And I think the the skills that we’re learning here, so I’m trying to distill this into maybe a couple clear thoughts. So first reaction [snorts] is you’re talking to your computer. I’m doing so a lot more. I’m probably not doing nearly as much as you are. because you’re doing like, you

28:19 are. because you’re doing like, you are. because you’re doing like,, you’re on hands on keyboard like know, you’re on hands on keyboard like 10% of what you used to be doing, which is great. I also find that when you talk to your computer, it goes faster. It’s it’s faster to get your ideas from head to computer, especially if you have ADD. Yeah. Yep. So the faster I can get it from here back to the computer now, the better., so a lot of there’s another program I was called Whisper AI. Have you heard of that? That’s just transcription. So that’s not so it can do real time, but yeah. Correct. But I’ve heard it’s really good about like being able to capture, you about like being able to capture,, your audio from what things that know, your audio from what things that you’re saying and bring it in and like

28:49 you’re saying and bring it in and like really do a good job of saying like the getting rid of the ums and the ands. And like like Mhm. Mhm. to your point Tommy like it just there’s this part of like you say words you can get your text directly into the system than the computer. Ah Ah but sometimes the intent of your words are a little bit different and you need a little bit of like a translation layer on top of them. So Tommy you’ll say hey I want to I want to I want to revise section two of this statement of work and then oh no wait I

29:19 statement of work and then oh no wait I didn’t want to do that I want to do this. So you maybe you’re you’re thinking basically it’s you allowing you to think out loud to something and it can take the raw text in but sometimes I needed to like quickly summarize it. So okay what I just talked to you about take that paragraph of words and now summarize it into something that’s relevant. relevant. Yes. Yes. And so that’s the part where I look at it going okay that’s interesting. that’s what I think Whisper does. And not to say that you need to go buy it. It’s a tool that you go buy. You can

29:49 It’s a tool that you go buy. You can you can get it from GitHub. There’s a ton of open- source versions of it. That’s my point was like there’s a lot of open source whisp but the words you’re looking for if you want to go get search for things that are like whisper related. Whisper AI was like the premiere of this. Yeah. Yeah. That there’s a lot of people have copied it now and you can go get it anywhere else. And there’s also a lot of free transcription services. So I believe Assembly AI I think is is the one company that’s doing it or maybe it’s LiveKit. There’s two kits or companies that are doing this. you get like 140 minutes of like free talking to

30:21 like 140 minutes of like free talking to it. Yeah. For free. So you don’t have to pay anything pretty much. And throughout throughout the month of a a work month, you you get most of it for free anyway. So anyways, that’s another really interesting thing out there as well. Going back to your your comment, Tommy, let’s talk about the skills piece. You’re saying you’re talking about skills in fabric for developers like physical skills you use with your agent. I am talking yes I’m talking about AI agentic skills not necessarily like my skill around building in DAX correct

30:51 correct here’s the the point I want to make is building in DAX is a skill and you have this in your knowledge in your head and I I was making a comment we I have a short on one of our comments Tommy that someone commented on I think we were talking about yeah I don’t want to write DAX anymore I’m just done with that like I’ I’d rather have the agent just figure it out and do it for me I could just describe what I want and have that do the DAX for me. And someone’s like, “No, an expert always must own and use and have and manage DAX.” I’m like, “Yes, you better understand it, but do I

31:23 “Yes, you better understand it, but do I need to be the expert in it?” Does the expert like, “Where does that skill and direct knowledge come from? In order for me to absorb that knowledge, I’ve got to go read a book from Marco Russo. I’ve got to do some tests on my own. And I’ve got to understand, I got to materialize the information and put it into my head. That takes some time. I can just throw an agent at an article, all right, book, at a PDF, and it how has the same and maybe even deeper knowledge because

31:53 and maybe even deeper knowledge because it remembers everything in that book and can go through everything even more detailed than I can. So I guess my point here though is Tommy is the skills that we’re talking about here either in for agents or for fabric those skills have to be formed by people in their heads when you’re working on and building things in semantic models. And I think this is a good segue for our main topic today because I really think that we need to have the conversation. You and I have talked in spades about skills and the glory of using AI skills.

32:25 skills and the glory of using AI skills. However, I think we need to have the conversation on how important they are for a fabric developer. How much they’re one and the same. I think my my point is they’re one and the same. Like the skills that I’m using to with agents are the same skills I have. I’m just working with an agent to develop a skill that I can reuse and I don’t have to continually tap my own skill to leverage this. So what I’m trying to say is like skills are still really important for both the developer without any agent and

32:57 both the developer without any agent and then with the agent you can accelerate your skills by leveraging what you already know and packaging it in a way that the agent can use it as well. So let’s try to answer three questions today I think is what is the difference between a technical skill and the agentic skill? How can we define a difference? How important are agentic skills for a fabric developer? And what are the cautions you should have using agentic skills? I will those are I think three areas I think we need to explore because if we cannot define a

33:28 explore because if we cannot define a difference between the technical skill I have and a clawed skill or an agentic skill then I think we have a problem because we’re going to run into a lot of slop. We’re going to run into a lot of error. So let’s I I would like to start there. I don’t know if you have a direction you want to go but it sounds fine. Can you tell me? Yeah, this is something we can definitely disagree about. Let’s go. I I have what? I think we completely agreed, but I actually watched something the other day and I think I’m going to play the devil’s

33:58 think I’m going to play the devil’s advocate in some of this. But first, let’s talk about the difference between a technical skill and a tech skill. Okay. What do you think the differences are, Tommy? Yeah, the technical skill is Yeah. Okay. [laughter] I was waiting for that. So, I should have talked faster. Should have paused and should have said I disagree. I disagree. Doing it wrong. [laughter] Let’s ask you the question. I’m wrong. Yeah. Yeah. before like the first two words are come out of your mouth. Dang it. Get two words out of my mouth. Why am I disagreeing with myself?

34:30 Why am I disagreeing with myself? when we talk about technical skills, you’re talking about a single implementation of something, right? I’m writing a DAX measure. I am writing it in this context. And skill is different than what I’m going to call wisdom, right? because I may know evaluation context and I may know all of the DAX functions available to me. However, my background experience and my years of working in different projects is also provided me wisdom is providing

35:01 is also provided me wisdom is providing me with experience on hey this is going to probably cause bloat. I know if I use a filter function here over this large of a model we’re probably going to run into some issues. So I may have the technical skill of being able to write it and understanding what to write in a given circumstance. But then there’s also the background experience of many many semantic models over the years that I’ve worked on and also the idea of what you’re trying to output and again well this user also has these

35:31 and again well this user also has these weird things. So I may have to write three measures to do that rather than trying to fit that all in one. When we’re talking about a agentic skill again the skill or the skill. md file because that’s what when we say skill here from an agentic point of view we’re talking about a skill. md and the resources involved. This is simply providing your agent additional context and instructions both it can be actual code and scripts. It can be formatting

36:01 code and scripts. It can be formatting and additional instructions and references to enhance that conversation. So there’s only so much you can fit in. You can do a lot with them. Again, it’s helping me build my statement of works. But if I were just to provide, let’s say an example, the SO skill, we’ll do something a little more less technical.

36:20 something a little more less technical. If I were to say, “Here’s a transcript. Write a statement of work.” Without my experience on what hours take on what the deliverables are, that skill will write all the deliverables. It will write all the hours without my input. Now, my input is my wisdom, is my experience that comes in to actually say, what, that’s not a deliverable. No, we can’t do that in two hours. We’re going to need six hours for that. go back and forth on this

36:50 that. go back and forth on this refinement. So when I think we talk about the user skill, the human skill, you have to tie in experience. When we’re talking about an anogentic skill, you’re talking about additional instructions and context. So Mike, can you give me your own definition of the difference between the user skill and agentic skill? A lot of what you’re describing here is, and maybe I’m oversimplifying what we’re talking about.

37:21 talking about. I believe what you described to me was a combination of rules and guardrails because in your mind you formed a bunch of rules and guardrails around how I write DAX, what’s acceptable, what’s not acceptable. You’ve defined some requirements around when someone so when someone listens to you, when someone talks to you, right, you’re running their words they’re saying to you through a rubric, a grading system. They said this, I know I in my past experience, I’ve got these things. I’m going to I’m going to build some rules

37:52 going to I’m going to build some rules or guardrails around that. Right? Also, Marco Russo and and training things for,, SQLBI, , SQLBI, he’s done a lot of studying on trying different things with DAX. Well, when you do when this pattern appears in right right this processing, use this pattern versus the other patterns because it’s just the fastest pattern, right?, so there there’s a lot of these patterns or rules or guardrails that have been appearing, right? So that’s what we’re that’s part of us. The other thing I heard you

38:22 of us. The other thing I heard you describing and talking about was it also sounds like Tommy you’re you were using a thing called you’re using memory to recall other instances and and the the concepts that you worked on in the past were related but not exactly identical. So you’re doing a little bit of projecting a little bit of these past experiences and drawing them forward into like this current building experience. Okay. My understanding from a skill is in the agentic world is it’s a combination of

38:52 agentic world is it’s a combination of those things. It’s a combination of guard rails and rules and it’s also a way of capturing memory. Now I’m going to say something really weird here because I believe the the agentic world of this right so when we talk about agentic skills versus people skills the agentic skills is a distilled or consolidated or it’s a a mirror image. It’s it’s a it’s a way of taking

39:22 image. It’s it’s a it’s a way of taking the human skill and bundling it in a way that an agent understands the skill and can reuse the skill over and over again. Now your skills Tommy when you’re talking with a client and you’re working with DAX and and building things with them directly for fabric or hey I really want to use pipelines and then I’m going to definitely want to get the notebooks right away right that’s knowledge based on what you’ve what you’ve built around the skills of of your system and how you’ve decided to design an architecture the agent doesn’t know that have any of that information right right so our job as developers now is if we

39:54 so our job as developers now is if we want to leverage that skill over and over again we’re taking knowledge out of our heads we’re pushing it into skills. And so you could think of the skill. [snorts] [snorts] Let me let me I’m I’m going to also go out the reservation here a little bit, right? Anything agentic, any harness you use, VS Code, Mhm. Mhm. C-Pilot in fabric, if you’re using cloud code, like these are all the the software that wraps around the model is called the harness. Yes. The only thing the harness is there

40:24 Yes. The only thing the harness is there to do is to guide the large language model and get it to be more deterministic in its output. And what you see now is custom agents, custom skills, all these additional written markdown files that support the skills that you’re talking about, Tommy. All those things are doing is it’s giving those are all parameters input parameters to a deterministic code written harness to allow it to be customized. So if if you think about the

40:54 customized. So if if you think about the tasks you do and what you work on, there’s likely a different harness you need for every experience and how you’re building with an AI. an AI. Let me give you another example. Sure. When I work on audio and video stuff, the programs, even me as a person, the person the software I use for audio video editing is different than the software I use for text and and table building. Sure. Sure. That’s because in each of those examples, you could think of the Adobe Illustrator or the Adobe Express for video editing and making content that

41:25 video editing and making content that has a harness around the software and the output that you want. So in the same way when we look at AI agents, everyone looks at, oh, it’s just cloud code. We’re going to use cloud code for everything. Cloud code is a framework. It’s the harness. And what happens is you need to extend and manipulate and change the harness by adding additional plugins, additional features to it that enable it to actually do different things, right? You can get cloud code to edit things. You can get cloud code to make social media posts, but you need other features of it. You

41:56 but you need other features of it. You need to be able to customize the rigid harness, the deterministic harness with this non-deterministic AI model to get consistent output. Right? So that’s why we have,, really clear instructions going to the agent. Hey, when I give you this task, you must write it in this format. Here’s my statement of work format. When I write this section of the statement of work, here’s the purpose. Here’s why it exists. Here’s what I want you to write. So to do that like a skill. So let’s go back all the way to your question now, right? What’s the difference between a

42:26 right? What’s the difference between a people skill and an agent skill? I would argue nothing. The skill though is the the agent skill is derived off of the knowledge and information that the human has learned and gained over time captured in a way an agent can leverage it. So that’s how I perceive it. So I think agent skills can be as good as what humans would be doing or people would be doing in certain areas. There is this area of like I think sales and sales opportunity where like Tommy like

42:57 sales opportunity where like Tommy like if someone’s talking to me about buying some software or doing something they don’t want to talk to an agent or an AI thing at this point of course they still want to talk to a person. They want to they want to have that real conversation the real connection between people that’s that still exists. But a lot of these other things like writing dask DAX building semantic models coming up with system model design consolidating multiple models down to one simple model like there’s a lot of skills we can develop and build that help us create these things and I this is an important note to where

43:27 and I this is an important note to where we are today because very very good points there. Let me let me focus on two things that you said or the first one is about what people actually care about from I think really the developer or the consultant. We’re no longer at the place where I think someone’s hiring a consultant or hiring a developer so much shouldn’t on just their skill alone that oh this person knows DAX really really well. Obviously that’s a part of it. Yeah, that’s part of it.

43:57 Yeah, that’s part of it. Yeah, that’s a part of it. But how do you hire like how how do you even hire like can you like there’s there’s no good DAX test right the only way I can the only way Tommy I can understand if DAX is to say here’s a problem solve it and then I look at your your your DAX output or I or I ask you very nuanced questions about DAX like hey what what is an implicit versus an explicit measure in DAX. It’s not even that. Yeah,

44:27 in DAX. It’s not even that. Yeah, it’s just it’s it’s specific questions to on tease out like the but the only way for me to hire you Tommy this is this is the challenge. If I don’t know what good DAX is and I’m trying to hire someone who does know what good DAX is, how do you have any reference to what the other person knows about what good DAX is? DAX is? Of course. Of course. Well, sorry. Keep going. going. No, no, no, no. You’re you’re good. And I think people hire me right now. I really do believe because they believe I can take them to where they want to go. And my skill is learning faster than most. Whatever your problem is, whatever

44:59 most. Whatever your problem is, whatever your architecture is with the experience that I have is I can learn faster than the average bear. Whatever you want to call it. Yep. And we’re going less to oh, Tommy the consultant is really good at DAX. There’s obviously there’s that’s already an assumption, right? And again, and I think where we’re going with the agentic skills is my ability, my value as a person, as a consultant or as an hire, as an employee comes or should come more

45:29 as an employee comes or should come more from I can take people where they want to go. I look at skills and an essential tool piece of anyone’s work. To be honest, Mike, if you if I were to hire right now and they said, “Yeah, I’m a fabric developer and I love AI.” I say, “What skill?” Like, “Let me see your code base. Let me see what you’re doing.” And they’re not using skills at all. I’m going to be very disappointed. I’m going to say, “I don’t think you’re really using AI.” Yep. Yep. To be honest, I and I don’t want to just see chat GPT in a in a

45:59 want to just see chat GPT in a in a chatbot because we’re at this point now, Mike, at least where I’m at, where skills are now not just a added benefit, a nice to have. You better be using them. You I have to is a strong way of saying it, but you should be. You absolutely should be. Whether it is as a architect, whether it is as a salesperson, especially if you’re in fabric right now, whether you’re using ones that people have

46:29 you’re using ones that people have created or you built your own skills, I don’t think we’re overhyping these either where they are part of just the process now. Would you agree with that? Would you say that? Would you really if you were to hire someone right now or look at your team and no one if someone wasn’t using skills, would you go that’s fine or would you go they should really know what this is? This is this is and I’m hire I’m saying you’re hiring for the best of the best. I don’t think

46:59 for the best of the best. I don’t think you can say this for every single person out there, but if I want to hire the best person out there, they dang well better be using skills because that means I know they’re focusing on other things, too. [snorts] [snorts] I feel like this is a conversation we’ve had multiple times before in the past, Tommy, about other various skills of fabric, right? Yeah. Yeah. One of them being I think one of the main points is if you are a fabric developer and you’re not using notebooks and Python and Spark, like you’re missing you’re missing the boat. You’re missing a really great

47:29 boat. You’re missing a really great opportunity. A lot of fabric runs on top of notebooks and it’s great for data engineering. Can’t get away from it now. Can’t get away from it. You can’t get away from it. No. Well, this is it was so impactful like so a lot of people again let’s let’s talk about the two communities of world here, right? So we’re also trying to blend two worlds of community, right? We’re talking about the PowerBI community world which was like Power Query report modeling and reports, right? We and slowly those items that were just purely PowerBI have

47:59 items that were just purely PowerBI have been stripped away. Data data marts are now gone, right? We don’t have those anymore. all those have been moved over to fabric. So what we’re seeing is we’re seeing this slow move of different toolings. power query dataf flows gen one is is being marked to eventually be done like it’s they’re not adding any new features to it. It’s on not I wouldn’t say life support but it’s like it’s it’s eventually going to be moved over to dataf flows gen 2 exclusively a fabric item. So right [snorts] what we’re seeing now is we’re seeing

48:28 what we’re seeing now is we’re seeing the pinch on if you’re only a PowerBI user only fabric is slowly absorbing all the little tiny features that they were okay maybe a first pass at them into the fabric world and one of the main skills was everyone was like well great I know SQL databases I know SQL I’m comfortable in that space I don’t want to go to notebooks and my argument is stop stop stop stop stop stop stop stop stop stop stop stop stop stop stop stop short shortcutting yourself you’re you’re you’re making excuses for things that are not even real. And I to your

49:00 that are not even real. And I to your point Tommy, this agent space, the skills now in fabric, right? Are you using the fabric MCP server to create items now? Are you using the PowerBI MCP server to then talk to and manipulate data models? these like these two things are incredibly powerful and I’m finding myself as an expert in the space I’m finding myself leaning more and more towards using these agentic tooling solutions on top of existing infrastructure and so I’m finding a much

49:30 infrastructure and so I’m finding a much better experience of hey agent go look at this workspace tell me what’s in it hey agent go read these notebooks tell me what’s going on there summarize it for me hey I found a problem in this I found a bug and I can just basically write up the bug bug and hand it back to the agent and it can then go do something like it’s it’s not just this is when I first saw chat GPT come out on a chat bubble my first reaction was this is powerful but it doesn’t do anything it doesn’t action on anything I wanted to do things I wanted to be very smart and give me

50:00 I wanted to be very smart and give me answers about questions and things but I needed to respond with now let’s give you access to some some tooling and let you go do the change pull down the file make the changes put it back up so Tommy I I’m going to wrap my thought in finalize my thought in users must be able to learn how to use agents. It’s going to be part of your new development pattern. This is a skill that is important enough that you must leverage this. And I’m the new

50:32 must leverage this. And I’m the new world of BI developers, the hiring companies, they are going to ask for if they’re not already, what agentic things do how to do? What agents are you using and how do you use them? And to your point, Tommy, if you say you’re using agents, the same thing that was with,, I use PowerBI, right? Or or I use Tableau Editor 3, like I know external tools. Yeah. You say that and then I need to I actually really need to like, okay, do we have a mutual understanding of what that definition means? When you say you

51:02 that definition means? When you say you that definition means? When you say how to use PowerBI, are we talking know how to use PowerBI, are we talking about how to build a report and add some bookmarks to it? or are we talking about how to build a report and you have good data modeling practices because those are two in my mind separate distinctly different skill sets and so I’ll find someone who says I know PowerBI and you’re just talking about I only know how to build reports but you have zero or little no knowledge around building the model and so that skill of that other part of this in my mind when I define how to PowerBI doesn’t fully wrap the whole picture and same

51:33 fully wrap the whole picture and same thing is happening now for agents right I know how to build things with agents what What does that mean? Are you building skills? Are you using MCP servers? Can you explain it to me? Do you understand some of the concepts so that you you’re you’re in it enough that it makes sense to you? I’m going to throw a wrench in this. I agree with you what you said, but I’m going to take this to the extreme because I think there’s another side of this, Mike, because I agree. I I everything you said I have no notes on. However, when you look at this whole spectrum of using AI for the fabric

52:04 spectrum of using AI for the fabric developer, there is this extreme part. If someone’s listening to us going, “Okay, I’ll do everything in AI.” Mike, I’ve been reading more and more about this issue about AI slop and more and more about the majority of these pull requests, majority of what AI is doing is not best practice and in fact has a lot of lot of fixes and work that needs to be done. And I think fabric’s going to be no exception to that. Let’s make the

52:34 exception to that. Let’s make the assumption. I think you and I can both make the assumption that agents aren’t perfect. perfect. So, let’s take this extreme and say, should someone only rely on skills? Where is the break point? Where is the threshold on I am relying on my own skills as a person, my technical skills, and I am letting agents run wild. Because if I were to hire someone say, “Yeah, 95% of what I do is agents and I

53:04 “Yeah, 95% of what I do is agents and I have workflows and AI and I don’t write DAX anymore.” Are you hiring that person or are you allowing that process to become become part of your workflow in your organization? Where’s the cut off point to how much skills and really utilizing and relying on skills for a person or for a team? Where is the cut off point? There was a there was a very small thought while I was unpacking this together here and we’re

53:34 this together here and we’re talking about agents and skills and things. There was a small thought around when when do you when do you cry uncle? When is too much? Yeah. Yeah. Right. And and so to your point that’s what I’m asking. Yeah. there there’s maybe like two thoughts I’m thinking. First thought is the more you use the agents one, you’re taking your knowledge and pushing them into agents. agents. Good. It’s it I’m thinking about solving different kinds of problems now. Not the

54:04 different kinds of problems now. Not the same problems. I’m solving different problems, which is great. That’s fine. That’s that that’s what I like to do. However, I don’t write code anymore. I’m not ever doing that. I’m writing less and less DAXs as I go. I’m distilling my knowledge into known skills, but I’m doing those other tasks less. It’s going to be a use it or lose it mentality as the agent picks up more of these remedial tasks. Maybe it’s going to continue to push me further and further away from me

54:35 further and further away from me actually doing it, which is I think okay, right? So, let’s one part of this. Okay. The second thought around this one is there’s a cost implication to this, right? I see a lot of tech companies right now letting go of people, letting go of talent and they’re keeping the remaining talent on to do more things and they’re throwing those people at there’s so the individuals that are staying on these companies, they’re leveraging agents to do the work of these other people that have left the

55:06 these other people that have left the companies. More work, different work. But what they’re also acrewing is they’ve removed a salary from their business business and you’re now exchanging part of that salary. Probably not the whole thing, but part of that salary now for some anthropic cost, right? So I’m I’m I’m shifting cost from a person human capital into AI capital now. And so so at some point you’ve reduced too much of the people workforce and you don’t have

55:37 the people workforce and you don’t have enough people to continue orchestrating and building and like learning that new things and you’re you’re doing so much management of the agents that you’re not actually able to like get forward move forward. This is this goes back to my comment I’ve made multiple times now. Uber has burned through their R&D budget in like a little bit less little bit more than a quarter of their whole year. So their whole year of R&D budget has been burned through in like the first quarter of the year because they’re spending so much more money on anthropic and their adoption of AI went from like 30% of engineers to like 68% of

56:07 30% of engineers to like 68% of engineers. it’s like it’s an incredible they found it so valuable and everyone started using it. Now comes the problem of optimizing it. And so Tommy that’s what I’ve done with my Azure resources with the MVP program. [laughter] the there’s so much more you so so you the there’s so much more you so so what is the threshold of what know what is the threshold of what you’re willing to spend on the AI so I think I think this is also a variable cost everything I see coming out now about AI it’s not just I’m running a single large language model all the time I’m running six of them at the same time

56:38 I’m running six of them at the same time I’m now building swarms of agents because they can build more things so not only am I paying let’s just say for example right if I run an AI for an hour it cost me 10 bucks well now if I’ve got six of them running for an hour. I’ve just 6xed my costs just by a couple lines of code. Now I is the output that I’m getting six times better because I have a whole swarm of things. Is it doing things that much faster? I think this is where the evaluation of leaders have to be like is this lemon worth a squeeze? And is there a threshold to say

57:09 squeeze? And is there a threshold to say we can’t continue spending so much more money on on AI things and we need to start looking back and going optimizing. Let me say this, Tommy. There’s a whole new world of AI optimized people optimizing for AI. So So that there’s a there’s a whole new skill coming. We’ve just shifted I’m a bunch of developers who write code to now I’m a bunch of developers who need to optimize how the agent runs efficiently. Yeah. Yeah. That work is just shifting to another

57:39 That work is just shifting to another team. There’s a concern I had with what you said too because you’re talking about this in the realm of the developer and if you were to basically plug and play what you said for a fabric team an organization right I think a lot of that doesn’t jive because you are making a lot of assumptions that the organization has said okay fabric team BI team go crazy with AI as much as you want to do right that’s I don’t see the story right now

58:09 that’s I don’t see the story right now I’m not saying that’s the story, but I’m saying based on what you said, that would have to be the assumption, right? So, So, no, I don’t think so. I think it has So, I don’t think that’s a fair assumption, but keep going. Okay. Yeah. So, there’s a problem though with as much as you would want to do with skills. And this is I think where we may be overhyping skills a bit or the reliance on this is I can’t if I’m in internally only utilize to your point you’re not using you’re not building anymore you’re

58:40 using you’re not building anymore you’re not writing anymore you’re allowing agentic things to do most of the things in fabric. Sure. Sure. Okay great well is someone at an organization a large organization also has an ability and if they do well talk about skills that you have to pass on. You talk about the knowledge that you have to pass on, right? I can’t just build in my own siloed box fabric lake houses and everything agentic if I’m the only one doing that. This is a problem that we’re going to constantly I think reach up on

59:11 going to constantly I think reach up on where me as an individual, if I’m building with other people or fabrics part of my organization, I can’t be the only one using skills because what happens when something breaks or changes, right? Do I have to tell them, oh, go to my GitHub repo and find the skills because that’s the only thing I’ve done. I don’t know what else happened. Right? Do you understand the governance issue here too on an individual? You and I can do this internally, but an individual at a company cannot rely on skills the same

59:41 company cannot rely on skills the same way you and I can. I also disagree with that statement as well. well. Go ahead. Let me hear this. So I don’t see right now there’s there’s what should happen and what is actually happening. What should happen is leadership should say skills things are important enough for us to learn and we’re going to have to push it out to our people. It have to it’ll have to come typically top down approaches work well. What I think in reality what happens is

60:13 What I think in reality what happens is someone at the middle to bottom level of like the organizational structure finds these things and becomes extremely productive. Let me give you let me give you a story. My brother-in-law works for a company. One of his team members in a region is in sales and he has figured out how to use chat GPT to increase the sales pitch. And he says, “Write me 10

60:36 pitch. And he says, “Write me 10 pitches.” Writes them down. Okay. Uses said 10 pitches. Which ones performed well? Okay. Feed that back to the agent. Here, these three did really well. Do it again. These two did well. Do it again. Do it again. He’s now one of the largest sales representatives of the entire company in that region because he’s using at he can at scale go faster than everyone else. Right. So awesome though. The reason leadership is not encouraging this. He’s just doing this on his own.

61:06 this. He’s just doing this on his own. He found something. He figured it out and he’s using it in his process. So what’s going to it’s going to be the same thing. This is the exact same thing that happened with PowerBI. PowerBI came in as a free tool. You downloaded desktop. There was people that were like I’m interested in this. And you’re going to have some people in your organization in the same way with PowerBI. And again, I think it’s the same pattern. Mike’s and Tommy’s the world said this is a gamecher. I need to learn this. This is exciting to me. I want to do this. We spent time. We made effort. We’ve

61:36 spent time. We made effort. We’ve learned it. We then researched it and we let it consume our thoughts for most of the time. Therefore, we got really good at it and here we are now running our own businesses on top of the PowerBI thing. This is the same pattern that’s going to happen for AI. It’s going to start a bit more in the grassroots area. People are going to figure out how to do things more efficiently. The large corporations, the large big companies are going to slowly steer the ship in the right direction, but it’s going to be these fast, nimble learners and people in the organization that are really going to set the direction of

62:06 really going to set the direction of this. Now, this is going to be the same thing that happened with PowerBI. In some cases, the company said, “Tommy, you’ve done a great job with this PowerBI thing. Let’s keep going with it. Let’s keep push. Let’s put fuel on this fire. Can I give you more employees? What does this look like? How do we grow it? This seems like it’s adding value.” And they accelerate by investing time and materials in what you’re doing time with PowerBI. This is the same thing with AI. with AI. It’s it’s it’s different though because the their ability to explain it, right? So let’s take your example here with

62:36 So let’s take your example here with PowerBI and the the the adoption the fire that PowerBI took at organizations and if I was the one going bottom up well I was explaining it to the team like dude there’s this thing called power query you can connect to this look at data we need to understand this and you’re beginning to gather in people from the re from relevant departments and whatever resources that you need to start building semantic models the right way. So now take this whole example with AI and the output may be the same. Maybe it’s a report but you can’t explain it

63:09 it’s a report but you can’t explain it right because you are only relying on skills. Nope. And you’re saying no well I just have a bunch of skills that run this fabric pipeline and look at all the data is good. good. Which means in order for that to get the adoption everyone else at the organization or who’s going to build has to use the same skills because you want to replicate that. That’s the only way. There’s no other way we’re going to say okay now you do it by as a human I’m going to use my skills you use the mix I feel like or my fear is we’re going to

63:40 feel like or my fear is we’re going to have this all or nothing approach if I am using skills to build the pipeline to build the the workflow the architecture I can’t then tell someone in another department or to do the same replicate my same work using the human work they’re going to have to use the agentic process process And to me that’s an issue. I don’t see it as an issue. Why Why I don’t see as an issue. I don’t see it

64:10 I don’t see as an issue. I don’t see it as an issue. I think the skills are not that’s not localizing specific things into people’s brains. It’s it’s skills skills are a plugin. Skills are easily readable. We can ask the agent what does this skill do? There there’s there’s no more and more black box. Correct. But I’m just saying so let me go back to your analogy right in the PowerBI education that you were doing to your team. Hey, look at this. I have Power Query. Hey, look at this. I’m I’m doing these things. Hey, look at this.

64:40 doing these things. Hey, look at this. Why are you doing any of that? Why didn’t you just keep doing what you were doing anyways? Well, it’s because PowerBI PowerBI gave you two things. It gave you automation. It provided efficiency. and it help you build things faster. Those three things, I see where you’re going with this. That the only reason you introduced PowerBI was for the if those three things didn’t tick the boxes there. More efficiency, automation, and building things faster, getting insights quicker, right? If that didn’t happen, you would not have spent any time in PowerBI.

65:10 not have spent any time in PowerBI. Period. End of story. You can’t argue with that one. This is this is the same thing that agents are doing, okay? okay? Agent experiences on top of fabric. What they’re doing is agents are we teach them in the same way we taught Power Query and PowerBI because agents can do things faster. It can do things automatically with automation. So basically I can give it a less I can give it less instructions. I’m not clicking buttons. I’m not going through UIs. I just tell it what I want.

65:40 through UIs. I just tell it what I want. And to your point, Tommy, we’re also able now to talk directly to it through our microphones. like we’re we’re saying the words of like the the the the amount of we’re we’ve changed the programming language. We’re no longer programming in CN. NET CN. NET SQL and other things and DAX. The programming language is now human thought English language. The language you speak in that’s the programming language. And so the reason we’re going to teach people to adopt this is because agents will make things more efficient. It will automate more of

66:12 more efficient. It will automate more of what we do. and I’ll be able to build faster. If you hold on to those three principles, everything else you’re describing is teachable. What skills are out there? How do I use them? All of those other areas that I think you’re concerned around, those are just educ to me, they’re just education bumps on the road. road. That is such a good point. I actually put that as a topic because I think you’re really that’s I think the crux of what I’m saying is my ability to pass the the gift of PowerBI to others was dashboard in a day, right? Yes. I could

66:43 dashboard in a day, right? Yes. I could skill up others agents in a day. So skills on Saturday thing like skills in a day like but something catchy. Yeah. This is the point you have to do that. Correct. And so I think this is the point I’m trying to make here is like the efficiency gain that we see right now with these skills inside fabric. And then this is why I think one of the reasons why my fabric precon with Matias went so well. We had 185 people in the session that signed up. People still wanting to get in that couldn’t get in because we didn’t have enough seats in the place. The whole room was packed.

67:14 the place. The whole room was packed. Every seat was packed and the back row was packed against the wall. So like we had a lot of people in there trying to people without tables still trying to follow along because people realize if I learn this skill I can go faster. I can automate more and I can I can build more stuff quickly. That’s that’s the thing that I’m seeing here. And to me this is this is a brand new way of thinking about how we build things. And over and over again I’m observing many many engineers saying but I need to see the code but I need to change I need to see what’s doing like I don’t think I can.

67:45 what’s doing like I don’t think I can. The the challenge is I don’t know how to direct the AI to build what I want because I don’t know what code it’s building and it I don’t know if it’s it’s right or not. And my comment to that is you’re thinking about AI wrong. You’re thinking about the problem incorrectly. If you are feeling you need to go dive into the code in order to understand what you need to do to prompt the AI.

68:15 need to do to prompt the AI. It just means all that is pointing to is you don’t know how to prompt AIs. You haven’t you haven’t spent enough time describing your requirements. No, it’s seriously you haven’t done enough. You haven’t done enough time deciding like how to write the requirements. And this is why I think Okay, we’re going to go full circle here. Yeah. Yeah. You’re not getting the output you desire because your input is wrong or your skills are wrong or your guardrails are wrong on the agent. The agent is nondeterministic. It’s the same thing I do to you, Tommy, or any other employee

68:46 do to you, Tommy, or any other employee that I have at my company. Like when I communicate to you, I give you like my level of thinking and I need to give you like the guardrails of my mind. Here’s what I’m thinking for this thing. Here’s how. So, I have to frame out a lot of extra words to get you aligned with the direction of where I’m going. And this is why we’ve talked about too, Mike, where you should be putting on your resume or this should be part of a resume saying that you have a skill around writing skills or being able to prompt. Prompt engineering is a technical skill. Absolutely. Well, we’re at time. Mike, I put in our speaking of

69:16 at time. Mike, I put in our speaking of agents, I put in our backlog a topic that I think would be great is talking about the individual agent skills versus working on a team. I think that deserves its own recognition on how do you organize and adopt around a team. But hey, Mike, this was an awesome I love talking about this. I think my closing thoughts here is we really are at a point now regardless of the reservations I may have said. Yeah. Yeah. Where you have to be using skills. I

69:47 Where you have to be using skills. I don’t see a way around it. I personally love it, but I don’t know what I would do without it now. And if you consider yourself a serious fabric developer, you should at least play around it. I think in 2026 in May, you should at least have some that you played around with. Yeah, I’m I’m going to I’m going to wrap on that one as well, Tommy. Let my final summary here is where am I finding the most value in what I’m using from Fabric and these new agent worlds?

70:17 from Fabric and these new agent worlds? The Fabric PowerBI MCP modeling server is powerful. I’m finding value from that. I’m also finding finding more value recently around the fabric MCP server. These are two things where there are tools and specific skills you can sign in and do things and create things and move around. I’m using the UI of fabric less and less. I’m using the UI of desktop less and less and instead I’m spending more more and more time on communicating with an agent that is understanding the intent and design of what I’m looking

70:47 intent and design of what I’m looking for and then I’m asking to go execute it in various files and forms and and things there. So all I’m going to say is this is a skill that everyone should learn and how we do it yet. We’re not quite there. Everyone has their aha moment. Tommy, you had you had yours maybe earlier than I did. My aha moment was like December and January time frame. For whatever reason, something for me clicked and I was like this is the only thing that I need to focus on. This is this is the evolution of

71:17 This is this is the evolution of I feel like seeing agents is the same feeling I’m having when I saw PowerBI for the first time. This is a gamecher and this is going to greatly change how businesses work. Agents can do a lot of things, but agents in con in concert with fabric and what we’re doing data engineering wise, this is going to rapidly change how you build stuff in the future. 100%. 100%. Okay, that being said, thank you so much for listening to the podcast. We appreciate all your ears for this hour and 10 minutes or so. So, we thank you for jumping in and listening to us on the show. Tommy, where else can you find

71:47 the show. Tommy, where else can you find the podcast? You can find us on Apple, Spotify, wherever you get your podcast. Make sure to subscribe and leave a rating. It helps us out a ton. Do you have a question, idea, or topic that you want us to talk about? on a future episode. Maybe more around skills, maybe more the other side of the coin. Well, head over to powerbi. tips/mpodcast. Leave your name and a great question. And finally, join us live every Tuesday and Thursday, a. m. Central, and join the conversation all powerbi. tips social media channels. Thank you all so much. Make sure you like and hit a thumbs up on this if you enjoy this

72:17 thumbs up on this if you enjoy this episode. We really appreciate it. It helps get it out to more people. Thank you all so much and we’ll see you next time. Explicit measures. [music] Pump it up. Be it high. Tommy and Mike lighting up the sky. Dance to the day to laugh in the mix. Fabric and A. I get your fix. [music] Explicit measures. Drop the beat now. Kings feel the crowd. Explicit [music] measures.

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.