Agents Helping with Data Governance – Ep.545

A mailbag question asks where agents fit into data governance. Both hosts answer carefully, because the honest version disappoints anyone hoping to automate the hard part. Governance is ownership, accountability, and people agreeing on definitions. What agents can do is remove the tedium that stops anyone from starting.

News & Announcements

• OneLake architectural guidance: a practical blueprint for the AI era — Microsoft’s guidance on building OneLake, laid out as patterns. One covers external data sharing — shortcutting Snowflake or Databricks tables into OneLake and building semantic models downstream with a single governed access point. Tommy’s only complaint is length: at 12 pages it’s a strong start but high level, and he remembers 145-page Power BI enterprise white papers.

• Fabric Task Flow Studio — Tommy’s own tool, built on Alex Powers’ Fabric task flows, for standing up demo environments fast: describe the pattern, get the artifacts, and import the task flow so the UI comes with it. He built it with Claude Code’s help and has been using it to spin up prospect demos in minutes.

• Chicagoland Power BI user group — July 16 — Live Rayfin app build, with discussion of which agents are involved and how the pieces fit, plus networking and Q&A.

Main Discussion

Topic: Defining data governance, then finding the narrow place agents genuinely help

Mike defines governance through his engineering years at Johnson Controls: a car battery carries marketing label data, physical dimensions, materials, and compliance standards, each owned by a different team that all have to agree. That’s master data management, and it looks nothing like a software feature.

• Agents have a small role — and it’s essential. Tommy is emphatic in both directions. Agents will not lead your governance program. They are very good at the mundane transparency work underneath it.

• Discoverability is the real win. What semantic models exist? What’s in the catalog? Which of these seven sales reports should someone actually use? An agent with an MCP connection can make that human-readable, which is exactly the work nobody has time for.

• Lineage inside a model is unreasonably hard to extract. Mike keeps returning to this: what does this measure do, which measures does it depend on, which tables does it touch? The information exists in the model and getting it out means real research time.

• Governance already exists — you’re adding Fabric to it. Mike reframes the usual question. Access control and business objects predate your Fabric tenant. The task is integrating Fabric into what’s there, unless the honest answer is you have no governance at all.

• Accountability isn’t only punitive. Tommy pushes back on the assumption that accountability means someone gets blamed. It also means knowing the pipelines work, the data flows correctly, and people can trust it.

• Data quality rules need a home. Does this column only contain integers? Did someone push a string into it? Where does that rule live, which notebook tests it, and if new data fails — do you flag it, quarantine it, delete it, and who gets notified? Mike notes OneLake doesn’t really support this today.

• Store your tenant settings in OneLake daily. A concrete, cheap practice: snapshot the settings so you can see what changed and when.

• Nobody ships the silver bullet. Microsoft will give you wrenches and sockets, not the finished solution, because governance maturity differs at every company. Mike points to FUAM — the Fabric unified admin monitoring solution — as an existing option that already pulls much of this together.

A 30-day governance MVP

Tommy’s opening move needs no agents at all: interview and investigate. Get a lay of the land on data trust and definitions — every measure, every report, the audit of what exists. Seven reports going to sales with sporadic views is a finding. Where is sales actually supposed to look? Ambiguous definitions and usage patterns are the two things he’d have agents surface in the first 30 days, while the discoverability work itself stays human.

Looking Forward

Audit one domain’s reports and measures for duplicate or ambiguous definitions this month — no agents required to start, and it tells you whether you have a governance problem or a tooling one.

Episode Transcript

0:01 measures. 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. Explicit measures. Drop the beat. Now, kings feel the crowd. Explicit measures. Hello and welcome back to the Explicit Measures podcast with Tommy and Mike. Hello everyone and welcome back. Hey Tommy, how’s it going? Oh, Mike, it’s going well. How are you

0:31 Oh, Mike, it’s going well. How are you doing? doing? I’m doing great., just trying to stay cool., the heat wave is coming through our Wisconsin area and now in Chicago as well. Should be pretty toasty today., but before we get into the news and all the other items are going on right now, let’s talk about the main topic today. I think our main topics around where does agents fit in helping with data governance? Do agents can agents help us out with data governance? What does this look like? what are patterns that we’re seeing evolve here?, and

1:01 that we’re seeing evolve here?, and how would we leverage agents to assist with this? State of governments is a quite a I think it’s a it’s a long path. It’s never really done and there’s a lot of work that needs to be continually updated. It’s it just doesn’t stop. You don’t just do it and I’ve completed it. Data governance is over and you move on. It’s this continual growing process I feel like in your organization. So, in le of that topic, we’re going to jump in and do some news. Tommy, what news did you find for us today? Oh, we got a few things. So, let’s start with the announcement, and I think this

1:31 with the announcement, and I think this is probably the coolest place to start. We are back with our Chicago and PowerBI Microsoft Fabric user group. This Thursday, there’s a few announcements here. The first one is this Thursday downtown office on Drive in Chicago. We are going to be talking about Fabric Apps in Rafen. And Mike, I believe you’re going to be the one presenting on Thursday, right? Yes, correct. So, if you want to come if you want to come join us there, please make sure

2:02 come join us there, please make sure you come join us. We’re going to do a build of a Rayfin app live. We’re going to do one of those. We’re going to kind to do one of those. We’re going to talk about concepts and how to of talk about concepts and how to leverage and build those apps directly. So, if you’re interested in the Fabric X apps ecosystem and you want to get some real-time viewing, someone building an app, how does it work? What kind an app, how does it work? What agents are we using? we’ll talk of agents are we using? we’ll talk about those topics there at the at the the user group. So, really excited. Again, we’re going to do the networking. You can ask questions to Mike or or me, but

2:32 questions to Mike or or me, but mostly Mike because he’s the one building., but the cool thing about this, Mike, too, is we have talked about the user group before, and I know we’ve been very sporadic with it, but we have our all of our user groups scheduled for August and September as well. So, we are going to go back to our monthly cadence. So, really hope you can join us if you’re in the area. We’ll have a lot of fun., so really excited about that. So, yeah, Mike, I really can’t wait to hear what you want what you’re going to do or see what you’re going to do, hear what you got to say, go through all

3:03 hear what you got to say, go through all those things. So, I’m hoping it’s a little bit like a choose your own adventure. We we’ll kind choose your own adventure. We we’ll let the audience pick some of of let the audience pick some of the the items we want to build or create while we’re there building the Rafen app, but just trying to show people it’s not intimidating. It’s it’s easy to get going and get started and it’s not too difficult to get started with working with Rafen, which I I’m really enjoying these days. Are So, you’re saying this is going to be a little like whose line is it anyways? Like, hey, I need I need a location. Who can give me a location? Yeah. Right. Exactly. Exactly. A little improv theater. you are in the

3:33 A little improv theater. you are in the second city, you’ll be in the second city. So, it’ll be perfect. So, really hope people can join us for that. Make sure if you’re hearing this on Tuesday, July 14th, you need to register soon because we need to send the list over to Microsoft so you can actually get into the building. so, you’ll need to bring your ID for that. So, really excited about that. So, that’s the first one. Mike, the next one is going to be another news article. And then I do have a really cool beat from the street. Yeah, Yeah, this is going to be a topic that you and

4:03 this is going to be a topic that you and I are absolutely probably going to dive into in more than one episode, but I want to give the announcement here. Microsoft just released the one lake architectural architectural guidance. And what this is is really Microsoft’s guidance around how to build one lake and what one lake looks like in the era of AI in the world that we live in now. Mike, I was just listening to one of our previous episodes where we talked about build for semantic models or build for agents since what anthropic has come out

4:34 agents since what anthropic has come out with what we’ve seen well from OSI in terms of how do you build a semantic model this is really great because it’s a little higher level when you think about one lake when you think about the different deployment ways so they really talk about three things in the guidance white paper strategic initi initiatives architectural patterns and platform capabilities. Mike, I just saw it this morning and the more I was reading through it, the more I realized this has to be a future episode.

5:04 this has to be a future episode. However, just I don’t know if you’ve been able to see the white paper at all or at least see the news article here. I’m so happy to see this because we have talked about we haven’t seen white papers yet around fabric and this is I think a really important step for Microsoft to say this is our best practice approach. this is our recommended approach. Yeah, you’re seeing some interesting things pop up on this new architecture. some of these architecture pieces are elements that I really resonate with. and I think also depending on where

5:35 and I think also depending on where you’re at organizationally, like how mature you are as an organization, different patterns that you’re going to see here presented are going to make more sense for you than others. Right? One of the examples would be is they have pattern number two, building a medallion architecture,, in an operating mode. they’re talking about like real-time analytics fact you like real-time analytics fact data factory analytics coming know data factory analytics coming bringing in data to bronze silver and gold I think this really works very well for organizations who don’t have a data bricks who don’t have a snowflake already existing haven’t spent the time or investment to to leverage those

6:06 or investment to to leverage those systems then there are other patterns they’re talk about here which they’re talking about using external data sharing getting data directly from snowflake or data bricks there are a lot of organizations that we work with that have either snowflake or data bricks and they’re leveraging the lakehouse to like shortcut or virtualize those tables into one lake and then build downstream artifacts of semantic models and all the governance and a single access point for all the data which I think also makes a lot of sense. The one that I’m most recently resonating is pattern number four which

6:38 resonating is pattern number four which is talking about platform simplifications. So it’s talking about apps and supporting apps that actually use you Microsoft Foundry, Microsoft M365, data agents and PowerBI all stitched together. The back end of all that runs on top of fabric and one lake, but the application is the the service area where you’re producing the information. So we build a ton of applications. This is I think a huge win for a lot of organizations. It simplifies a lot of things and I think they have to start talking about

7:09 they have to start talking about application building because Rayfen’s now here. that’s really an application that’s now built into fabric. So they have to have more of that story there. So I really like this article. I’m the the author is Ofer. Ofer is a personal friend of mine from Microsoft. Really smart gentleman. Has been working a ton of years at Microsoft and building out really trusted onele stories around what’s being built there. So, great article. He’s he’s really experienced in this space and we really want to push

7:39 this space and we really want to push this article and this white paper because I think this is really relevant. Again, a lot of this Tommy we’ve talked about in the past is how do you measure up? Like what you’re doing something as a company, where do you fit against other companies? What’s what is your what’s the measuring stick that we’re using here? And I think having white papers like this allows you to go through your organization and say and figure out where do we sit on the scale of how mature are we in in a one link governance plan strategy whatever that

8:11 governance plan strategy whatever that looks like. And I think that’s really useful here in my opinion. Yeah. And the one thing I have to admit I was a little disappointed because it’s only 12 pages and some of you may be thinking 12 pages is a lot. Well, if you’re old like me and Mike, the initial PowerBI white papers around EnterpriseBI and when Premium came out, they were like 145 pages and they were dense. So, this is I think this is a great start. It’s it is very high level., it it doesn’t go so much into roles,

8:42 , it it doesn’t go so much into roles, responsibilities, the different assignments that you do to people, but I think it’s a great start because we need to start seeing this from the source from Microsoft. This is I think where you and I come in a lot where this is our lives. This is what we live for. But I really think this is an important step for companies to your point just to see hey where what’s our index? Where do we where do we land here? So really yeah I think I think this is a message that’s also very clear for me looking at what Microsoft’s producing in the space.

9:13 Microsoft’s producing in the space. Microsoft’s clearly saying one lake is where they’re making a large investment and they’re going to continue to make one lake this better story here and I think we’re talking about a topic today around data governance. Data governance is a very important topic. There’s lots of things to unpack of like what does that mean? How do we understand it? How do we share data that has been governed? I think this there is a data governance icon on this chart that’s talking about one lake co-pilot and governance. It’s there. It’s listed

9:44 and governance. It’s there. It’s listed on on the diagram here. So I I really do feel that we’re in a place right here where where the one link is getting more love, more attention. and I think if Microsoft is smart here, they’re going to make it as easy as possible for you to bring your data to one lake. Yeah. Right. Cuz once you bring it there, everything else downstream just works and bolts on. So I think strategically the more they can make one lake easy to use, quick to start with, get going

10:16 use, quick to start with, get going right away, I think the better way you’ll have longterm for fabric and everything else that you’re building. It’s funny, Mike, because I was just talking to a team last week and they were saying, “Well, what’s really the selling point of fabric?” And it’s really one like if I were to do my ranking right now of all the features in fabric and in terms of the importance number one will be semantic model but number two is probably one like and in terms of the capabilities the back end the ease of use and just the capabilities there. So really happy to see what they’re doing here. All right I

10:47 see what they’re doing here. All right I got one more thing. This is actually a beat from the street and Mike this is actually something you Mike and Alex Powers talked about a couple months ago. Alex Powers, who you may know as nothing but the cell, came on and talked about his new feature called fabric task flows, which is an awesome feature to help build out almost basically the architecture and the artifacts just from a simple prompt. it’s over it’s under the Microsoft handle in GitHub and it’s a really amazing feature. All you

11:18 it’s a really amazing feature. All you have to do is say I need to build I have 30 stores and I’m trying to get all the data in. we want to build a report, whatever the case may be. Or maybe we’re doing real time and I need telemet telemetrics around it. Well, I’ve been using it and I I really enjoyed it, but I’m a user. I like user interfaces as much as I use cloud code. I always enjoy especially with the amount of steps that I thought the task flows required went through and we I built out what’s called fabric task flow studio. This is a forked repository of Alex’s task

11:51 forked repository of Alex’s task flows and it really is an entire user interface that goes through the seven different phases that Alex built out. So you can actually go back and edit, you provide the prompt, you can save projects and go back to them and it allows you to say, what, I actually want to view all the things that were created in the design phase or let me retest the design phase. I actually set it up too. So you can use GitHub copilot or cloud code. you can choose which handle you want to do. And I I’ve been actually using this to

12:22 And I I’ve been actually using this to just do help build out a lot of demos. has been the coolest thing is I have I have a client or a prospect that we’re talking about things like what would be easy. Let me build this out pretty quickly. And I actually built it out too to actually import the task flow itself. So you can actually get the user interface as well. Mike, we’re getting to this point now where one obviously this was not just built by me handling npms myself but very much with cloud code help building it but you realize now this ease of use in building

12:54 realize now this ease of use in building fabric and that’s really what I want to bring up to you we can build as quickly as we want in fabric using AI and what Alex has been able to do what I’ve been saying Since this I saw the light the writing was on the wall Tommy for me back in like January, February of this year was like this is this is going to change things. This is substantially changing what I think I need to be building and how I need to be building stuff, right? I totally agree with that statement. It’s it’s easier than ever

13:25 It’s it’s easier than ever to describe to some agent thing. Here’s what I’m trying to accomplish. I want to get this done. Yeah. Keep going, Tommy. No. So, what I was going to say like what’s cool is you can build those things out, but I think this really leans on being successful with this is the strategic side is you have to think strategic. And there’s a difference between doing strategy and doing tasks. Mike, I had all this in transcripts from again the calls I was having. I have cloud code. I built me out a skill

13:55 cloud code. I built me out a skill that’s actually a task flow prompt builder. So basically hey what’s the best way to work with task flows or fabric task flows and it knows basically the architecture that I wanted and Mike the technical side we know obviously what AI can do but that does not mean that our jobs are easy I I would say I think this is the big misconception because you have to think about this on a long-term point of view. So what’s your take on that?

14:25 your take on that? Well, one, I’m looking at your tool and going, “This is amazing.”, I also saw some of the conversation going back and forth between you and Alex a little bit here. And Alex is like, “Man, it’d be my vision for this thing was to make it like a Google homepage where you just have like a simple input box and just describe what you want and boom, out pops some stuff from Fabric.” So, I’m going to need to give this a bit of a run for its money here, Tommy, a little bit., because,, I do tons of demos. I’m always making up sample data sets and sample workspaces and all kinds of interesting things. And how fun would it be to not only have an agent build

14:58 it be to not only have an agent build something with Rayfin, but actually show and highlight maybe even this week we’ll see if I can get something together quick enough. But you something together quick enough. But I always have a need for different know I always have a need for different types of models. Hey, I need a a lakehouse with a semantic model that’s direct to some sample data. Just make up some stuff. Just make up a three dimensions and a factual table. Put it all together for me. Like we always use adventure works all the time or Ktoso which sometimes it works sometimes it doesn’t but sometimes you want the data to have like a bump in it

15:28 want the data to have like a bump in it or something interesting or something fun. Well now you can describe that to the agent. It can then try to interpret or build out something that’s interesting on a particular product inside the data set. So so you can be much more creative there but then you can describe all the things that you want. Okay, lakehouse semantic model with rowle security deployed here this way d and then you get to the point of okay great now that you have all these things available to you you can then go back and now build your rafin example or your powerb reports on top of this so

16:00 your powerb reports on top of this so it takes a lot of those ideas and and this is what I think is really important here I was just I literally screenshotted an image the other day on I think it was on LinkedIn or or or Twitter or something that had it I I had seen it before and I have not been able to like articulate this into words or I saw it and it was I needed to just hold on to it. It’s this image of before AI and then after AI and to your point Tommy before AI it has this the

16:31 Tommy before AI it has this the people that had ideas was very few the people that needed to execute the ideas was larger but not enormous and then the usage of the idea was was massive right so I think of the idea was online web reporting right so then Tableau comes out the idea was handled by a couple individuals They had a team behind it to execute it and then boom, software is done. You go buy it, right? Same thing for PowerBI, right? We start with this idea, you have a team behind it to build it and then you distribute to millions of people.

17:02 you distribute to millions of people. Well, after AI, the whole design process has flipped, right? The the idea side is huge because now everyone has an idea. Everyone in their brothers, oh, I got this idea. Let’s build this. This is all new. So everyone’s crowding into the ideas space and there’s probably even less people to execute the idea and I feel this is true for me because even in my own workflow I had like three or four ideas before AI and now I have like a hundred of them and I have more and more git repos popping up full

17:34 more and more git repos popping up full projects prototypes here and Tommy this is one of these ideas this is a great idea and you’re taking someone else’s idea you’re refining it and then thankfully you’re able to execute on it Right. So,, all right. We need we need more agents to help us execute these ideas into like something that we can go give out. And I me personally, I’m loving GitHub. GitHub is like my new social media platform. I’m favoring and starring. There’s tons of cool stuff out there., and I feel like that’s the way of seeing people who

18:04 like that’s the way of seeing people who are actually executing on projects. But the downside of all this is because of the proliferation of the ideas phase, we now have a problem of everyone has a thing, right? The number of projects I’ve seen around whisper AI. Well, everyone’s got like a local cloud version or local version running. Which one’s better? Which one do that the other one the other one doesn’t? Do I have to try out five of them to figure out which one’s good or not? Like which one bubbles to the top now? Now it

18:35 which one bubbles to the top now? Now it becomes like a game of, hey, there’s now six different free tools. Which one should I use? Which one works for me? for me? And so this is another problem I think we’re we’re running into is there’s not enough people to actually use all the tools at this point. 100%. And I think also too, some of those some of the ideas are great, but they can be ambiguous with Whisper, right? You don’t have like a feature spec like this. Whisper does one, you spec like this. Whisper does one,,, feature one, feature two know,, feature one, feature two doesn’t do feature three. And I think to your point too, I sent you a video the

19:07 your point too, I sent you a video the other day,, just about where companies are going with token usage. We can talk about that. We’ll put that in the parking lot, but there’s a lot of things we’re running into. I think in the meantime, this goes back to the strategic side, not just build everything or go any idea that you have, but I have been trying to be much more targeted around what I am building and what I’m using AI for because it’s just so essential or you can spend a lot of time doing nothing. But I feel like also in both this regard though, Tommy, where Tommy and I are building and this is just an observation between us as we

19:38 is just an observation between us as we talk and what we’ve been building together now. I see what you build. You see what I’m building. We’re not using AI to tell us answers about our data. This is Yeah, we’re we’re not it’s not leadership. Get this out of your head that you’re just going to throw a bunch of data at an AI and get something useful out of it. Like it does not work. We’re not finding value in it. That’s not where the value mark is right now. Now, maybe in the future someone comes up with something amazing and it becomes more useful, becomes more

20:08 becomes more GPUs. Yeah. So I don’t know. I don’t know what’s going to happen in the future. I can’t tolerate. But right now, the way I look at the landscape, that’s not the value ad. The value ad is saying in your workflow, when people step into something that is un not understood, right? I have a semantic model. I don’t understand it. Use the AI to help you get information, learn, educate. And this is where I think right now a key moment is the AI can understand code

20:39 moment is the AI can understand code much better than we can. It can summarize a lot more information better than we can. And so using the AI to help you create experiences that users need, that’s the key. That’s where I’m finding value. And this is what I think is a prime example, Tommy, of what you did, right? I want to use Alex’s tool to rebuild a fabric environment. Mhm. This pattern, this pattern,, lakehouse, direct lake, semantic model,

21:09 lakehouse, direct lake, semantic model, role of security, report. Boom. Describe it. Boom. Get it. The whole answer is there. Now, it’s all, it’s not all right, but the example, the architecture is there in place. You can start wiring in your own data and then move it from there. So, this creator agent space, I got to be honest, Tommy, I was working in a notebook the other day and it was airing out in the middle of the notebook and I was like, dog gone it. I don’t want to I don’t want I wasn’t I was on my phone trying to run it. So that one that was like not ideal to begin with, right? right? But there’s not an easy way of giving

21:42 But there’s not an easy way of giving like hey let me drop in claude to go pull the notebook and go get the error from it and help me work on it. So if I was on my desktop I would have gone to like VS Code. I would have like synchronized the notebook, pulled it down and then had the AI work like whatever AI I want to work on the notebook with me. And I also find that better like the co-pilot AIS just don’t cut it for me because they’re whatever model they’re using, it may be smart, it may not be smart. I don’t know what it’s doing. I I prefer to know this this problem is

22:13 I I prefer to know this this problem is hard enough that I need,, Opus 46 or I’m going to use GPT55 or I’m going to use DeepSeek 40 Pro. I want to pick what I want to use on that particular issue because I have some level of experience around what model fits the right scenario. Yeah. and let me just quickly add to that and then we’ll get to the main topic. What’s neat about this too and the I think there’s a big part not just the build but a little to your point to the refinement side of things. Like you

22:44 the refinement side of things. Like you find something that works to an extent but then what else can you do with it? And it’s interesting with the Task Flow Studio in particular, I set it up so there’s a chat there. It’s like a chat like feature and I was curious because I deployed everything that I wanted, but to your point it’s all blank. There’s no actual data in it. It’s just the artifacts for it. But I’m like, well, I am connected to cloud code. I am connected to the fabric MCP. I am connected to,, all the fabric agent that Alex created. So I

23:15 the fabric agent that Alex created. So I got curious. Yes. So everything was deployed and in that project because it say those projects I was like hey I’m working in industry X I’m not going to give the industry away here. I said based on this architecture can you just put dummy data in there based on this industry and I said let me think about this. Yes I can. Next thing I know I go to powerbi. com and fabric

23:37 I know I go to powerbi. com and fabric and I see the lakehouse has all this data in it and like are you kidding me? Like are you kidding? So, we’re at this point now where where and let me let me add one note here. I guess Tommy, I’m guessing part of the reason where your agent pulled up some of the stuff from was you had a conversation with the customer and they talked about some of the things. Yeah. Yeah. About like this is the data we deal with and like the context I’m going to make up some stuff here like hey I’m in the the clothing industry and I have a warehouse and I have a whole bunch of things sitting on

24:07 have a whole bunch of things sitting on racks and this is the struggle I have and I have inventory coming in. I have sales coming in. I have to go find like they can describe in general terms and you can go back to the AI agent and say here’s the transcript of this meeting. Go scrape through this transcript andor summary and figure out what data would be most relevant to them and then mock up something that might be similar. Again, you’re just trying to get close so that the customer can understand like where you’re coming from, right? That is such a good point here too and that goes back to the harness idea. So yeah, Mike, we’re living in a

24:37 idea. So yeah, Mike, we’re living in a great age and I think yeah, we’ll we’ll pause there because I think that could be another topic, but Mike, a whole another topic. We can get there, whole another topic. So, but let’s get to the main topic. Data agents in data agents in governance. All right, so maybe we should just quickly define data governance first. Well, actually, just kidding. This is probably a question that actually came through as a mailbag. So, let me let me let you read the let’s read the question first, Tommy. What I’d like to do is I’d like to read the question. Let’s define data governance because I think there’s a lot of

25:07 a lot of opinions of what that means and maybe you give your definition, Tommy, I’ll give my definition and then we’ll talk about Okay. Where do we see agents fitting and helping with this? Perfect. All right. Here we go. Hi guys. Love the show. I was in Atlanta and was in Mike’s workshop on Tuesday with Matthias. Matthias. Woo. I’m building an agent that pulls fabric audit and model APIs to collect grants, direct PowerBI links, semantic model sharing, and app audiences, and then normalizing those records into a

25:38 then normalizing those records into a central store for a workspace free bird’s eye powerbi dashboard. I haven’t integrated team shareoint yet. That’s a future wish list so I can later map which sites contain Azure Active Directory groups or links are. Can you ever see a native tenant level solution for this? What are your thoughts and how would you achieve this? And yeah, this is a great question and I know the question that you had before we just dive in. You notice that in the question they didn’t actually say or

26:10 question they didn’t actually say or Taylor B did not say data governance. However, this is such a big part of data governance. So Mike, I’ll I’ll pause there because I think you have a definition first. yeah, let’s go through those different elements that you wanted to define for our our our episode today. Yeah, I just want to give us some context on what I think data governance is. I think everyone can talk about data governance. people can say a lot of different things. I really want to focus on on the idea of every table, every

26:42 the idea of every table, every table, column,, there’s data governance is making sure that there’s an ownership to what measures and columns exist and how did they get there. I think it’s is what I’m going to be leaning on for my data governance talk, right?, data governance is always an evolving process. There’s always new data being added. You add new programs to your system. So, it’s a little bit of a living organism, I would say, but the system, the system you put in place

27:12 system, the system you put in place around data governance needs to be consistent and reusable. One of the things I’d like to just point out here is there’s a challenge happening right now when you look at your tables as they come into the lakehouse. And I’m going to just focus on fabric that you could do a lot of other patterns here, but I’m going to focus on purely the the governance story around fabric. When you bring in tables to lakehouse, you have to interpret how did those tables get there? What are you doing with the data? Is this data slowly

27:42 doing with the data? Is this data slowly changing dimensions or is this data just the current record or the current value of whatever that table is? So there is there is every step along the way every time you touch the data you’re doing some business logic or rules around that right yeah where’s that where is that information captured who can look it up how do you expose how the dim customers table arrived there what defines an actual customer and we can get into the nuances of all this other like little detaily things of you other like little detaily things of what the customer and do does every

28:14 know what the customer and do does every department define customer the same way or not, I’m not going to be too hard on the actual definition of like what is a customer, but just the fact that the organization needs to talk about it and figure out do you want everyone to have the same definition or do you allow some flexibility in your company to have different definitions of what is a customer right something like that so I don’t want to debate that particularly but at the end of the day someone needs to own it to own it or multiple people need to own it and

28:45 or multiple people need to own it and when I was in engineering we had data governance problems and in engineering there was we were building at the time I was at Johnson Controls we were building batteries for cars and so if you think about a battery that goes into your car there’s lots of different things that go on the battery there’s a label that goes on the front has marketing information on it what does the marketing look like how does the what does the label what size of it what materials it made of there is how heavy what’s the dimensions is it is it up to standards right that’s

29:15 is it up to standards right that’s engineering information that holds to that information, right? They they’re responsible for doing the testing to make sure that that battery is doing what you say it will do. Then there was the finance team. Well, how do we charge for it? What pricing of is it?, and then there’s all these different other teams where,, the same size battery would have different companies buying it, right? We’d have like a an AutoZone or an O’Reillys or an Advanced Auto Parts. Like different companies would buy these batteries from us and label them with their own labels. Great.

29:46 label them with their own labels. Great. Cuz they’re they’re not going to build batteries. They want to buy it from us. So what are all the unique characteristics? The product skew. What do they call it on the shelf? So all of these different teams are talking about the same thing. We’re talk we’re all talking about the battery, right? right? And so each team needed to have some level of responsibility to update their portion of data. And I think of it this way. There’s one row of data representing that particular battery. Mhm. Mhm. In that row of data, there are many

30:17 In that row of data, there are many columns of data. And some of those columns are owned by different teams. I shouldn’t be telling you, Tommy, if I’m in engineering, I shouldn’t be telling you, Tommy, how to produce the final price of the product, right? That’s that’s someone else’s job to figure out margins and things. I’ll tell you what it costs. I’ll roll up the cost of the battery. That may be something I roll up, but you’re in charge of figuring out what the sale price is and how much margin we need to make on it. And so to me, I look at data governance is data governance is this idea of many

30:47 is data governance is this idea of many different teams coming together and agreeing upon products details and things around that may be more of like master data management, data governance that way. that way. That’s one of just many examples around the data engineering, how do you get data in everything else there. So, let me just pause there, Tommy. What’s your definition on data governance? Would you add anything or take away from mine? I I would and I this is what I love about our job too. There’s so many strokes here and I’ll just touch on one thing that you said here is that that

31:17 thing that you said here is that that ownership side. For me, Mike, there’s four pillars and I and I preached this from the rooftops when I when I work with clients around this. There’s four pillars that really make up data governance. So, I’m just going to run through them in no particular order. If there’s one that you want to touch on, let’s do it. Sure. But these four pillars are what makes it up a healthy data governance at an organization. It’s going to be the enablement and empowerment of data at a company. Teams be able to use data. Teams be able to consume data. The promotion of data initiatives at the organization. What is the priority of

31:48 organization. What is the priority of what we need to do? How do we make sure those things are done? Whether it’s integrating systems, whether it’s getting the right dashboards or building up the team. accountability and responsibility through the whole pipeline of data input, through the pipeline of what our definitions are. And finally, it’s cultural and literacy. How do we make sure to your point that we’re all speaking the same language when we’re looking at everything? So again, enablement, empowerment, the promotion of data initiatives, accountability, responsibility, cultural, cultural, and literacy are

32:19 cultural, cultural, and literacy are really the things that make up data governance because it’s not just the data team, the BI team who does data governance. It needs to come from leadership. It needs to come from representatives from each department head that is going through what are the pain points around how they’re consuming and using data. How do we make sure those things get accomplished in the long term? Let’s say you want to actually have AI run through all your PDFs, right? And you want to make sure that is better for data input. Well, you can just say that, but one, how do you

32:50 can just say that, but one, how do you actually ensure that lands as a priority? Is it more important than what we’re doing today? If we already have bad semantic models, who’s in charge of that? So those four pillars for me are what stands up a data governance platform at an organization. So I don’t know if you want to touch on those or if you want me to say them again. again. Nope, those are good. Okay. I I want to I want to double down on those points as well. I also want to point out there is a governance page inside the Microsoft fabric adoption road map. So another another area that’s really good for source of information

33:21 really good for source of information here is there is a Microsoft fabric adoption roadmap section or area called governance and inside there there’s actually a data governance institute. and I like there there’s like two simple definitions in here that I think resonate with what we’re both saying. I gave a very specific example. I probably went too deep too fast on my definition because that’s my experience of like how I interacted with data governance like my mental model of what what it looks like. This one’s much more broad and general in terms but I think the same thing applies here. the idea

33:52 the same thing applies here. the idea is the very simple definition they have is data governance is the exercise of decision-m and authority for data related matters. And then they have a little bit longer of a definition. Data governance is a system of decision rights and accountabilities for informationational related processes. I like that one cuz I think it’s process related for a lot of this. It’s executed according to an agreed upon model to describe what actions and with what information.

34:22 actions and with what information. who can take what actions and with what information and when under what circumstances using what methods. So it’s basically defining how we use data, how we leverage and and work with data. And the one area I’d like to probably most importantly point at here is the accountabilities section. There’s a lot of people, a lot of different people, different teams, different people that all need to understand how and what they own around

34:52 understand how and what they own around the data. And I think as a company, we have to be able to let we have to be able to have a process that allows different teams to own different portions of data governance. who’s responsible when there’s a problem, what is our process we follow to ask questions about that data, right? right? That’s that’s key here, I think. No, I a thousand% because I think accountability goes a few ways, right? Because I think most organizations, at least the ones I talked to when I I

35:23 least the ones I talked to when I I start working, they have a misconception

35:24 start working, they have a misconception of data governance where what actually let me ask you this. Let’s see if we can do a we can think alike again. What is the common word when people hear data governance they associate? If you did word association, there’s a single word people always think data governance is painful. painful. I was going to say another feature to one. Yes. Yes. Or but no. Yeah. no. Yeah. What do they think of data governance as? What’s another associated word? Oh, I I’ll give you what letter starts with. Starts with S.

35:55 Starts with S. Security. Security. Yeah. Yeah. All the time. Maybe what they’re thinking about. Yeah. That’s that right like it’s like who can see what right and that’s they think data governance is but that’s a small that’s a small view of what data governance is that’s what I’m saying it’s a misconception and again we you and I have talked about just in this beginning of the topic Mike we’re talking about how do we make sure people are doing what they need to do how do we make sure if how the data flows that we can point it back to us to someone and I don’t want to just say that as the negative like if it’s wrong we’re going

36:26 negative like if it’s wrong we’re going to chastise someone but when the projects are done right and we know the data is flowing correctly that people can trust it. We know where those different pipelines are. If people are inputting information wrong, there needs to be accountability and this goes all into those different aspects here. So let’s take that now with the mailback question here. Yeah, about how can I agents actually help with this, right? Because a lot of data governance is not on a computer to me. It is the committee and

36:56 computer to me. It is the committee and it’s the people, right? I always when I start data governance projects and I think similar to you there’s a we I always tend to start with workshops and assigning the committee where are the different roles that people are going to have do we have a stakeholder do we have the committee the leadership committee do we have the people are going to then execute on what the initiatives are put out and none of that has to be done on a computer that’s not technology the internet again that’s really the people however it gets to the point though where we need to define these things we

37:27 where we need to define these things we need to put them into perspective right you have to prioritize what you’re going to do and that’s the project phase of this so for me I think about this part and I think about aligning this with the knowledge center or the center of excellence Mike and I have an association here agents are a great assistant as a secretary as a transcript side of this that’s one aspect there but I’m going to let me hit the ball over to you here when you think

37:57 the ball over to you here when you think about data governance, when you think about the people side of this, is that how you see agents being the most helpful or are you more looking at this from a technological point of view, the backend point of view? I think there’s like two major pieces here that I think we either don’t have in fabric or it’s not quite there yet. Right. So, let’s let’s just talk about Yeah., let if I had to ask questions about the data,

38:28 about the data, right? Why did this number come out to this number? How did this table get here? What’s the lineage of this table? I’m not sure if we have a lot of good resources where the code that we actually have produced inside the semantic model like just think about it the m a lot of a lot of what we do Tommy is we conceptually think about our data what does it need to do how do we need to shape it how does it fit together and then what we do is we take that

38:58 then what we do is we take that knowledge and we hardcode that into actual code we make real pipelines, real code and that is a representation of the decided agreed upon business logic around how that table gets there, right? What are we filtering out? How are we loading the data? That’s just knowledge, but it’s all being baked into a process. And so I want to really lean on particularly what this data

39:28 particularly what this data governance institute is talking about is the whole goal of data governance is to make better decisions right right it’s to remove friction getting things done it protects your needs of the stakeholders stakeholders say here’s our main objective our our key leaders are saying we need to understand our revenue we need to understand our expenses we need to understand our shipping whatever the thing is there’s a there’s a broader task ask at hand and all this stuff is supposed to help guide the data into

39:58 supposed to help guide the data into those decisions. And so one of the the last piece of this I think they talked about was ensuring transparency of your process, right? Do we does data governance actually allow people to see and trust and understand this is how the data got there? Because a lot of times I think Tommy, we want the data to say what we want it to say. [snorts] [snorts] I want the data to be like, “Oh man, if I just do X, I get more sales, right? right? I want that to be true, but the data may

40:28 I want that to be true, but the data may not actually show it.” And so, I’m going to interrogate the data. I’m going to massage the data. I’m going to go look at I’m going to try to like really figure out what’s going in the data so I could hopefully understand some correlation. If I put in this level of effort, if we increase our sales team, if we spend more money on ads, if we do these other things, does the output come the same way? Right? So I think governance is really about that transparency that trusting being able to have that now okay let’s come back to what you were said about agents where does agents fit in this right there’s a lot of people and process involved here

40:59 and process involved here I think the agents are really good at listening to the the conversations between real people talking about data governance Tommy you and I get on a call we’re asking about well I don’t trust this table I don’t know what’s going on here I’m having difficulty with this can you help me understand that this number doesn’t look right. Can you help me figure out what’s going on here? Right? We can have that conversation, but the AI can summarize that down and produce, okay, well, we we, what did this conversation produce for us around

41:29 this conversation produce for us around data governance? Is there something we need to improve in our process? Is there something that we missed? Is there something that we we didn’t connect on for this particular user? So, I think AI could be used there in one in one fashion. I think the data agent is supposed to be somewhat supportive around helping users ask questions about how did this data get here? What does this table do? How do these measures work? That’s another part of transparency. I’m

42:00 That’s another part of transparency. I’m building a bunch of workloads that are doing very similar things like I want to data governance have data governance around a semantic model, right? Who can see the tables? Who can see the relationships? who’s allowed to see the measures and how they’re written. All that information should be captured somewhere. It should be easily a visible to the end user. And if I give an end user a thin report or a PowerBI pageionate report, there’s zero visibility Tommy to what’s in the model. What does this measure do? How does this measure rely

42:30 measure do? How does this measure rely upon other measures or have visibility to tables? I think that’s important to people. I think they care about that stuff. I it’s funny because I I want to touch on the transparency side here because I want to say something and you may not you may or may not agree with this but I think agents have a very small role but an incredibly essential one and let’s not get let’s not get misunderstood. Agents are not going to lead your data governance program and I think when people hear this they’re think like oh well we can just

43:00 think like oh well we can just build a bunch of agents to basically build it out. I think they’re going to apply roles that I think agents are already very good at, but that transparency side is probably the hardest thing about data governance in itself because it’s the thing you have to keep up on. For example, you have a let’s say you’re beginning to build your data governance program. Well, we want to know definitions. Okay. Well, usually that means people have to always keep track of what’s in the semantic model from all these different teams. It’s almost an impossible task. However, even

43:32 almost an impossible task. However, even though it’s mundane, right? It’s not necessarily you’re building up the architecture and you’re building out everything. That’s part of the trust that you talked about, but that’s not easily achievable. Well, that’s a great opportunity for what’s already out there, right? An MCP to connect to that an agent can connect to to always read what the definitions are and then to push that back to a context harness, right? whether it’s SharePoint or Confluent or notion wherever the essential source is this

44:02 essential source is this relationship between the MCP or the fabric an agent’s ability to look at fabric and then wherever we’re storing our context I think is a huge win here that again almost has not been achievable and I know you and I don’t remember his name but you and someone else tried to build out a way to do definitions and document control. I tried to build out something in Power Apps, but it’s nearly impossible to keep track on. So the transparency side is huge there where let’s utilize what’s

44:34 huge there where let’s utilize what’s already out there and rather than recreating the wheel initiatives are going to be by the people but the mundane task being just to do the transparency to say what is my data catalog what are the semantic models out there how can I make this human readable so I know if I’m looking for sales what are all those reports out there well that’s a great opportunity with an agent with an MCP to do So that’s really where I think the transparency side which again it’s one of those it’s not

45:04 again it’s one of those it’s not important until it’s the most important thing like to always plug things in like that where transparency can be such a big help from agent. I’m not saying that’s the only use case. However, if you were just starting off my argument would be that’s the best use case. Yeah. And I think you’re telling me a lot of this is interpretation level for me. Right. I want the agent to interpret what we’re saying, right? to your point, there’s key objectives, there’s key governance things that you’re trying to communicate. Yes, we should definitely

45:34 communicate. Yes, we should definitely do that. Let’s communicate those key governance, those key areas of of knowledge there, right? But how do we give that to our users? What’s the best way for them to understand the policy? can we give out the policy to an agent or document the policy so that the agent can read and understand what that policy looks like? There’s a level of tier one support here that I think we could be pretty well served by these agents just gaining that initial level

46:04 agents just gaining that initial level of trust of trust right right what for example let me just give another proposal here let’s imagine you have a semantic model inside that semantic model we have a number of relationships between tables and we have measures measures today Tommy what do we have today that will allow you to ask a question of how does this measure interact with the tables? What dimension columns can be used with this measure? We have nothing. We have we have nothing that really

46:34 We have we have nothing that really helps us understand the relationship between the columns and all the relationships between tables for that particular measure. This is an area that’s the governance how the data is related, right? Ease of operation with data that exists inside the semantic model. it’s there. Extracting that out without spending tons of time researching on top of the model, that’s tricky. And you’re just talking about the semantic model here. So, as we get near

47:04 semantic model here. So, as we get near the end, we have more time, but just one thing that you brought up as well was there’s a whole other layer here that gets everything much more complicated

47:13 gets everything much more complicated unfortunately because you’ve mentioned the semantic model. However, let’s talk about what does fabric how does fabric play in a data governance story? One, I I think you wrote this question, but one does fabric even have a data governance story and I think this is a huge part because everything it’s hard enough our jobs have been hard enough to try to do data governance with just semantic models man like in terms of trying to keep on track of the reports that are out there. Who is it for? What purpose is it served? And then

47:44 for? What purpose is it served? And then the layer of the definitions and the columns and what’s important there, right? That’s a job in itself that I don’t think to this day I in a company have a really great handle on. Okay. Now add fabric to this, right? Add little lakeous, add,, the the different transformations that are being done. Where are they being done? Because it’s not just power query now. And in my head there becomes this entanglement because you’re not just talking now about all these different artifacts but

48:15 about all these different artifacts but then it goes back to well who’s data governance for right and there is data governance for the architectural team the team building but there’s also data governance for the consumer. So this becomes a very tangled web in my head but let’s let’s back it up then. How does you actually have fabric at a data governor story here? And I want to bring up a few points here where this has never been more important to me on how you actually build up your committee

48:46 how you actually build up your committee or the the people who are involved in in the data governance story because you cannot just have people with some limited knowledge of technology or not a leadership role at the organization. If you do not have people who have the ability to speak business and the technology, I find it very difficult in the age of fabric for you to ever have a good handle on this. And I go back to the people side of this, Mike, because I need to have someone from sales, you need to have someone from sales,, a vice president or a, you

49:18 know, a vice president or a, you know, a vice president or a,, someone who understands the know, someone who understands the business, but I need to have someone who is going to be able to speak that, but understands fabric’s the,, architecture. To me, I find it very difficult here because I can’t get to agents yet until we get to where does fabrics how does fabric complicate things or would you even agree with that? Would you even say fabric complicates it? Maybe a better way to ask the question is how does adding fabric the layer of fabric on top of

49:48 fabric the layer of fabric on top of data governance change how we think about data governance? I want to go the other way around. I think I think data governance is the consistent thing that we’ve been doing all along. I would say,, I think you were just saying what I heard you saying was what happens when we add data governance to fabric. I think it’s the other way around. around. Whether you have it or not, you already have some form of data governance. It’s already there hopefully hopefully whether whether you’re using fabric or not,

50:18 not, right? It’s it’s in the original systems that you build, right? It’s in your business objects that you’ve already created. It’s in your how do you give access to data to different individuals across the company who has access right? So that data governance I believe already exists. I think it’s more about how do you add fabric into your data governance. I think that’s the way I would I’d rather take it whether you like it or not. You just may not have any data governance. It may be very loosey goosey and not very well defined.

50:49 loosey goosey and not very well defined. Well, okay. So what does that mean for your fabric usage? How does fabric push into that new governance pattern? And so so that’s maybe where I would lean on is looking at that angle of things and saying, okay, if we step back and say, okay, some of the data governance pieces where do we fit them in? Right? This is where I’ I’d like to look at like the one lake catalog.

51:21 one lake catalog. One lake catalog is interesting to me because it does some of the initial phases of what data governance would look like. Meaning here’s all the lakeouses you have access to. Here’s all the reports you have access to. Here’s all the semantic models you have access to. And inside the one lake catalog, we start having a story around giving access and and governing the data into individual users. Now, I I fully feel like the data catalog isn’t robust

51:51 like the data catalog isn’t robust enough to to [snorts] do really what we want ultimately for true data governance and and what we really want here, but it’s a step in the right direction. It’s going to give you some assistance around understanding what’s in a particular semantic model, seeing some of the data., one of the things I really like recently was this ability, the one lake catalog now has an exploration in it. You go to the one lake catalog, you click on a table. Hey, I want to explore this data. An exploration shows up and you can actually see and run queries against that data and see in real time

52:22 against that data and see in real time like what’s coming into that data. You can pick different columns, see what’s there, what’s there. I think that’s very relevant, right? We could even go into like really nuanced areas of when you load in data, does this column only have an integer in it? Did someone not accidentally shove in a string to it? Right? What is that rule? Where does that rule go? what notebook or process do you have in place to test that rule? So there’s like really nuanced areas here that could be data governance, but fabric one leg doesn’t really support

52:53 fabric one leg doesn’t really support any of that. And so back to the user’s question here is if we if we look in again on what the user was questioning here was, hey, I’m trying to get an agent that helps me pull fabric data and model APIs. I’m collecting direct PowerBI links. I’m correct. I’m collecting semantic model sharing app audiences and normalizing them into a central place. That’s data governance. You’re building a process around who has access to what items.

53:23 around who has access to what items. This is a very small part of the bigger data governance story. And I would also argue too, Tommy, there’s not a lot of good documentation around this. This is FUAM. This is why we build a solution. What did you say? FUAM. FUAM. Fabric unified admin monitoring. Okay. Gotcha. It’s it’s a it’s the FUAM system is Microsoft’s version of how do you load a bunch of pipelines, hit a bunch of API calls, and get a common structure of data down so you can use it and go look at it., this is that’s that’s more what we’re talking

53:54 that’s that’s more what we’re talking about. Who’s exporting data? Who’s not? And are there sh are there security groups being used? Do those same security groups have access to SharePoint site? There’s this whole there’s a huge picture of your organization. Who has access to what data? one I think small portion of what data governance is. Yeah. I I almost got offended because I was like well fam to you too, man. Like but now I but now I shut your fam. You Yeah, you shut your fam. Shut a good I think that there’s there’s two elements you’re talking about here. I

54:24 elements you’re talking about here. I like that. I love that you bring up the one lake catalog because I think the people side of this in fabric have the ability more than ever before to focus on two things. You focus on the right data. We can be very targeted correct on yes what we want to build in the pipeline for hey how does quota come in with Salesforce and our C and our other systems well let’s tie that together that’s a huge important task for us let’s make sure that we have that defined so we can see from source to report how we get the right data and initiatives it’s never

54:56 right data and initiatives it’s never been easier to your point well we talk to this application all the time how important is that initiative we can do that in fabric the Asian side you bring up something interesting with one lake catalog catalog quality, right? Not just defining items, but with lake houses, right? Well, data quality can be something you can build or have an agent assist you on, right? Let’s go a perfect example. I have seven sales regions and I need to make sure there’s always only seven sales sales only seven. There’s no undefined or

55:27 only seven. There’s no undefined or unknown or someone doesn’t change some of the data, right? the relationship with the data committee says what is the most important data quality that we need to make sure make sure product names and sales regions okay let’s now feed that to the agent every day it runs that lakehouse those tables to do what are the distinct values what are those sales so that’s something we can monitor if that ever goes query yeah you like this right yeah I like that I like this one so let’s go down this route a little bit more like where does agents fit here or how do

55:57 where does agents fit here or how do they fit well in this right I think this is a is a This is a great point to fit into. Right. Right. And this is also let’s go back to that one lake architecture diagram. Right. There is a one link architecture diagram here that’s actionbased eventing. Right. So this is the real- time intelligence. This is real time information. Right? As you run your data do we have the ability to as the data loads detect what new data shows up and in that new data run a test on it. And if something doesn’t meet the test

56:27 if something doesn’t meet the test notify someone. And again, this is going back to our data governance, which is okay, I’m bringing in net new data. Something in that net new data is wrong, right? right? Can we flag it, quarantine it, do we delete it? That’s that’s the that is the data governance we’re talking about. And then also, if you’re going to allow it to go through, or you’re going to quarantine it, it, who do we notify? Like, this all should be processed. We should set this up. Who do we notify to say, “Tommy,, you’re

56:58 do we notify to say, “Tommy,, you’re in charge of the sales data. Someone on your team entered in this information. It’s not right. It doesn’t link up correctly. Here’s a handful of records that we found that don’t match our agreed upon data quality. Go fix it.” And then that should be something we can give a task back. We agree. Like, so this is a contract, right, Tommy? We agreed that we’re only going to have seven regions and there was going to be every piece of data is going to have all that’s zero of AI that’s zero agent that’s zero actually of real data

57:28 that’s zero actually of real data from the committee the initiatives. Yeah. Yeah. Yes. That’s the initiative. That’s the agreement between like we’re writing this down. We’re we’re we’re having a data contract between you the data provider and me watching that process or helping to produce that data back to people because that way better data in means better insights and data out as well. So, I love that. And this is where the agents, I think, make sense to monitor, look at, listen to those data qualities. When there’s an anomaly showing up, send a message. Yeah. Say, “Hey, by Mike’s the one who put in a new region,

57:59 Mike’s the one who put in a new region, talk to Mike thing or send Mike an email because I think that again the by itself it’s not going to happen without the initiatives. To your point, I like this a lot. And I I I think the data quality, the data definitions, like well, who’s in charge of approving definitions, right? And I think both those elements are such an easy win. So let let’s go to this, Mike, because I man there’s so much more here that we could talk about. What do you think is possible in 30 days? And I’m going to I

58:29 possible in 30 days? And I’m going to I I’ll start here, but if you were to say what’s an MVP, a minimal viable product to get data governance with agents running in your company. again, we can make some assumptions here about the size of the company, but let’s just start here. So, I I’ll begin and I’ll think here. Yeah, let me This is a a good question. Let me just see if I can pull I’m going to pull some notes here on some stuff, but go ahead, Tommy. I want you to start. Okay, so I always start if I’m starting

58:59 Okay, so I always start if I’m starting data governance from the ground up. Again, I’m going to start with a

59:01 Again, I’m going to start with a workshop. I’m going to start with that committee making sure we have those people. Let’s assume we already have the people. Let’s understand biggest pain points, the biggest long-term initiatives that we’re trying to do. I’m going to assume data quality and definitions are a big part of that. And I would set up let’s let’s understand the lay of the land. How many problems do we actually have? Sales is complaining a lot. Okay. Well, where are those problems coming from? I would want to be able to stand up the following in 30 days. The first thing would be our

59:31 30 days. The first thing would be our data quality standards. Do we actually have any? And based on the amount of erroneous or bad data coming in, is there accountability? I would want to stand up my accountability first off understanding all the data that we’re flowing into fabric. Can we point it back both from the peer person doing the transformations and the person inputting the data on where that’s coming from? The second part that I think is an easy win in that 30 days is the data trust and definitions. Let’s

60:01 the data trust and definitions. Let’s look at all the measures and all the reports and the audits that we have. Well, we have seven reports going to sales and the views are sporadic. Is that the way we want to go? Where is sales actually going to find that information? And that data quality and the data trust or the data literacy is something that I can set up with agents both saying my usage and both saying ambiguous definitions. And so that’s what I would look at 30 days if I had to say outside of what the long-term goals are just to get a lay of

60:32 long-term goals are just to get a lay of the land to say where are we really starting from is what I would want to do in those 30 days. Yeah, I think I’m going to agree a lot with where you’re at on this. I think there’s definitely some discoverability. I’m I’m also going to maybe assume here most companies have some level of data governance already. I would argue yes but I I just how mature you’re in that governance level that would probably vary widely between organizations right and even inside a

61:03 organizations right and even inside a single organization certain teams are going to have much more capability around data governed governed data than other parts so if I’m talking about 30 days if if you are hearing the buzz term first I think I would align on in that early part of the 30 days really getting to leadership and saying do we really agree upon what data governance means to our team because at the end of the day data governance is a expense to the business I’ve drawn this

61:35 expense to the business I’ve drawn this chart many times for for firms and consulting things I think of data governance being like an asmtoical line right it so if you want 100% clean governed data it’s going to be really expensive your your your costs will get really expensive to have 100% clean auditable data. One of my examples I use here is you look at banks. Banks can’t lose your money. They have to have very clear auditable processes of when you get a transfer of

62:05 processes of when you get a transfer of a dollar into your bank account, they have to make sure they never lose it. There can’t be any data quality issues on their side. So there’s backups of backups and checks on upon checks and making sure everything works as expected because soon as someone starts losing money, people get sued, right? So you money, people get sued, right? So the the cost to and this is why know the the cost to and this is why banks and systems that the banks run are really old and very expensive because it’s very expensive to test new things, things, right? So I I think in general banks are

62:37 right? So I I think in general banks are a good example of like you have to have really good data quality in their space. Now we’re not all running banks. I understand this. but the the price of that data governance goes up the higher you want to get to that data quality. So I think going back to my comment talk to leadership ask them like what do we think data governance is? Are we having any major issues with data governance today currently? What are our users of our data struggling with? Hey, I really don’t understand these models Tommy’s building. We have so many different reports coming down from Excel

63:07 different reports coming down from Excel about sales and sales numbers. I don’t know which number to trust. Right? There’s probably issues in your process today. And so you find those issues and you start saying, “Okay, is this a data governance problem? Is this something we need to start working on together?” And I need leadership to buy into. we’re willing to spend money, time, effort, people to work on this. If we don’t have that number one, just don’t worry about it because

63:37 just don’t worry about it because let me Yeah. Right., if we we’ve already we’ve already failed before we started. Let me virtually high-five you here real quick, Mike, because because I think it’s a tellt tell. We’re talking about agents and governance, but I would say about if I were to guesstimate 70% of what we just said as our initiatives were people, the people skills. Mike, if you and I didn’t have people skills, you cannot get to that point of getting agents to run. And

64:07 that point of getting agents to run. And I think that was such a important thing of what you mentioned here. You talked about discoverability. You talked about really interviewing and investigating where we’re at, which requires no agents at all to me. And I think that’s such an important part. So Mike, I really enjoy this conversation. Do you have anything else that you wanted to touch on before we conclude? If you read the article from Microsoft around data governance specifically, there’s different halfway down the article, actually it’s a really long article, believe it or not, but there’s there’s a really good approaches.

64:37 there’s a really good approaches. There’s three different approaches that Microsoft recommends. You can fully roll out fabric and then come back and then work on governance planning. You can governance plan everything up front and then roll out fabric. The option I like the best is a little bit of like give and take. and take. Build a lightweight version of governance. Roll out some of fabric. Come back review. Do you need to add anything for governance? Roll out some more fabric. Look at your governance. So do some fabric. Do some governance. Do some fabric. Do some governance. I think this is a voling experience where you’re

65:09 this is a voling experience where you’re going to figure out how to shape these things together. you need to have leadership buyin. I think that’s so important and so key., and then you important and so key., and then to answer the question directly know to answer the question directly from our user who submitted, Taylor who submitted a great question here. Wonderful. Thank you for submitting. Do we ever see tenantwide governance on who has access to what items? I don’t know if we’re ever going to get that. I think fabric’s really big. I think Microsoft would agree that this is going to be useful to people, but I think at the end of the day, Microsoft’s not

65:41 the end of the day, Microsoft’s not really motivated. Like if you talk about what Microsoft’s goals are, they’re designed around giving more monthly active users. And for Microsoft to spend a ton of time helping you figure out who has access to what item, that’s not really going to drive more monthly active users. I think at this point they’re going to try to do more real time analytics and RTI and, you real time analytics and RTI and,, let’s throw an ontology at you, know, let’s throw an ontology at you, whatever that means. Like there’s always going to be other things. There’s these bells and whistles they’re going to continue to throw at you with things., so I don’t think Microsoft’s

66:13 things., so I don’t think Microsoft’s going to give us exactly what we want and I think businesses are so unique in how they think about governance and where they are on governance maturity. We need to think of better ways of doing this., I do feel like you want to partner with people who have done some of this. Tommy, you’ve done a lot of data governance. I have rolled out the Yeah. Yeah. Yep. I have done I’ve rolled out the FUAM the fabric unified admin monitoring solution which I think does a lot of these things already helps you pull that data together and at least pull it into

66:43 data together and at least pull it into a single pile. Carlos Solutions, my company, has also built a custom monitoring solution that’s lightweight, just grabs a handful of APIs and grabs all the operations and logs those every day, right?, one data governance thing I think would be really relevant for companies is every day you should be going to grab all of your tenant settings and storing them in a one lake because that way you can see from an auditability standpoint did any of my settings change modify modify who from one day to the next what

67:13 who from one day to the next what changed I think that’s something that you need to have audit about Microsoft doesn’t give this to us they give us an API they let us run pipelines so so there really does need to be like a data governance workspace of how we start implementing this data governance. Now the other interesting thing here too is where does agents where do agents fit in this and I really do think that we Carlos solutions and Tommy I know you as well we’re building software and solutions that is incorporating AI into

67:45 solutions that is incorporating AI into our process. Tommy, your example today, the task flow assistant portal, great example, right? We’re talking to an agent, we’re helping it think through what it’s doing, and then we’re asking the agent to build out stuff or samples or architectures for us as well. as well. So, I think I think what we’re going to find is we’re going to continue to find better ways of incorporating agents into our existing process, into our existing governance. we’re getting better at this. It’s it’s getting more useful to

68:15 this. It’s it’s getting more useful to us. but right now, fabric doesn’t really have the single silver bullet to get all this resolved at once. You’ve got to be creative here. And to your point, I don’t think they’re ever going to create the single solution that your company exactly will need. They’re going to give you all of the wrenches and sockets and tools that you need to produce it, but it’s going to be up to you to figure out what’s the off-the-shelf solution and either finding a partner or someone who’s already built one of these that can help you get this going quickly. So, anyways, that’s my

68:45 quickly. So, anyways, that’s my thought on data governance. You want to wrap there, Tommy? Does that sound good? I think No, I think that’s good., I I do love your attempt to try to remember the word studio. By the way, I I really tried hard. It’s studio. studio. You’re throwing some darts. No. Yeah, but it dude I love this conversation today. So, no, let’s wrap it up. it up. All right, with that being said, thank you very much for listening around our conversation around data governance and how that works with potentially AI agents., stay tuned. I guess we’re going to have to keep watching Fabric and figuring out where else we can put

69:16 and figuring out where else we can put and pull in agents., if you do want to see some different agent experiences, go check out Power Designer. So, Power Designer is a workload that you can go use today inside Fabric., it’s free. You can go use it for free. if you go pay for the the tips plus membership, you can even add your own agent. You can go add agents from Foundry directly into Fabric and use them with these workloads, which I think is really interesting. So, if you want to see other experiences not co-pilot related, you’re more than welcome to go check out our workload. And we’re trying to think of a new way of adding agents in

69:46 think of a new way of adding agents in here. And if Tommy and I come up with some really cool data governance stuff, maybe we build another workload item that incorporates an agent around governance things. Who knows? Come up with an idea. Let us know in the comments down below. What would you want to see? What stuff would you like us to build? That could be we’ll we’ll we’ll live stream a build of something like this. Would be cool. Anyways, with that being said, Tommy, where else can you find the podcast? You can find us on Apple, Spotify, wherever. 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

70:17 idea, or topic that you want us to talk about on a future episode? 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 on all of PowerBI tips social media channels. Thank you all so much and we’ll see you next time. Explicit measures pump it up high and lighting up the sky. Dance to the laughs in the mix and I get your fix. fix. Explicit measures drop the beat now

70:49 Explicit measures drop the beat now the crowd. Explicit 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.